VISVESVARAYA TECHNOLOGICAL UNIVERSITY

Jnana Sangama, Belagavi- 590018

A
TECHNICAL SEMINAR REPORT
on

“CONVOLUTION NEURAL NETWORKS”

Submitted in partial fulfillment
Of
TECHNICAL
in
INFORMATION SCIENCE AND ENGINEERING
VIII SEMESTER TECHNICAL SEMINAR (18CSS84)

By
S VARSHITHA 1HK19IS082

Under the guidance of

Dr.V. Balaji Vijayan
Associate Professor

Department of Information Science and Engineering

2022 – 2023

HKBK COLLEGE OF ENGINEERING

22/1, Nagawara, Bengaluru – 560045
E-mail: [email protected], URL: www.hkbk.edu.in
 DEPARTMENT OF INFORMATION SCIENCE AND ENGINEERING

PROGRAMME EDUCATIONAL OBJECTIVES (PEOs)

PEO 1: To Empower Students through Wholesome Education to achieve academic excellent
education in the field of Information Science and Engineering.
PEO 2: To Provide Students with in-depth disciplinary knowledge in engineering fundamentals
that require to succeed in Information Science and Engineering.
PEO 3: To Create Highly Qualified Professionals in multi-disciplinary areas with the knowledge
of Information Technologies, Services Globally
PEO 4: To Inculcate in Students Professional and Ethical attitude with a strong character with
effective communication skills, teamwork skills, multidisciplinary approach, and an ability to
relate Engineering issues to broader social context.
PEO 5: To Provide Students with an academic environment aware of advanced technological
growth leading to life-long learning through innovation and research with professional ethics that
uplifts mankind

PROGRAM SPECIFIC OUTCOMES(PSOs)

Professional Skills:
An ability to identify and analyse requirements, and in designing and implementing well-
tested technology solutions for rapidly changing computing problems and information
system environments.

Problem-Solving Skills:
An ability to Design, develop and optimize solutions for information systems employing
fundamentals of system hardware & software, graph theory, finite automata, data storage
and communication networks.

Collaborative Skills:
An ability to communicate and develop leadership skills and work effectively in team
environments. They are capable of collaborating to design and implement well tested
solutions for rapidly changing computing problems and information system environments.

Successful Career and Entrepreneurship Skills:

An ability to adapt for innovation and changes and be successful in ethical professional
careers along with the impact of computing on society, and platforms in creating innovative
career paths to be an entrepreneur, and a zest for higher studies.
 Bengaluru – 560045

DEPARTMENT OF INFORMATION SCIENCE AND ENGINEERING

VISVESVARAYA TECHNOLOGICAL UNIVERSITY

TECHNICAL SEMINAR REPORT

ON
“CONVOLUTION NEURAL NETWORK”

Submitted by partial fulfillment of the

TECHNICAL SEMINAR (18CSS83)

VIII Semester, Department of Information Science and Engineering

2022 – 2023

SUBMITTED BY :

S VARSHITHA 1HK19IS082

ii
 ACKNOWLEDGEMENT
I would like to place my regards and acknowledgment to all who helped in making this project
possible. I thank all those whose guidance served as a beacon of light and crowned our efforts with
success.

First of all I would take this opportunity to express our heartfelt gratitude to the management
committee - Chairman Mr. C. M. Ibrahim, Director Mr. C.M. Faiz Mohammed and the in charge
Principal Dr. Tabassum Ara for all the infrastructures provided to complete the TECHNICAL
SEMINAR in time.

I deeply indebted to Dr. A Syed Mustafa, HOD, Information Science and Engineering for the
ineffable encouragement he provided for the successful completion of the project.

A special and an earnest word of thanks to the technical seminar guide, Dr.V.Balaji Vijayan and
our coordinator Dr. Komala C R for constant assistance, support, patience, endurance and
constructive suggestions for the betterment of the project.

I’m extremely thankful to the teaching and non-teaching staff of the Department of Information
Science and Engineering for their valuable guidance and cooperation throughout our dissertation.

I thank my parents for their support and guidance provided to us to finish the internship well ahead
of time. I thank my friends who lent their support in every way possible to make sure the mini
project has been completed. Last, but not least I would like to thank God for giving us this
opportunity to do everything in the appropriate time to finish this project.

S VARSHITHA (1HK19IS082)

iv
 ABSTRACT

Convolution Neural Network are a deep learning neural network that are designed for processing and
analysis of data with a grid-like structure, such as photos, videos, and audio signals, it is a subset of
deep neural networks. The design and operation of the brain's visual cortex, which processes visual
data, served as an inspiration for CNNs. A succession of convolutional layer of a CNN process the
input data by applying filters to it in order to extract local features. To create the final output, these
features are then merged and routed through one or more fully connected layers. Backpropagation is a
technique for training convolutional layers that modifies the settings of the filters to reduce the
discrepancy between the expected output and the actual output. Many applications, including image
classification, object identification, facial recognition, and natural language processing, have made
extensive use of CNNs. Due to their capacity to learn hierarchical representations of the input data,
which allows them to recognise complicated patterns and relationships, they have attained state-of-the-
art performance in numerous tasks. In general, CNNs have transformed computer vision and are now a
necessary tool for many machine learning applications.

v
 TABLE OF CONTENTS

Chapter No. Chapter Title Page No.

Acknowledgement iv

Abstract v
List of Figures viii

List of Abbreviations ix
1 Introduction 1
1.1 Introduction 1
1.2 What is CNN 3
1.3 Objectives 4
1.4 Principles of CNN 5

2 Technical Background 6
2.1 Evolution of CNN 6
2.2 Why CNN for Image Classification 7
2.3 Feature Extraction 8
2.4 Types of CNN Layers 9
2.5 How CNN Works 12
2.6 Characteristics of CNN 13
2.7 Types of CNN 14

3 CNN Architecture 15
3.1 Introduction to CNN Architecture 15
3.2 Types of CNN Architecture 19
3.3 Challenges in CNN Architecture 21
3.4 Benefits of CNN Architecture 22
3.5 Why is CNN Important 23

4 Training Process 24
4.1 Data Pre – Processing and Augmentation 24
4.2 Parameter Initialization 27
4.3 Regularization to CNN 28
4.4 Optimizer Selection 29

vi
5 Advancement in CNN 35
5.1 Advancement in CNN Architecture 35
5.2 Application Areas of CNN 39
5.3 Benefits of Convolution Neural Network 41
5.4 Future of Convolution Neural Network 41

6 Conclusion 43
Scope of Future work 43
References 44

vii
 LIST OF FIGURES
Fig No. Figure Name Page No.
1.1 ANN Layers 1
1.2 Example of Classification of CNN 11
1.3 Slide window for element 5
1.4 2D Convolution Principle 5
2.1 Evolution of CNN 6
2.2 CNN Image Classification 7
2.3 Example of Feature Extraction 8
2.4 Convolutional Layer 9
2.5 Max Pooling Layer 10
2.6 Fully Connected Layer 11
3.1 CNN Architecture 15
3.2 Down Sampling Technique 16
3.3 Convolution Layer 17
3.4 Pooling Layer 18
3.5 Connected Layer 18
3.6 LeNet Architecture 19
3.7 AlexNet Architecture 20
3.8 VGG Net Architecture 20
3.9 End to End Learning 22
3.10 Memory Sharing 23
4.1 Mean Subtraction 24
4.2 Normalization of Data 25
4.3 Example of Raw Training Data 26
4.4 Under – Fitting, Over – Fitting 28
4.5 Effects of LR Value on Process 29
4.6 Working Principle of Gradient Descent 30
4.7 Gradient Descent Graph 31
5.1 End to End model of FCN 36
5.2 End to End model of SegNet 37
5.3 End to End model of DeConvNet 37
5.4 End to End model of Deep Mask 38
viii
 LIST OF ABBREVIATIONS
ANN - Artificial Neural Network
FNN - Forward Neural Network
RBM - Restricted Boltzmann Machines
RNN - Recurrent Neural Network
CNN - Convolution Neural Network
CV - Computer Vision
RGB - Red Green Blue
FC - Fully Connected
NN - Neural Network
FCN - Fully Convolution Network
ResNet - Deep Residual Network
MNIST - Modified National Institute of Standards and Technology database
2D - 2 Dimensional
RCNN - Recurrent Convolution Neural Network
ReLu - Rectifier Linear Unit
POOL - Pooling Layer
GPU - Graphics Processing Unit
CPU - Control Processing Unit
SVM - Support Vector Machine
SDG - Stochastic Gradient Descent
IOT - Internet of Things

ix