PCB DEFECT DETECTION USING YOLOv5

Submitted in partial fulfilment of the requirements for the

degree of

Bachelor of Technology
in
Electronics and Communication Engineering
by

Priyanshu Shahi (2100970310122)

Priyanshu Yadav (2100970310123)

Under the Guidance of

Dr. Shahid Eqbal

Assiociate Professor

Department of Electronics and Communication Engineering

(B. Tech ECE – Accredited by NBA),
Galgotias College of Engineering and Technology, Greater
Noida
(Affiliated to DR.A.P.J Abdul Kalam Technical University,
Lucknow)

January 2024
 DECLARATION

We hereby declare that the report entitled “BER analysis of K-PSK with OFDM RO-
FSO system over double generalized fading distribution” submitted by us, for the
award of the degree of Bachelor of Technology in Electronics and Communication
Engineering to Galgotias College of Engineering and Technology, Greater Noida
affiliated to Dr. A.P.J. Abdul Kalam Technical University, Lucknow is a record of
bonafide work carried out by us under the supervision of Mr. Aksay Kumar.

We further declare that the work reported in this report has not been submitted and will
not be submitted, either in part or in full, for the award of any other degree or diploma
in this institute or any other institute or university.

Place : Greater Noida

Date : 12-01-2024 Signature of the Candidates

1
2

1
3 CERTIFICATE

This is to certify that the report entitled “BER analysis of K-PSK with OFDM RO-
FSO system over double generalized fading distribution” submitted by Suyash
Trivedi and Tanushree Singh from the Department of Electronics and Communication
Engineering, Galgotias College of Engineering and Technology, Greater Noida
affiliated to DR. A.P.J. Abdul Kalam Technical University (AKTU), Lucknow, for the
award of the degree of Bachelor of Technology in Electronics and Communication
Engineering, is a record of bona fide work carried out by him under my supervision, as
per AKTU code of academic and research ethics.

The contents of this report have not been submitted and will not be submitted, either in
part or in full, for the award of any other degree or diploma in this institute or any other
institute or university. The thesis fulfils the requirements and regulations of the
University and in my opinion meets the necessary standards for submission.

Place: Greater Noida

Date: Signature of the Guide

The report is satisfactory/unsatisfactory

Approved by

Head of the Department

2
 CHAPTER 1
Introduction

With more than 100 years of history, printed circuit boards (PCBs) are one of the key
concepts in electronics, as they form the substrate for the electrical connections
between the different electronic components attached. The main advantage of circuit
boards is that they reduce the probability of wiring and assembly errors and their
manufacturing can be highly automated. With the advancing development of
semiconductor design and manufacturing technology, PCBs are becoming ultrathin,
high-density, and multilayer to accommodate high-performance components. The
quality of a PCB depends on the quality of each line and hole, and problems in any
one of the thousands of lines and holes on a board can affect the quality of the final
product.

In the early days, manual inspection of PCB defects was performed. The operator
used a magnifying glass or a calibrated microscope to determine whether the PCB
was qualified. The disadvantages of this method are high cost, discontinuous defect
discovery, and difficulty in data collection. With the increasing production of PCB
boards, this method is no longer economical or practical.

In recent years, much research has been conducted on detecting PCB defects. With
the development of machine vision technology, automated visual PCB defect
detection has been gradually accepted by the industry and is being widely used.
Different vision-based defect detection methods have been studied and introduced in
the literature for automatic PCB inspection [1–3]. In recent years, an automated
optical inspection (AOI) technique has been developed to test and check PCBs for
potential defects [4, 5], such as surface, dimension defects, or component placement
defects. As PCBs are becoming more and more complicated, manual checks for
defects are becoming difficult and time-consuming, so AOI plays an important role
in verifying the quality of the printed circuit board after assembly and significantly
affects the line’s production quality and efficiency.

On the other hand, with the development of deep learning, many researchers have
proved that deep learning technology can be used to detect PCB defects with higher
accuracy, which results in higher productivity [6, 7].

In this paper, an improved YOLOv5 network is proposed, which ensures that the
3
model size can meet the requirements of embedded deployment with an improved
ability to detect defects in real time.

1.1. PRINTED CIRCUIT BOARD (PCB)

A printed circuit board (PCB) is an electronic assembly that uses copper conductors
to create electrical connections between components. PCBs also provide mechanical
support for electronic components so that a device can be mounted in an enclosure.

All PCBs are built from alternating layers of conductive copper with layers of
electrically insulating material. Conductive features on printed circuit boards include
copper traces, pads, and conductive planes. The mechanical structure is made up of
the insulating material laminated between the layers of conductors. The overall
structure is plated and covered with a non-conductive solder mask, and silk screen is
printed on top of the solder mask to provide a legend for electronic components. After
these fabrication steps are completed, the bare board is sent into printed circuit board
assembly, where components are soldered to the board and the PCBA can be tested.

During manufacturing, the inner copper layers are etched, leaving the intended traces
of copper for connecting components in the circuit board. Multiple etched layers are
laminated in succession until the printed circuit board stack-up is complete. This is
the overall process used in PCB design fabrication, where the bare board is formed
before passing through a printed circuit board's assembly process.

1.1.1 HISTORY OF PCB

In the past, electronics were designed and assembled from small integrated circuits
(ICs) and discrete components, which were connected together using wires, and the
components were mounted to a rigid substrate. This original substrate was initially a
material called bakelite, which was used to replace the top ply on a sheet of
plywood. The number of wires was so great that they could get tangled or inhabit a
large space within a design. Debugging was difficult and reliability suffered.
Manufacturing was also slow, where multiple components and their wired connections
were manually soldered.

Old PCBs often looked like the system shown below, where copper traces are exposed

4
and many through-holes are used to hold large, bulky components.

Today, standard designs can have many small components, such as tiny ICs, very small
passive components, and advanced chips with very high pin counts. It's impossible to
manually connect all of these components together with soldered wires, so the copper
connections are deposited directly on insulating substrates as described in the above
manufacturing process. Many of today's devices are advanced high density interconnect
(HDI) designs with thousands of connections and multiple electrical interfaces,
powering everything from smartphones to heart rate monitors to rockets.

1.1.2 KEY FUNCTIONS OF PCB

• Mechanical Support: Provides a sturdy platform to mount and support

electronic components.

• Electrical Connectivity: Copper traces connect different components, enabling

electrical signals to flow correctly.

• Signal Integrity: Ensures minimal interference or signal loss during

transmission.

• Heat Dissipation: Some PCBs are designed to dissipate heat generated by

high-power components.

• Compact Design: Allows complex circuits to be implemented in smaller

spaces.

1.2 YOLOv5
YOLOv5 is a model in the You Only Look Once (YOLO) family of computer vision
models. YOLOv5 is commonly used for detecting objects. YOLOv5 comes in four
main versions: small (s), medium (m), large (l), and extra large (x), each offering
progressively higher accuracy rates. Each variant also takes a different amount of time
to train.

The YOLOv5 repository is a natural extension of the YOLOv3 PyTorch

repository by Glenn Jocher. The YOLOv3 PyTorch repository was a popular
destination for developers to port YOLOv3 Darknet weights to PyTorch and then move
forward to production. Many (including our vision team at Roboflow) liked the ease of

5
use the PyTorch branch and would use this outlet for deployment.

After fully replicating the model architecture and training procedure of YOLOv3,
Ultralytics began to make research improvements alongside repository design changes
with the goal of empowering thousands of developers to train and deploy their own
custom object detectors to detect any object in the world. This is a goal we share here
at Roboflow.

The following points shows some of the updates that were made by Ultralytics:

• June 22, 2020: PANet updates: new heads, reduced parameters, faster inference
and improved mAP 364fcfd.

• June 19, 2020: FP16 as new default for smaller checkpoints and faster inference
d4c6674.

• June 9, 2020: CSP updates: improved speed, size, and accuracy (credit to
@WongKinYiu for CSP).

• May 27, 2020: Public release of repo. YOLOv5 models are SOTA among all
known YOLO implementations.

• April 1, 2020: Start development of future YOLOV3/YOLOV4-based PyTorch

models in a range of compound- scaled sizes.

These advancements were originally termed YOLOv4 due to the recent release of
YOLOv4 in the Darknet framework, the model was renamed to YOLOv5 to avoid
version collisions. There was quite a bit of debate around the YOLOv5 naming in the
beginning and we published an article comparing YOLOv4 and YOLOv5, where you
can run both models side by side on your own data.

We abstain from custom dataset comparisons in this article and just discuss the new
technologies and metrics that the YOLO researchers are publishing on GitHub
discussions.

It is worth noting, since the repository was published, significant research progress has
occurred in YOLOv5, which we expect to continue, and might give some justification
to the YOLO-"moniker". Thus, YOLOv5 is by no means a finished model: it will evolve
over time.

6
1.2.1 YOLOv5 ARCHITECTURE

Object detection, a use case for which YOLOv5 is designed, involves creating features
from input images. These features are then fed through a prediction system to draw
boxes around objects and predict their classes.

Fig 1.1 The anatomy of an object detector (citation)

The YOLO model was the first object detector to connect the procedure of predicting
bounding boxes with class labels in an end to end differentiable network.

The YOLO network consists of three main pieces.

• Backbone: A convolutional neural network that aggregates and forms image

features at different granularities.

• Neck: A series of layers to mix and combine image features to pass them
forward to prediction.

• Head: Consumes features from the neck and takes box and class prediction
steps.

Fig 1.1 Another picture of the object detection process (citation from YOLOv4)

7
With that said, there are many approaches one can take to combining different
architectures at each major component. The contributions of YOLOv4 and YOLOv5
are foremost to integrate breakthroughs in other areas of computer vision and prove that
as a collection, they improve YOLO object detection.

1.2.2 WORKING OF YOLOv5

• Input: The model takes an input image (e.g., 640x640 pixels).

• Feature Extraction: The backbone extracts features from the image using
convolutional layers.

• Feature Fusion: The neck combines multi-scale features to detect objects of

various sizes.

• Prediction: The head outputs three types of predictions:

o Bounding Boxes (x, y, width, height)

o Objectness Score (probability of object presence)

o Class Probabilities (e.g., cat, dog, PCB defect, etc.)

• Non-Maximum Suppression (NMS): Removes duplicate bounding boxes and

keeps the best prediction.

• One Pass, Many Predictions: YOLOv5 processes the entire image in one
forward pass, making it real-time.

1.2.3 YOLOv5 TRAINING WORKFLOW

Step 1: Dataset Preparation

• Organize data in YOLO format: Images and corresponding annotation files

(.txt).

• Use tools like LabelImg or Roboflow for annotation.

Step 2: Configuration File

• Define dataset.yaml (path to data, classes, etc.).

• Adjust hyperparameters (e.g., epochs, batch size, learning rate).

Step 3: Training

8
 • Code:-

!python train.py --img 640 --batch 16 --epochs 100 --data dataset.yaml --

weights yolov5s.pt

Step 4: Validation

• Evaluate the model performance using metrics like mAP (mean Average
Precision).

Step 5: Inference

• Test the model on new images or videos.

• Code:-
!python detect.py --weights best.pt --img 640 --conf 0.4 --source data/images/

1.2.4 DATA AUGMENTATION IN YOLOv5

With each training batch, YOLOv5 passes training data through a data loader, which
augments data online. The data loader makes three kinds of augmentations:

• Scaling.

• Color space adjustments.

• Mosaic augmentation.

The most novel of these being mosaic data augmentation, which combines four images
into four tiles of random ratio.

The mosaic data loader is native to the YOLOv3 PyTorch and now YOLOv5 repo.

Mosaic augmentation is especially useful for the popular COCO object

detection benchmark, helping the model learn to address the well known "small object
problem" - where small objects are not as accurately detected as larger objects.

It is worth noting that it worth experimenting with your own series of augmentations to
maximize performance on your custom task.

1.2.5 AUTO LEARNING BOUNDING BOX ANCHORS

In the YOLOv3 PyTorch repo, Glenn Jocher introduced the idea of learning anchor
boxes based on the distribution of bounding boxes in the custom dataset with K-means

9
and genetic learning algorithms. This is very important for custom tasks, because the
distribution of bounding box sizes and locations may be dramatically different than the
preset bounding box anchors in the COCO dataset.

In order to make box predictions, the YOLOv5 network predicts bounding boxes as
deviations from a list of anchor box dimensions.

The most extreme difference in anchor boxes may occur if we are trying to detect
something like giraffes that are very tall and skinny or manta rays that are very wide
and flat. All YOLO anchor boxes are auto-learned in YOLOv5 when you input your
custom data.

The following code illustrates an example of anchors that have been learned from
training data in a YOLOv5 configuration file.

# parameters

nc: 80 # number of classes

depth_multiple: 0.33 # model depth multiple

width_multiple: 0.50 # layer channel multiple

# anchors

anchors:

- [116,90, 156,198, 373,326] # P5/32

- [30,61, 62,45, 59,119] # P4/16

- [10,13, 16,30, 33,23] # P3/8

# YOLOv5 backbone

1.2.6 16 BIT FLOATING POINT PRECISION

The PyTorch framework allows the ability to half the floating point precision in training
and inference from 32 bit to 16 bit precision. When used with YOLOv5, this
significantly speeds up the inference time of models.

However, the speed improvements are only available on select GPUs at this point -

10
namely, V100 and T4. That said, NVIDIA has written intent to improve their coverage
of this efficiency boost.

1.2.7 NEW MODEL CONFIGURATION FILES

YOLOv5 formulates model configuration in .yaml, as opposed to the .cfg files used in
Darknet. The main difference between these two formats is that the .yaml file is
condensed to just specify the different layers in the network and then multiplies those
by the number of layers in the block.

The new ‘.yaml’ format looks like the following:

# parameters

nc: 80 # number of classes

depth_multiple: 0.33 # model depth multiple

width_multiple: 0.50 # layer channel multiple

# anchors

anchors:

- [116,90, 156,198, 373,326] # P5/32

- [30,61, 62,45, 59,119] # P4/16

- [10,13, 16,30, 33,23] # P3/8

# YOLOv5 backbone

backbone:

# [from, number, module, args]

[[-1, 1, Focus, [64, 3]], # 0-P1/2

[-1, 1, Conv, [128, 3, 2]], # 1-P2/4

[-1, 3, BottleneckCSP, [128]],

[-1, 1, Conv, [256, 3, 2]], # 3-P3/8

[-1, 9, BottleneckCSP, [256]],

11
 [-1, 1, Conv, [512, 3, 2]], # 5-P4/16

[-1, 9, BottleneckCSP, [512]],

[-1, 1, Conv, [1024, 3, 2]], # 7-P5/32

[-1, 1, SPP, [1024, [5, 9, 13]]],

# YOLOv5 head

head:

[[-1, 3, BottleneckCSP, [1024, False]], # 9

[-1, 1, Conv, [512, 1, 1]],

[-1, 1, nn.Upsample, [None, 2, 'nearest']],

[[-1, 6], 1, Concat, [1]], # cat backbone P4

[-1, 3, BottleneckCSP, [512, False]], # 13

[-1, 1, Conv, [256, 1, 1]],

[-1, 1, nn.Upsample, [None, 2, 'nearest']],

[[-1, 4], 1, Concat, [1]], # cat backbone P3

[-1, 3, BottleneckCSP, [256, False]],

[-1, 1, nn.Conv2d, [na * (nc + 5), 1, 1]], # 18 (P3/8-small)

[-2, 1, Conv, [256, 3, 2]],

[[-1, 14], 1, Concat, [1]], # cat head P4

[-1, 3, BottleneckCSP, [512, False]],

[-1, 1, nn.Conv2d, [na * (nc + 5), 1, 1]], # 22 (P4/16-medium)

[-2, 1, Conv, [512, 3, 2]],

[[-1, 10], 1, Concat, [1]], # cat head P5

12
 [-1, 3, BottleneckCSP, [1024, False]],

[-1, 1, nn.Conv2d, [na * (nc + 5), 1, 1]], # 26 (P5/32-large)

[[], 1, Detect, [nc, anchors]], # Detect(P5, P4, P3)

1.2.8 YOLOv5 LABELING FORMAT: YOLOv5 PyTORCH TXT

The YOLOv5 PyTorch TXT annotation format is similar to YOLO Darknet TXT but
with the addition of a YAML file containing model configuration and class values.

Depending on the annotation tool you use, you'll need to make sure to convert
annotations to work with YOLOv5. Roboflow allows you to input 27 different labeling
formats and export them to work with YOLOv5. If you uses VOTT, LabelImg, CVAT,
or another tool, you can convert those labels to work with YOLOv5.

Ultralytics, the creator of YOLOv5, partners with Robflow as the suggested YOLOv5
labeling tool.

1.2.9 ADVANTAGES OF YOLOv5

• Real-Time Detection: Extremely fast inference speeds, suitable for live

applications.

• High Accuracy: Improved feature extraction and fusion techniques.

• Easy to Use: Pre-trained models and PyTorch implementation simplify

deployment.

• Scalability: Multiple model variants for different computational needs.

• Custom Training: Easy to fine-tune for custom datasets (e.g., PCB defect
detection).

1.2.10 LIMITATIONS OF YOLOv5

• Struggles with detecting tiny objects in high-resolution images.

13
• Might require GPU resources for efficient performance.

• Performance depends heavily on hyperparameter tuning and dataset quality.

14
 CHAPTER 2
Literature Review
There is certainly an unprecedented demand in the world currently for avalanche
bandwidth and a high data scale that is crucial for all of these. In order to do this, RoF
automation is known as most effective notable technology, here for transmission of data
of dm method is employed in which data is delivered parallel rather than serial which
reduces the inter mark references and it is less prone to attenuation. When contrasted to
FSO-based optical wireless connections, RF does not make budgetary sight for business
operators or corporations noticing to broaden optical mesh.

S. No REFERENCES FINDINGS LIMITATIONS

1. Bekkali, A., Naila, C.B., Transmitting The Ro-FSO
Kazaura, K., Wakamori, K. and performance of a system efficiency is
Matsumoto, M., 2010. dm signal over a very much
Transmission analysis of an turbulent free sensitive for the
OFDM-based wireless service space optics atmospheric
over turbulent radio on FSO channel. turbulent flow and
links modelled by gamma– the optical power
gamma distributions. IEEE
photonics journal, 2(3), pp.510-
520.[11]

2. Armstrong, J., 2009. OFDM for Using dm systems . Higher peak to

optical to obtain good typical power ratio
communications. Journal of performance in plus sensitivity to
light wave technology, 27(3), optical systems. rate of recurrence
pp.189-204.[12] offset and the stage
noise.

3. Nistazakis, H.E., Stassinakis, Bit error rate Worst bit error rate
A.N., Muhammad, S.S. and performance of performance
Tombras, G.S., 2014. BER multi hop RO- corresponds to the
estimation for multi-hop Ro- FSO systems guidelines of the
FSO QAM or PSK OFDM comprises of GG biggest modulation

15
 communication systems over distribution. format used.
gamma-gamma or exponentially
modelled turbulence
channels. Optics & Laser
Technology, 64, pp.106-
112.[13]

The basic transmission efficiency of an OFDM indicator over the violent Ro-FSO
channel over average CNDR, BEP and likelihood of failure conditions. The author [11]
conferred a mathematical base for optimizing OFDM, Ro-FSO linkage and close-
formed expressions for device BEP and the likelihood of failure, taking into
consideration the optical noise and nonlinear enforcement.

Describing a typical transmitting side and receiving side of OFDM, and describe the
transmission functions for primary signal blocks. Signals having time domain and
frequency are described in the system at variable factors. This particular process is
described by considering the result of a simple two-way channel on the signal
transmitting element due to the fact that the subcarrier frequency has gone, OFDM is
an extremely guaranteed optical communication technology, but there are some further
restrictions that have opened up for many new and interesting research platforms.[12]

We extracted all enclosed mathematical emphasis for the specific valuation of BER for
every possible compact, for each entity Ro-FSO hyperlink and also for the complete
mathematical appearance of the aggregate multi-hop system. Author inspected all the
impact of subcarrier numbers on the OFDM, the modulation format, the nonlinearity of
the LD, the optically modulated catalogue, CNDR values and the cardinal of nodes
(relays), the pursuance of the BER for individual Ro-FSO hyperlink entity and the
entire relayed optical wireless link. The particular expressions obtained may be used
for evaluation of the Bit error rates plus the design of any possible Ro-FSO quadrature
amplitude modulation, PSK OFDM or relay system.[13]

S. No REFERENCES FINDINGS LIMITATIONS

4. Dabiri, M.T, Saber, M.J. and Gamma Evaluating the Bit
Sadough, S.M.S., 2016, distribution and error rate
September. BER performance pointing error is performance for

16
 of OFDM-based wireless studied. FSOOFDM system
service over radio-on-FSO having turbulence
links in the presence of effects and pointing
turbulence and pointing errors.
errors. In 2016 8th
International Symposium on
Telecommunications
(IST) (pp. 30-33). IEEE.[14]

5. Nistazakis, H.E., Stassinakis, The Ro-Fso Not really extended

A.N., Sandalidis, H.G. and scenario that use d for considering more
Tombras, G.S., 2014. QAM the orthogonal complicated radio
and PSK OFDM Ro-FSO frequency division over fiber space
Over $ M $-Turbulence multiplexing optics problems.
Induced Fading technique is
Channels. IEEE Photonics evaluated
Journal, 7(1), pp.1-11.[15]

6. Varotsos, G.K., Nistazaki, Extracted the Increased ABER and

H.E. and Tombras, G.S., mathematical OP values.
2017. OFDM Ro-FSO links declarations for
with relays over turbulence the metrics
channel and nonzero associated with
boresight pointing errors. J.
the Probability of
Communication, 12(12),
outage (OP) plus
pp.644-660.[16]
the Average Bit
Error Rate

(ABER) of
analyzed Ro-FSO
implementations
along with the DF

Relays.

Here, in this paper[14] the study of the characteristics of a Ber-dependent OFDM

17
building with quadrature luxurious modulation, the optical light ray activity of the FSO
section is disturbed by atmospheric turbulence and pointing errors. Obtained a book
close to the form of the Ro-FSO M-QAM OFDM model, all atmospheric interruption
and the pointing failure. In this work, the consequence of a directing error is ignored,
while in some of the Ro-FSO apps, trans receivers are usually on the highest frame to
get a kind of sight. In this type of systems, the imbalance caused by the building sway,
can weaken the performance of the hyperlink.

The specific attainment of OFDM signals in distinctive propagation tracks has been
tested here. The Standard subject are incorporated, where results of OFDM signal
among turbulence with gamma distributions have been analysed, here the previous
work on the multi-hop Ro-FSO framework presupposing Gamma-Gamma distributions
and turbulence modulating exponentially, Here, a high-altitude platform (HAP) mesh
was used to transmit Wi-MAX to much of Earth's space. However, the best of the
author's recognition is that the assessment of model M-turbulence is not available for
open specialized materials, rather than the value of that distribution.[15]

The specific joint impact of both Malaga modelled turbulence and non-zero foresight
pointing faults on execution of a multi-hop DF relayed to the N-subcarrier L-QAM
OFDM Ro-FSO system is investigated. As a result, accurate enough closed expressions
are extracted for the essential ABER and the outage probability performance metrics
having such Ro-FSO configurations.[16]

S. No References Findings Limitations

7. Nino, M.P.., Nistazakis, H.E, Average BER and Does not perform
Leitgeb., E. and Tombra’s, G.S., the probability of well in adverse
2019. Spatial diversity for QAM outage is atmospheric
OFDM Ro-FSO link with estimated. turbulence
nonzero boresight pointing error
over atmospheric turbulence
channel. Journal of modern
optics, 66(3), pp.241-251.[17]

8. Kashani, M.A, Uysal., M. and Closed form Gamma-Gamma

Kavehrad, M., 2015. A novel remark model is not

18
 characteristic channel model for for a disruption suitable for far
turbulence induced fading in probability and ranges.
free-space optical the standard bit
systems. Journal of Light wave error rate
Technology, 33(11), pp.2303- expressions of an
2312.[18] FSO structure
over turbulence
route is studied.

9. AlQuwaie, H., Ansari, I.S. and A brand new Capacity gets

Alouini, M.S., 2015. On the consolidate model reducing as the
performance of free-space that will detail for pointing errors
optical communication systems the specific value is becoming
over double generalized gamma impacts of high.
channels. IEEE journals on directing errors
selected fields in and kind of
communications33(9), pp.1829- receiving metal
1840.[19] detector used.

The author here [17] represent good OFDM Ro-FSO conversation system with the
spatial diversity of the MRC on the receiving side is being tested. It is absolutely worth
noting that particular spatially diverse Ro-FSO hyperlink uses a number of laser sources
precisely where all of them associated with it are directed to a particular receiver. In
this manner, the particular optical dissipation collected for specific variety portion is
awaited to increase considerably. Thus, the efficiency of the hyperlink is evaluated in
provision of the mean bit rate as well as its failure probability (OP) when the
transmission is not appropriate for both barometric disturbances and boresight pointing
errors having non-zero values.

The proposed thesis [18] tells completely new and deterministic reporting model known
as Dual Generalized Gamma (Double G) is used here for any particular fluctuation in
irradiance. The particular model used is usually validated for all ranges of disturbance
conditions (either weak or vigorous) and includes almost of current record figures for
specific irradiance fluctuations in literature having particular cases. In addition, the

19
evaluation of the presented design includes models named Gamma-Gamma and
Double-Weibull. This architecture also defines a good perfect match for simulation
parameter and is not twice as bigger to the other two versions, which appear a difference
from simulation details under several conditions.

The Pdf File of a specific irradiance below the effect associated with directing errors
greater than the dual GG disturbance station. The PDF document matches the results of
Kashaniaussi que al. in all possible ways, whenever the error handling techniques have
absolutely no effect below the IM / DD recognition method. As a result, we extracted
unified plus closed-form expressions taking into account statistical attributes associated
with a particular end-to - end SNR associated with I a sole FSO link (ii) a mixed RF-
FSO swap transmitting system, which was identified by pointing errors directly for
consideration. In addition, all the link efficiency offered is evaluated for each technique
by providing closed-form expressions.[19]

S. No REFERENCES FINDINGS LIMITATIONS

10. Prabu, K. and Thakkar, S., 2018. Extracted a closed The performance is
Analysis of FSO link with time type statement of altered using the
diversity over M-distribution a particular FSO technique of time
channel model with pointing hyperlink over diversity.
errors and GVD effects. Optics Malaga-
Communications, 421, pp.115- distribution
124.[20] design with
pointing mistakes
including GVD
results.

11. Prabu, K., Gupta, S. and Jaiswal, Misalignment Performance is

S., 2018. Impact of pointing error effects is degraded by using
errors and turbulence effects on used. pointing errors.
POLSK and coherent OWC-
based FSO system over
generalized turbulence channel
model. Photonic Network
Communications, 36(1), pp.96-

20
 105.[21]

12. Duan, M., Wang, P., Liu, X., Li, The ABER Low LDPC rate
Y., Chen, W. and Li, A., 2018. performance is obtains better
Average bit error rate analyzed. ABER
performance analysis of a low- performance.
density parity-check-coded
orthogonal frequency-division
multiplexing FSO system under
Málaga distribution considering
atmospheric attenuation and
pointing errors. Applied
optics, 57(19), pp.5505-
5513.[22]

The specific M-distribution channel design is designed to show barometric turbulence,

and\characteristics is enhanced by various strategies. In this, the chances of fade on-off
keying (OOK) based FSO set up with the future diversity is analysed by pointing errors,
including the effect of GVD over the Malaga dispensation model.[20]

The specific analytical expressions for the combined distribution associated with
scintillation and top errors have currently been achieved, additionally the centred
moments linked to the specific general possibility of submission are generally derived.
Standard Bit Error Cost (ABER) and Decode-and-Forward based multi-hops website
and site FSO communication program making use of combined path reduction results,
misalignment errors (electronic., fading) plus atmospheric turbulence-induced drop-
down casting simply by the Meters distribution are usually reflected in fine detail.[21]

The particular performance of A.B.E.R is usually analysed for different turbulence

strength values, weather surrounding, normality ray broadness and jitter, disparate
values of the assessment of Malaga values for the Δ parameter. The Mont Carlo
simulation is used to check the faultlessness of the suggested A.B.E.R version. In
addition, unique codes are summarized to enhanced system performance parameters.
The particular result shows that, in terms of equal disturbance, climate and directing
error problems, even if or not the L.D.P.C unique cipher are taken into narration or not,

21
the work of the 16-QAM OFDM is generally no better than the 16-PSK OFDM than
that of the Malaga falling channel, as well as the ALLERDINGS output decline along
with the higher one in respect of both modulation strategies.[22]

S. No REFERENCES FINDINGS LIMITATIONS

13. Vaishali and Sancheti, S., 2018. Handles the To construct a
Investigation between impairments reliable and
Performances of Free Space which are pervasive channel.
Optical Communication Links required to be
Under Atmospheric resolved for the
Turbulence. JCM, 13(7), formation of
pp.368-374.[23] link.

14. Aminikashani, M., Kavehrad, M. Performance of SIMO system for

and Gu, W., 2016, February. FSO systems is FSO are better than
Error performance analysis of examined. other atmospheric
FSO links with equal gain turbulence models.
diversity receivers over double
generalized gamma fading
channels. In Broadband Access
Communication Technologies
X (Vol. 9772, p. 97720R).
International Society for Optics
and Photonics.[24]

15. Sahota, J.K. and Dhawan, D., Coherent-based

2018. Reducing the effect of Homodyne High scintillation
scintillation in FSO system using detection is value
coherent based homodyne proposed for
detection. Optik, 171, pp.20- modified
26.[25] systems.

The aim of this review [23] is to scrutinize the dominant gaps present as obstruction in
the line of designing an optimized FSO constipuent. This paper shows the extreme
endowment and impairments that need to be addressed for the settlement of the FSO

22
linkages. Major output parameters are deliberate, such as performance of error bit rate,
optimization of failure proportion, distinct modulation style, increase in beam analysis,
spatial diversity, mitigation moderation, etc. that deal straight with transmission
performance, channelling and receiving segment of the FSO system.

The distribution of the specific amount of Double GRVs is a good analytical solution
thus, we have obtained a novel best bound for submission of the amount of Double G
dispersed RVs. The multiplication of Dual GG distributed RVs is first achieved for each
particular distribution. Then, on the basis of well-known inequities for both arithmetic
and geometrical means, the closed formed upper bound collaboration, together with the
specific distribution, produced a specific amount of Double GG distributed RVs. These
distribution properties help us to study performance of error rates in the FSO
system.[24]

A brand new FSO system continues to be used to reduce the effect of scintillation during
transmission. The adapted program consists of the Homodyne method used in the
receiving area of Free Space Systems. Results have shown that the decrement of the bit
error value is substantially low after adopting Homodyne detection. Scrutiny of the
assessment, both direct and consistent. Homodyne detection is done, through which we
verify that Homodyne is very capable to use rather than a direct way. The outcomes
also show that BER values for the Direct Detection procure was 3.33e −1 where, with
respect to the detection of homodyne, the achieved value was 1.55e−5 for medium
strength scintillation.[25]

S. No REFERENCES FINDINGS LIMITATIONS

16. Yang, L., Alouini, M.S. and Performance of a Value of BER
Ansari, I.S., 2018. two-way differs for gamma
Asymptotic performance communicating fading channels
analysis of two-way relaying systems directing
FSO networks with nonzero errors having non
boresight pointing errors zero values over
over double-generalized Double
gamma fading
generalized
channels. IEEE Transactions
fading stations.
on Vehicular

23
 Technology, 67(8), pp.7800-
7805.[26]

17. Chatzidiamantis, N.D., BER and Outage Proposed a

Sandalidis, H.G., Probability statistical new
Karagiannidis, G.K., Channel model
Kotsopoulos, S.A. and (Double-Weibull)
Matthaiou, M., 2010, April. which does not
New results on turbulence describe irradiance
modeling for free-space fluctuations in
optical systems. In 2010 weak turbulence.
17th International
Conference on
Telecommunications (pp.
487-
492). IEEE.[27]

18. Jurado-Navas, A., Garrido- Different statistical Gamma-Gamma

Balsells, J.M., Paris, J.F., model pdf does not fit with
Puerta-Notario, A. and simulation data.
Awrejcewicz, J., 2011. A
unifying statistical model for
atmospheric optical
scintillation. Numerical
simulations of physical and
engineering
processes, 181.[28]

Author proposes a channel medium, known as Double-Weibull, to explain irradiant

variations in Frequency bands for strong and moderate turbulence effects.
The stochastic strategy used is positioned on scintillation thesis and is extracted from
two random variables of Weibull. Closed-form probability distributions as well as
cumulative density activity are presented for Meijer 's G-operations.[26]

In this, the comparison takes place between the current model and the classic gamma

24
model, from which we analyse the accuracy through a range of counterfeit where
together plane and spherical patterns are recognized. Lastly, the execution of the Fso
system on the Double-Weibull disturbance approach is figured out, they also derived
the terms for the bit error rates, considering Intensity-modulation having on off key.[27]

In this author develops a new statistical illustration for irradiate variations of

confounded optical wave front transmitting over a turbulent medium in which most of
irradiance fluctuations fall under homogeneous and isotropic turbulences. For an
unbounded optical wave front under all turbulence conditions it brings out closed form
and mathematical versatile terms for principal channel statistics which make this
approach as main advantage of this technique. Moreover, it consolidates
maximum proposed statistical models for their bibliography-derived radiance
fluctuations, providing, an addition, excellent agreement with that of the
experiment.[28]

S. No REFERENCES FINDINGS LIMITATIONS

19. Prabu, K., Kumar, BER, Channel Performance of outage
D.S. and Srinivas capacity and probability is not good in
T., 2014. Performance Outage DPSK.
analysis of FSO links under probability
strong atmospheric
turbulence conditions using
various modulation
schemes. Optik-
International Journal for
Light and Electron
Optics, 125(19), pp.5573-
5581.[29]

20. Singhal, P., Gupta, P. and BER Alamouti-coded system

Rana, P., 2015, April. Basic has less BER performance
concept of free space optics than repetition-coded.
communication (FSO): An
overview. In 2015

25
 International Conference
on Communications and
Signal Processing (ICCSP )
(pp.
0439-0442). IEEE.[30]

21. Jagadeesh, V.K., Channel Derived

Palliyembil, V. capacity and closed form
Muthuchidambaranathan, P. Bit error rate expression
and Bui, F.M., 2015. Free performance
space optical is not good
communication using for
subcarrier intensity moderate
modulation through and strong
generalized turbulence turbulence
channel with pointing error. conditions.
Microwave and Optical
Technology Letters,57(8),
pp.1958-1961. [31]

In Today's world, wireless technology is becoming a leading technique for wireless

networks, FSO is a system which utilizes observable to near-infrared (NIR) light
propagates through the space to send data. FSO attract a lot of attention as advanced
engineering technology to overcome the last mile of bottle - neck obstacles in locally
area access structures due to its higher bandwidth. Minor cost required in non-licensed
spectrum, comparably less power utilization and safety compared to RF systems. This
thesis carried out a research of the different atmospheric variation of the FSO channel.
This research involves the analysis of fog models like Kim Model and Kruse Model, as
well as the analysis of wet and dry snow over FSO linkages.[29]

Atmospheric turbulence is a remarkable weakening component in FSO systems. Spatial

diversity approaches offer a reassurance technique to mitigating turbulence-induce
fadding. In the whole work, the author examines the inaccurate output of the FSO
interface with spatially diversity on the various turbulence platforms characterize in the
Double Generalized section. It is also a modern generalized mathematical model

26
enveloping every turbulence situation. Here the composer presumes the intensity of
modulation/direct exposition with switching keying and BER values examine of the
SIMO, MISO and MIMO FSO frameworks on this modern channel model.[30]

Here are un-wired frameworks proficient of giving high-pace, reliable and value
competent communication among two individuals. The channel for free space
channel is interpreted in provision of the Malaga distribution. It is a standardized
representation that integrates both the reputation of the channel and the earlier versions.
Throughout the course of this analysis, the efficiency of the SIM FSO system is
evaluated by taking into accounts the adverse consequence of path fluctuations and point
absurdity. The termination shows the devaluation of channel performance of system and
also how the m-distribution model can be synthesized.[31]

S. No REFERENCES FINDINGS LIMITATIONS

22. Esmail, M.A., Fathallah, H. and Attenuation in Attenuation
Alouini, M.S., 2016.An FSO increase as dust
experimental study of FSO link Communicatio particles increases
performance in desert environment n System
IEEECommunicationsLetters,20(9
), pp.1888-1891. [32]

23. Peppas, K.P., 2011. A simple, Outage Simple closed form

accurate approximation to the sum probability and approximation, pdf
of Gamma–Gamma variates and Bit error rate is of GG is efficiently
applications in MIMO free-space investigated. by α-μ distribution
optical systems. IEEE Photonics
Technology Letters, 23(13),
pp.839-841. [33]

24. Tang, Y., Zhou, X., Zhang, Z. and An optimal bit Analysis was not
Tian, Q., 2011, September. detection made for moderate
Performance analysis of a two-way technique based and weak
network-coded free space optical on Bayesian turbulence
relay scheme over strong estimation for conditions.
turbulence channels. In 2011 IEEE

27
 Vehicular Technology Conference receiving nodes
(VTC Fall) (pp. 1-5). IEEE [34] in the
disappearance
of Chanel State
information is
formed.

FSO contact systems are over-whelmed by airborne dust substances in arid and semi-
arid provinces. The appearance of these fragment in the sky significantly impacts
optical linkage, diminished its compatibility and provoke a power disruption.
Consequently, they examine the output of the FSO in this regard under dust blizzard.
The author constructs a chamber to contained this explicit habitat and to accomplish
computation. From the empirical observations, experiential formulate and suggest an
empirical consolidation for weakening of the represented as a part of the extent of
discernability. The corollary spectable adequate output for FSO interactions, beneath
medium and low dust, with a possible extension of hundreds of meters to slight
kilometres.[32]

Here author planned a simple, reliable, close-form analogy to the probability density
function of aggregate of complimentary, normally distributed gamma-gamma (G-G)
probability distributions. The PDF of the G-G sum can be shown to be efficient
estimated by the PDF of the 5-007-μ distribution. On the basis of this approach, simple,
accurate approximations for significant performance benchmarks of multi-input multi-
output systems instalment over G-G fading’s are displayed. The feasibility of system
method is composed of a variety of numerically evaluated and computer simulation
outcomes.[33]

Here, we suggest the concept of a network-coded cooperation (Net CC) sending to

layout for FSO and TWRN chain. Achieved network-coded spatial diversity
improvements will effectually improve the accuracy of FSO communication over heavy
turbulence channels. In contrast to Traditional presumption of full channel state (CSI)
suggestions previously identified at destination, acquiring a closely-form contrivance
for this by communicative benefits of Meijer's G-function, defined by a gamma-gamma
heavy per- tube ration prototype.[34]

28
 S. No REFERENCES FINDINGS LIMITATIONS
25. Sandalidis, H.G., Tsiftsis, T.A. Error Beam wander and
and Karagiannidis, G.K., 2009. performance of tilt consideration
Optical wireless the heterodyne effects are
communications with DPSK and neglected.
heterodyne detection over optical wireless
turbulence channels with communications
pointing errors. Journal of light
system operating
wave technology, 27(20),
under different
pp.4440-4445. [35]
intensity
fluctuations is
studied. The new
Closed-form
features for
estimates of
random
deformation of
the channel of
propagation are
imitative and the
implementation
of the bit-error
rate (BER) is
examined for all
of the above
fading effects.

Heterodyne differential phase-shift keying optical wireless communication device

working under various intensity circumstances, Error correction equation. In particular,
the proliferating signal is believed to suffer from the combined effects of the fading,
inconsistency and path-loss-induced atmospheric turbulence. New closed-form
functions for spontaneous transmission channel reverberation statistics are examined,

29
and the bit error rate (BER) efficiency is monitored for all of the fading effects
described above. Obtained simulations for determining the error output of OW setups
with atmospheric effects and inconsistency have been shown. In addition, it is also
considered that non-linear computation determines the optimal frequency range that
achieves the optimal BER value for the average power ratio.[35]

30
 CHAPTER 3
Work Completed in the 7th Semester

3.1 PROPOSED MODEL

3.1.1 IMPROVED YOLOv5S NETWORK FRAMEWORK

The YOLOv5 method adopted YOLOv4’s CSPDarkNet53 as the backbone feature

extraction network and added some new features [24, 25]. The superior flexibility of
YOLOv5 makes it convenient for edge computing. Depending on the numbers of
feature extraction modules and convolution kernels at specific locations of the network,
the YOLOv5 family comprises four network models, namely, YOLOv5s, YOLOv5m,
YOLOv5l, and YOLOv5x. The corresponding numbers of network parameters are 7.2
× 106, 2.12 × 107, 4.65 × 107, and 8.67 × 107, and the model sizes are 16.5 Mb, 49 Mb,
109.1 Mb, and 205.7 Mb, respectively [26]. As the number of network model
parameters increases, the complexity of the basic network model increases, and the
accuracy of the model is improved, but more computational and hardware resources are
required, and inference speed decreases.

The basic YOLOv5 framework can be divided into four parts: input, backbone, neck,
and prediction. Mosaic data enhancement, adaptive anchor box operation, and picture
scaling are used to process the input dataset. The backbone mainly comprises CBS, C3,
and SPPF, which increase the local receptive field range, improve the network speed
by slicing the input image, and enhance the learning ability and ensure the accuracy of
the network while reducing the number of operations. Compared with the backbone,
the components of the neck are relatively simple and basically consist of CBS,
upsample, Concat, and C3. The FPN+PAN structure is used in the neck to integrate
different levels of features. FPN layers convey semantic features from top to bottom,
and PAN uses up- and downsampling operations to combine the low- and high-level
features. The prediction part includes bounding box loss and nonmaximum inhibition.
YOLOv5 uses GIOU_Loss [27] as its loss function. Compared with IOU_Loss [28],
GIOU_Loss solves the problem of nonoverlapping bounding boxes.

In this paper, an efficient channel attention (ECA-Net) [29] mechanism and the Focal-
EIOU loss [30] function are introduced to improve the PCB defect detection

31
performance. Moreover, the k-means++ clustering [31] algorithm is adopted to
optimize the selection of the initial clustering centre to obtain suitable anchor boxes.
The improved YOLOv5s network structure is shown in Fig3.1.

Fig 3.1 Structure of improved YOLOv5s network.

3.1.2 K-MEANS++ OPTIMIZATION

YOLOv5 uses k-means [32] to cluster the anchor boxes, but this method has certain
limitations. k-means is a heuristic method, i.e., it cannot guarantee convergence to a
global optimum, as the selection of the initial centre will directly impact the clustering
results. k-means randomly selects sample points as clustering centres, which can easily
lead to local convergence or require an increased number of iterations before
convergence is achieved. On the other hand, the initial a priori anchor frame parameters
of YOLOv5 are obtained using k-means clustering on the universal target detection
datasets. However, PCB defect types only have six categories, and the size of the a
priori frame needs to be redesigned.

Therefore, the k-means++ method is used to optimize the selection of the initial
clustering centres and obtain suitable anchor boxes for PCB defect detection. k-
means++ converts the selection of initial points into a probability problem, which
facilitates the estimation of better initial clustering centres and accelerates the method’s
convergence. The specific steps of k-means++ are as follows:

32
(a) A central point x is chosen randomly from dataset X

(b) For each data point, the Euclidean square distance D(x) from the nearest previously
chosen centroid is calculated

(c) The probability of each sampling point P(x) selected as a new cluster centre is
calculated separately using the following equation:

(1)

(d) Steps b and c are repeated until k centroids have been sampled

(e) The classical k-means algorithm is applied until there are no more clustering centre
changes

3.1.3 BOUNDARY LOSS FUNCTION

The loss function measures the distance between the predicted and the expected output
of the neural network [33, 34]. The closer the predicted output is to the expected one,
the smaller the value of the loss function. YOLOv5 uses the GIOU loss function to
calculate the box loss. However, when the height and width of the two prediction boxes
are the same and on the same level, the GIOU will degenerate into an IOU, which will
not reflect the coincidence between the two boxes accurately. In addition, the
convergence speed of GIOU is slow, and the regression is not accurate enough. In this
paper, the Focal-EIOU loss function was used instead of GIOU.

Focal-EIOU measures the discrepancies of three geometric factors in bounding box

regression. Moreover, focal loss for regression is designed to solve the problem that the
loss value fluctuates significantly due to low sample quality. The Focal-EIOU loss
function is shown in

(2)

(3)

where γ is a parameter that controls the degree of outlier inhibition, LIOU is the IOU
loss, Ldis is the distance loss, and Lasp is the aspect loss. Cw and Ch indicate the width
and height of the minimum bounding box covering the predicted and ground-truth

33
boxes.

3.1.4 ATTENTION MECHANISM NETWORK

PCB defect detection through deep learning requires effective feature extraction.
However, extracting and identifying features using generic methods is difficult because
the defects of bare PCBs are too small relative to the whole image. When high-
resolution images are used, the detection speed of convolutional networks is slow.
Hence, it is challenging to meet the real-time detection requirement of practical
applications. The attention module can learn where the feature information that needs
to be focused on is in the feature space and can guide the network to extract defect
features [35]. For example, in the SE-NET mechanism attention module [36], the input
features go through the global mean pooling channel by channel, then through two fully
connected layers, and finally through a sigmoid nonlinearity to generate the weights of
each channel. These two fully connected layers capture nonlinear interactions across
channels, which reduces dimensionality effectively. Although this strategy has
achieved good application results, dimensionality reduction inevitably causes side
effects and is not conducive to capturing the dependencies between channels.

In this work, an efficient channel attention mechanism network, namely, ECA-Net, is

introduced to improve cross-channel attention. ECA-Net is a convolutional neural
network with a novel channel attention mechanism that improves the channel attention
module of SE-NET and follows a local cross-channel interaction strategy without
dimensionality reduction, which can be implemented effectively using one-dimensional
convolution. Further, ECA-Net proposes a method to adaptively select the size of the
1D convolution kernels to determine the coverage of local cross-channel interactions,
thereby reducing model complexity while improving performance. Figure 3.2 shows
that the ECA module performs global average pooling [37] to obtain features without
dimensionality reduction. Then, the ECA module uses fast 1D convolution to capture
local cross-channel interactions, and the 1D convolution parameter k is generated
according to the size of the input channel. The sigmoid function generates the weight
ratio of each channel. Finally, all feature channels are combined using an element-wise
product. As a result, the effective features are enhanced and the ineffective features are
suppressed.

34
 Fig 3.2 Diagram of ECA module.

The morphological features are contained in feature maps of different scales, which are
extracted by the backbone network. For small targets, the foreground information in the
feature map is sparser. To focus on useful defect targets, the ECA module was
embedded into the backbone network, connected in series to the C3 module, and named
as C3-ECA. Then, the C3 module of the original YOLOv5 backbone was replaced.

3.2 EXPERIMENTS AND ANALYSIS

3.2.1 DATASET AND ENVIRONMENT

In this study, the popular public PCB defect dataset proposed by Ding et al. [38] was
used. Examples of the dataset are shown in Figure 3.3. The PCB defects in the dataset
are classified as missing holes, spurious copper, short circuits, burrs, mouse bites, and
open circuits. The PCB dataset distribution statistics are shown in Table 1.

35
Fig3.3.1 Missing hole

Fig3.3.2 Mouse bite

36
 Fig3.3.3 Open circuit

Fig3.3.4 Short

37
 Fig3.3.5 Spur

Fig3.3.6 Spurious copper

38
Defect type Missing Mouse Open Short Spur Spurious
hole bite circuit copper

No. of 115 115 116 116 115 116

images

No. of 497 482 482 491 488 503

defects

Table 3.1 Category of the PCB dataset.

The resolution of each raw image in the dataset was 2777 × 2138. The amount of data
in the original dataset was relatively small, which easily leads to overfitting during
training, so data augmentation techniques were adopted before training. The images
were then cropped into 600 × 600 subimages to form the training, validation, and test
sets with 6720, 2881, and 1067 images, respectively. The initial value of the learning
rate is 0.01 and was reduced using the cosine annealing strategy. The batch size was set
to 16, and a total of 300 epochs were used for training. In addition, image-level data
augmentation methods were also used, such as the augment HSV, random affine,
mosaic, image scale, and other transforms. Our experiments were performed on a Linux
4.15.0-142-generic Ubuntu 18.04 with Intel quad-core i7-7700HQ, 2.80 GHz CPU,
16 GB memory, and NVIDIA GeForce GTX1080Ti (11 GB).

3.2.2 EVALUATION METHOD

The evaluations were carried out to verify the effectiveness of the proposed model.
The main indicators are selected: precision (P), recall (R), average precision (AP), and
mean average precision (mAP). The corresponding equations are as follows:

( 4)

39
where Tp represents the number of defects correctly recognized by the detection
model, FP represents the number of incorrect or unrecognized defects, FN expresses the
number of mistakenly detected targets, P is the precision, and R is the recall. The
average precision AP represents the integral of the precision rate to the recall rate. The
average of AP of all categories is the mean average precision mAP, which measures the
performance of the entire model.

3.2.3 EXPERIMENTAL RESULTS

Ablation experiments were carried out to verify the effectiveness of the presented
method using three groups of networks. The objective evaluation and comparison
results of the test model are shown in Table 3.2.

Metric Test set [email protected] (%) Up (%)

Original YOLOv5 1,067 95.29 0

+ k-means++ 1,067 95.61 0.32

Loss function

+ EIOU 1,067 96.54 0.93

+ SIOU 1,067 96.78 1.17

+ Focal-EIOU 1,067 96.97 1.36

Attention mechanism

+ SE 1,067 98.68 1.71

+ ECA 1,067 98.74 1.77

+ C3-ECA 1,067 99.10 2.13

Table 3.2 Ablation study on improved YOLOv5.

40
The original YOLOv5 defect detection model has a more comprehensive receptive field
design range and achieved more general detection results on the PCB defect dataset
compared to the other highly performing defect detection models. Improved choices
were obtained for the anchor boxes using the k-means++ method, which lead to an
increase of the mAP value by 0.32 percentage points. To explain the reason for this,
YOLOv5 adopts the k-means method to cluster the anchor boxes generated using the
COCO dataset and adopts a generic method to adjust the anchor boxes during the
training process. However, the first thing to be completed before k-means clustering is
to initialize the cluster centres. At the same time, it is precisely this reason that acts as
a shortcoming of the k-means clustering algorithm, i.e., that convergence depends
heavily on the initialization of the cluster centres. The adoption of the k-means++ can
alleviate this problem effectively and improves the detection accuracy and effect to
some extent. In addition, the effects of difference loss functions such as EIOU [30],
SIOU [39], and Focal-EIOU were investigated experimentally, and it was found that
the effect of Focal-EIOU was the best, as it increased mAP by 1.36 percentage points.
Due to the enhanced contribution of the most probably anchor box in the model’s
optimization, Focus-EIOU has stronger robustness to small datasets and noisy boxes
and yielded the best mAP results compared with other loss functions such as GIOU,
EIOU, and SIOU. The attention mechanism module also improved the performance of
the model. Similar to adding an attention module such as SE or ECA in front of the
SPPF layer in the backbone, ECA was embedded directly in the C3 module, and the
combined module was named C3-ECA. The results of the ablation experiment showed
that C3-ECA led to the most obvious improvement of the mAP value, at 2.13
percentage points.

Figure 3.4 shows that the performance of the model is same across training and
validation, which suggests that the model does not overfit the data. It can be seen from
the results that the loss function value had a downward trend during the training process.
Before the training batch reached 20, the loss function value dropped rapidly, while the
precision, recall rate, and average accuracy rapidly improved. On the other hand, the
model has a high precision and a high recall, which suggests that the model has low
false negatives.

41
 Fig3.4 Training and validation curves of the proposed method.

Figure 3.5 shows the precision, recall, F1, and P-R curves of the proposed model. It can
be seen that the improved model achieved good recognition results for all types of
defects. The P-R curves of the corresponding categories for the six defect types in the
dataset are depicted. It can be seen from the figure that the average AP value of the
method reaches 99.1%, which shows excellent detection performance.

42
Fig3.5.1 Precision-confidence curve (Testing results of proposed method)

43
Fig3.5.2 Recall-confidence curve (Testing results of proposed method)

44
Fig3.5.3 F1-confidence curve (Testing results of proposed method)

45
 Fig3.5.4 Precision-recall curve (Testing results of proposed method)

To further verify the detection effect, the proposed method was compared on the PCB
dataset with advanced object detection methods, including Faster R-CNN [40], SSD
[41], RetinaNet [42], YOLOv4 [43], and YOLOv5s. Faster R-CNN is a representative
two-stage deep learning object detector. SSD draws on the anchor mechanism of Faster
R-CNN and the end-to-end one-step structure of the YOLO method, in which object
classification and location regression are performed directly in the convolution stage.
RetinaNet is a unified network consisting of a backbone CNN network and two task-
specific subnets, namely, a classification subnet and a box regression subnet. YOLOv4
is a real-time object detection model published in April 2020 that achieved state-of-the-
art performance on the COCO dataset. The comparison results for the accuracy, speed,
and model size are shown in Table 3.

46
Method [email protected] (%) [email protected]:.95 (%) Speed (FPS) Parameters (MB)

Faster R-CNN 95.32 64.5 7 528

SDD 84.67 55.7 41 149

RetinaNet 89.7 61.2 23 157

YOLOv4 97.63 63.3 62 248

YOLOv5s 98.72 66.1 87 14.5

Proposed method 99.1 69.3 86 17

Table 3.3 Comparison of methods on the PCB defect dataset.

It can be seen in Table 3.3 that, compared with the traditional methods, the proposed
model has obvious advantages and achieves a higher FPS rate. First, the k-means++
clustering method is used to determine the anchor box size of the PCB defect dataset
and improve the accuracy of the model for small object defect positioning. Second, the
Focal-EIOU loss function is adopted to replace the original confidence loss and
compensate for the imbalance between positive and negative sample classes. Finally,
the C3-ECA module is added to the optimized YOLOv5 network to improve the
weights of the more important feature channels. Through the above improvements, the
optimized model’s mAP increased by 3.78, 14.43, 9.4, 1.47, and 0.38 percentage points
compared to Faster R-CNN, SDD, RetinaNet, YOLOv4, and YOLOv5s, respectively,
and the detection FPS meets the real-time requirements of engineering applications.

47
 REFRENCES
1. Popoola, W.O., Ghassemlooy, Z., Allen, J.I.H., Leitgeb, E. and Gao, S., 2008.
Free-space optical communication employing subcarrier modulation and spatial
diversity in atmospheric turbulence channel. IET optoelectronics, 2(1), pp.16-
23.

2. Nistazakis, H.E., Assimakopoulos, V.D. and Tombras, G.S., 2011. Performance

estimation of free space optical links over negative exponential atmospheric
turbulence channels. Optik, 122(24), pp.2191-2194

3. Sandalidis, H.G., Tsiftsis, T.A., Karagiannidis, G.K. and Uysal, M., 2008. BER
performance of FSO links over strong atmospheric turbulence channels with
pointing errors. IEEE Communications Letters, 12(1), pp.44-46.
4. A. G. Bell, “On the production and reproduction of sound by light”, American
Journal of Science, vol. 118, no. Third Series, pp. 305–324, October 1880.
5. Anbarasi, K., Hemanth, C. and Sangeetha, R.G., 2017. A review on channel
models in free space optical communication systems. Optics & Laser
Technology, 97, pp.161-171

6. Armstrong, J., 2009. OFDM for optical communications. Journal of light wave
technology, 27(3), pp.189-204

7. Tsiftsis, T.A., Sandalidis, H.G., Karagiannidis, G.K. and Uysal, M., 2009.
Optical wireless links with spatial diversity over strong atmospheric turbulence
channels. IEEE Transactions on Wireless Communications, 8(2), pp.951-957

8. Kaushal, H. and Kaddoum, G., 2017. Optical communication in space:

challenges and mitigation techniques. IEEE communications surveys &
tutorials, 19(1), pp.57-96
9. Amphawan, A., Chaudhary, S. and B Gupta, B., 2015. Secure MDM-OFDM-
Ro-FSO System Using HG Modes. International Journal of Sensors Wireless
Communications and Control, 5(1), pp.13-18.

10. Amphawan, A., Chaudhary, S. and Chan, V.W.S., 2014. 2 x 20 Gbps-40 GHz
OFDM Ro-FSO transmission with mode division multiplexing. Journal of the
European Optical Society-Rapid publications, 9.

11. Chaudhary, S., Kapoor, R. and Sharma, A., 2019. Empirical evaluation of 4
QAM and 4 PSK in OFDM-based inter-satellite communication

48
 system. Journal of Optical Communications, 40(2), pp.143-147.

12. Yi, X., Yao, M. and Wang, X., 2017. MIMO FSO communication using
subcarrier intensity modulation over double generalized gamma fading. Optics
Communications, 382, pp.64-72.

13. Krishnan, P., Jana, U. and Ashok kumar, B.K., 2018. Asymptotic bit-error rate
analysis of quadrature amplitude modulation and phase-shift keying with
OFDM Ro-FSO over M turbulence in the presence of pointing errors. IET
Communications, 12(16), pp.2046-2051.

14. Bekkali, A., Naila, C.B., Kazaura, K., Wakamori, K. and Matsumoto, M., 2010.
Transmission analysis of an OFDM-based wireless service over turbulent radio
on FSO links modelled by gamma–gamma distributions. IEEE photonics
journal, 2(3), pp.510-520.

15. Armstrong, J., 2009. OFDM for optical communications. Journal of light wave
technology, 27(3), pp.189-204.

16. Nistazakis, H.E., Stassinakis, A.N., Muhammad, S.S. and Tombras, G.S., 2014.
BER estimation for multi-hop Ro-FSO QAM or PSK OFDM communication
systems over gamma-gamma or exponentially modelled turbulence

17. Dabiri, M.T, Saber, M.J. and Sadough, S.M.S., 2016, September. BER
performance of OFDM-based wireless service over radio-on-FSO links in the
presence of turbulence and pointing errors. In 2016 8th International
Symposium on Telecommunications (IST) (pp. 30-33). IEEE.

18. Nistazakis, H.E., Stassinakis, A.N., Sandalidis, H.G. and Tombras, G.S., 2014.
QAM and PSK OFDM Ro-FSO Over $ M $-Turbulence Induced Fading
Channels. IEEE Photonics Journal, 7(1), pp.1-11.

19. Varotsos, G.K., Nistazaki, H.E. and Tombras, G.S., 2017. OFDM Ro-FSO links
with relays over turbulence channel and nonzero boresight pointing errors. J.
Communication, 12(12), pp.644-660.

20. Nino, M.P. Nistazakis, H.E, Leitgeb., E. and Tombra’s, G.S., 2019. Spatial
diversity for QAM OFDM Ro-FSO link with nonzero boresight pointing error
over atmospheric turbulence channel. Journal of modern optics, 66(3), pp.241-
251.

21. Kashani, M.A, Uysal., M. and Kavehrad, M, M., 2015. A novel characteristic

49
 channel model for turbulence induced fading in free-space optical
systems. Journal of Light wave Technology, 33(11), pp.2303-2312.

22. AlQuwaie, H., Ansari, I.S. and Alouini, M.S., 2015. On the performance of free-
space optical communication systems over double generalized gamma
channels. IEEE journals on selected fields in communications33(9), pp.1829-
1840.

23. Prabu, K. and Thakkar, S., 2018. Analysis of FSO link with time diversity over
M-distribution channel model with pointing errors and GVD effects. Optics
Communications, 421, pp.115-124.

24. Prabu, K., Gupta, S. and Jaiswal, S., 2018. Impact of pointing errors and
turbulence effects on POLSK and coherent OWC-based FSO system over
generalized turbulence channel model. Photonic Network
Communications, 36(1), pp.96-105.

25. Duan, M., Wang, P., Liu, X., Li, Y., Chen, W. and Li, A., 2018. Average bit
error rate performance analysis of a low-density parity-check-coded orthogonal
frequency-division multiplexing FSO system under Málaga distribution
considering atmospheric attenuation and pointing errors. Applied
optics, 57(19), pp.5505-5513.

26. Vaishali and Sancheti, S., 2018. Investigation between Performances of Free
Space Optical Communication Links Under Atmospheric
Turbulence. JCM, 13(7), pp.368-374.

27. Aminikashani, M., Kavehrad, M. and Gu, W., 2016, February. Error
performance analysis of FSO links with equal gain diversity receivers over
double generalized gamma fading channels. In Broadband Access
Communication Technologies X (Vol. 9772, p. 97720R). International Society
for Optics and Photonics.

28. Sahota, J.K. and Dhawan, D., 2018. Reducing the effect of scintillation in FSO
system using coherent based homodyne detection. Optik, 171, pp.20-26.

29. Yang, L., Alouini, M.S. and Ansari, I.S., 2018. Asymptotic performance
analysis of two-way relaying FSO networks with nonzero boresight pointing
errors over double-generalized gamma fading channels. IEEE Transactions on
Vehicular Technology, 67(8), pp.7800-7805.

50
30. Chatzidiamantis, N.D., Sandalidis, H.G., Karagiannidis, G.K., Kotsopoulos,
S.A. and Matthaiou, M., 2010, April. New results on turbulence modeling for
free-space optical systems. In 2010 17th International Conference on
Telecommunications (pp. 487-492). IEEE.

31. Jurado-Navas, A., Garrido-Balsells, J.M., Paris, J.F., Puerta-Notario, A. and

Awrejcewicz, J., 2011. A unifying statistical model for atmospheric optical
scintillation. Numerical simulations of physical and engineering processes, 181.

32. Prabu, K., Kumar, D.S. and Srinivas, T., 2014. Performance analysis of FSO
links under strong atmospheric turbulence conditions using various modulation
schemes. Optik, 125(19), pp.5573-5581.

33. Singhal, P., Gupta, P. and Rana, P., 2015, April. Basic concept of free space
optics communication (FSO): An overview. In 2015 International Conference
on Communications and Signal Processing (ICCSP) (pp.0439-0442). IEEE.

34. Jagadeesh, V.K., Palliyembil, V., Muthuchidambaranathan, P. and Bui, F.M.,

2015. Free space optical communication using subcarrier intensity modulation
through generalized turbulence channel with pointing error. Microwave and
Optical Technology Letters, 57(8), pp.1958-1961.

35. Esmail, M.A., Fathallah, H. and Alouini, M.S., 2016. An experimental study of
FSO link performance in desert environment. IEEE Communications Letters,
20(9), pp.1888-1891.

36. Peppas, K.P., 2011. A simple, accurate approximation to the sum of Gamma–
Gamma variates and applications in MIMO free-space optical systems. IEEE
Photonics Technology Letters, 23(13), pp.839-841.

37. Tang, Y., Zhou, X., Zhang, Z. and Tian, Q., 2011, September. Performance
analysis of a two-way network-coded free space optical relay scheme over
strong turbulence channels. In 2011 IEEE Vehicular Technology Conference
(VTC Fall) (pp. 1-5). IEEE

Copy of Base Paper

51
 Publication Detail of 7th Semester(if any)

Piyush Jain, Jayanthi N, Lakshmanan M, Suyash Trivedi, Tanushree Singh, “Bit Error
Rate Analysis of K-PSK Modulation with OFDM RoFSO System over Double

52
Generalized Gamma Turbulence Channel”, Accepted in IEEE International
Conference on Advances in Computing, Communication Control and Networking –
ICACCCN (ICAC3N-20), GCET, Greater Noida, India, December 18 & 19, 2020.

53