Scribd
Upload
44 views14 pages

Scope of The Project

The document describes a project to detect electronic components like ICs, capacitors, and resistors in printed circuit board (PCB) images. It provides details on the dataset of over 14,000 annotated PCB images, data preprocessing steps like tiling to aid detection of small components, model training using YOLOv8 and SAHI for improved small object detection, and challenges faced including limited data and small component sizes. Accuracy was improved through modifications to separately train on cropped individual components.

Uploaded by

Aadithya Ramlal
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
44 views14 pages

Scope of The Project

The document describes a project to detect electronic components like ICs, capacitors, and resistors in printed circuit board (PCB) images. It provides details on the dataset of over 14,000 annotated PCB images, data preprocessing steps like tiling to aid detection of small components, model training using YOLOv8 and SAHI for improved small object detection, and challenges faced including limited data and small component sizes. Accuracy was improved through modifications to separately train on cropped individual components.

Uploaded by

Aadithya Ramlal
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 14

SCOPE OF THE PROJECT

• Detect electronic components in


Printed Circuit Board(PCB) images
• IC , CAPACITOR, RESISTOR
• Find the location of the identified
components with respect to FIDUCIAL
points present in the PCB.
DATASET

• Total image count – 14716 PCB images annotated with IC, Resistors & Capacitors.
• Image Resolution – 640 X 640 Pixels
• Number of Channels – 3 (RGB)
• DATASET
EDA

• INFERENCE:
• Resistors and Capacitors are smaller in size compared to IC’s
• There is a high intra-class variance, where components are the same but look very different
due to the large variety of types available.
• There is also low inter-class variance, where components that are different look very similar
due to similar packaging
Preprocessing

• To tackle the detection of small objects such as Resistors and Capacitors, TILING is applied
• Tiling helps to detect small objects by looking at specific locations of the image
without sacrificing the resolution.
• PREPROCESSING APPLIED:
• Tiling 2 X 2
Augmentations and WHY?

• To increase the size and diversity of the dataset, we applied various


augmentations such as:
• Horizontal and Vertical Flip
• Bounding box Flip (HORIZONTAL)
• Rotation between -8° to +8°
• FLIPS( Horizontal and Vertical ) helps the model to be insensitive to the
subject’s orientation
• ROTATION helps model to detect objects even when the camera or subject
are not perfectly aligned
AUGMENTATIONS

FLIP ROTATION
DATASET SPLIT

1. Training set - 11772


2. Validation set - 601
3. Testing set - 564

10%
Training
10%
Validation

Testing

80%
MODEL AND ITS ARCHITECTURE

• YOLO (YOU ONLY LOOK ONCE) : YOLOv8 is the newest state-of-the-art YOLO model
that can be used for object detection, image classification, and instance segmentation
tasks. YOLOv8 was developed by Ultralytics .

• We chose YOLOv8m since it gives better accuracy in a faster time


• YOLO ARCHITECTURE
RESULTS

• EPOCHS – 41
• TRAINING TIME – AROUND 14hrs
• Batch size -8
• mAP- 0.38 ( ALL COMPONENTS)
CNTD

• The predictions were not good enough for small objects (Resistors, capacitors), so we
found out an OPEN SOURCE
SAHI (Slicing Aided Hyper Inference)

LINK- SAHI
Slicing Aided Hyper Inference (SAHI), an open-source solution that
provides a generic slicing aided inference and fine-tuning process for small
object recognition. During the fine-tuning and inference stages, a slicing-
based architecture is used.
WITH AND WITHOUT SAHI
CHALLENGES FACED

1. DATASET PREPARATION:
• Image quality and lighting
-Manual
• Component variability (INTER-CLASS VARIABILITY)
-Using 3 channel (RGB)
• Overlapping components
• Similar appearance between some types of resistors and capacitors

-APPLIED PREPROCESSIING AND AUGMENTATIONS TO MAKE THE MODEL LEARN THE


FEATURES BETTER

2. Due to the small size of Resistors, the model found it difficult to learn.
3. More Training time and lack of GPU due to the size of dataset.
-Training in parts
4. Limited DOMAIN data.
STEPS TO IMPROVE ACCURACY

MODIFICATIONS TO THE DATASET:


• Use individual components to train the model and detect the
components on the Region of Interest ( ROI) and try to
classify the components.
• Generate individual components by cropping from the PCB
images.

You might also like