Description
Computer vision and image analysis are indispensable components of every automated environment. Modern machine vision and image analysis techniques play key roles in automation and quality assurance. Working environments can be improved significantly if we integrate computer vision and image analysis techniques. The more advancement in innovation and research in computer vision and image processing, the greater the efficiency of machines as well as humans. Computer Vision and Image Analysis for Industry 4.0 focuses on the roles of computer vision and image analysis for 4.0 IR-related technologies. The text proposes a variety of techniques for disease detection and prediction, text recognition and signature verification, image captioning, flood level assessment, crops classifications and fabrication of smart eye-controlled wheelchairs.
Table of Contents
1. BN-HTRD: A Benchmark Dataset for Document Level Offline Bangla Handwritten Text Recognition (HTR) and Line Seg-mentation
Md. Ataur Rahman, Nazifa Tabassum, Mitu Paul, Riya Pal and Mohammad Khairul Islam
INTRODUCTION
RELATED WORK
DATA ANNOTATION
Data Collection and the Source
Data Distribution
Annotation Guidelines
Annotation Scheme and Agreement
Data Correction
LINE SEGMENTATION: METHODOLOGY
Thresholding and Edge Detection
Morphological Operation and Noise Removal
Hough Line Detection 9 1.4.4 Hough Circle Removal
Bounding Box
OPTICS Clustering
Line Extraction and Cropping
RESULTS AND EVALUATION
Evaluation Metrics
Line Segmentation Results
CONCLUSION AND FUTURE WORK
2. A New Approach Using Convolutional Neural Network for Crops and Weeds Classification
Nawmee Razia Rahman and Md. Nazrul Islam Mondal
INTRODUCTION
CONVOLUTIONAL NEURAL NETWORK
THE PROPOSED MODEL
Data Source
Dataset Description
Work Procedure
Data Preprocessing
Experimental Setup and Evaluation Metrics
RESULT AND DISCUSSION
CONCLUSION
3. Lemon Fruits Detection and Instance Segmentation Under Orchard Environment Using Mask R-CNN and YOLOv5
S M Shahriar Sharif Rahat, Manjara Hasin Al Pitom, Mridula Mahzabun, and Md. Shamsuzzaman INTRODUCTION
LITERATURE REVIEW
Texture, color and Shape based fruits detection
Machine learning based fruits detection
MATERIALS AND METHODS
Image data acquisition
Image pre-processing
Model architecture
Model training
RESULT ANALYSIS AND COMPARISON
Result analysis
Discussion
CONCLUSION
4. A Deep Learning Approach in Detailed Fingerprint Identifica-tion
Mohiuddin Ahmed, Abu Sayeed, Azmain Yakin Srizon, Md Rakibul Haque, and Md. Mehedi Hasan INTRODUCTION
RELATED WORKS
DATASET
METHODOLOGY
Convolutional Neural Network Model
EXPERIMENTAL SETUP AND IMPLEMENTATION
Hyperparameters Optimization
Evaluation Criteria
RESULTS AND DISCUSSION
Gender Classification
Hand Classification
Finger Classification
CONCLUSION
5. Probing Skin Lesions and Performing Classification of Skin Cancer Using EfficientNet while Resolving Class Imbalance Using SMOTE
Md Rakibul Haque, Azmain Yakin Srizon, and Mohiuddin Ahmed
INTRODUCTION
METHODOLOGY
Dataset Description
SMOTE
Efficient-Net
PROPOSED APPROACH
Resolving Class Imbalance Using SMOTE
Extracting Complex and Versatile Features Using Efficient-NetB0
EXPERIMENTAL ANALYSIS
Experimental Setup
Classification Result
Understanding the Misclassifications
CONCLUSION
6. Advanced GradCAM++: Improved Visual Explanations of CNN’s decision in Diabetic Retinopathy
Md. Shafayat Jamil, Sirdarta Prashad Banik, G. M. Atiqur Rahaman, and Sajib Saha
INTRODUCTION
BACKGROUND
Convolutional Neural Networks (CNNs)
Visualizing CNNs
PROPOSED VISUALIZATION TECHNIQUE
EXPERIMENTS AND RESULTS
Training CNN model for disease level grading of DR
Visualizing CNN through GradCAM++ and proposed method
CONCLUSION
7. Bangla Sign Language Recognition Using Concatenated BdSL Network
Thasin Abedin, Khondokar S. S. Prottoy, Ayana Moshruba, and Safayat Bin Hakim
INTRODUCTION
LITERATURE REVIEW
METHODOLOGY
Data Preprocessing
Proposed Architecture
Image Network
Pose Estimation Network
Concatenated BDSL Network
Training Method
RESULTS
Dataset And Experimental Setup
Performance of Concatenated BDSL Network
DISCUSSION AND FUTURE SCOPE
8. ChestXRNet: A Multi-class Deep Convolutional Neural Net-works for Detecting Abnormalities in Chest X-Ray Images
Ahmad Sabbir Chowdhury and Aseef Iqbal
INTRODUCTION
RELATED WORK
METHODOLOGY
Data Preprocessing
Data Augmentation
Proposed ChestXRNet Model
Proposed Transfer Learning Methods for Benchmarking
Callbacks in Keras
RESULT ANALYSIS AND DISCUSSION
Data Description and Datasets
Experimental Setup
ChestXRNet Model’s Training, Validation Accuracy and Loss
Result Comparison Between ChestXRNet and Other PreTrained Models
Model Evaluation and Prediction
CONCLUSION
9. Achieving Human Level Performance on the Original Om-niglot Challenge
Shamim Ibne Shahid
INTRODUCTION
RELATED WORK
METHODOLOGY
EVALUATION ON OMNIGLOT
EVALUATION ON MNIST
CONCLUSION
10. A Real-Time Classification Model for Bengali Character Recognition in Air-Writing
Mohammed Abdul Kader, Muhammad Ahsan Ullah, and Md Saiful Islam
INTRODUCTION
METHODOLOGY
Data Acquisition
Feature Extraction
Classification model
RESULT AND ANALYSIS
CONCLUSION AND FUTURE WORK
11. A Deep Learning Approach for Covid-19 Detection in Chest X-Rays
SK. Shalauddin Kabir, Mohammad Farhad Bulbul, Fee Faysal Ahmed, Syed Galib, and Hazrat Ali INTRODUCTION
LITERATURE REVIEW
DATASET DESCRIPTION
Data collection
Dataset creation
PROPOSED METHODOLOGY
Proposed Algorithm
Preprocessing: Image resize and normalization
Augmentation of Images
Deep Neural Networks and Transfer-learning
Fine-tuning
Experimental Setup
Model Evaluation
RESULTS AND DISCUSSION
Evaluation
Results on first setting
Results on second setting
Result on third setting
CONCLUSION
12. Automatic Image Captioning Using Deep Learning
Toshiba Kamruzzaman, Abdul Matin, Tasfia Seuti, and Md. Rakibul Islam
INTRODUCTION
LITERATURE REVIEW
MODEL ARCHITECTURE
Encoder
Decoder
Model-1: Base Model (LSTM: Long-Short Term Memory)
Model-2: Transformer Model (BERT Integration)
Model-3: Our Model (BERT with LSTM and dense layer)
EXPERIMENTAL SETUP
Dataset
Hyperparameters
RESULT ANALYSIS
Qualitative Analysis
Model-1: Base Model (LSTM: Long-Short Term Memory)
Model-2: Transformer Model (BERT Integration)
Model-3: Our Model (BERT with LSTM and dense layer)
Quantitative Analysis
CONCLUSION
13. A Convolutional Neural Network Based Approach to Recog-nize Bangla Handwritten Characters
Mohammad Golam Mortuza, Saiful Islam, Md. Humayun Kabir, and Uipil Chong
INTRODUCTION
RELATED WORK
METHODOLOGY AND SYSTEM ARCHITECTURE
DATASET
RESULT ANALYSIS
CONCLUSION AND FUTURE WORK
14. Flood Region Detection Based on K-Means Algorithm and Color Probability
Promiti Chakraborty, Sabiha Anan, and Kaushik Deb
INTRODUCTION
LITERATURE REVIEW
OUTLINE OF METHODOLOGY
Background Subtraction
Dynamic K-Means Clustering Algorithm
Connected Component Labelling
Morphological Closing
Color Probability
Edge Density
EXPERIMENTAL RESULT ANALYSIS
CONCLUSION AND FUTURE WORK
15. Fabrication of Smart Eye Controlled Wheelchair for Disabled Person
Md. Anisur Rahman, Md. Abdur Rahman, Md. Imteaz Ahmed, and Md. Iftekher Hossain
INTRODUCTION
RELATED WORK
NOVELTY AND CONTRIBUTION
ORGANIZATION OF THE PAPER
SYSTEM DESIGN
Hardware configuration
Software configuration
METHODOLOGY
RESULT
CONCLUSION AND FUTURE WORK
