This book proposes various deep learning models featuring how deep learning algorithms have been applied and used in real-life settings. The complexity of real-world scenarios and constraints imposed by the environment, together with budgetary and resource limitations, have posed great challenges to engineers and developers alike, to come up with solutions to meet these demands. This book presents case studies undertaken by its contributors to overcome these problems. These studies can be used as references for designers when applying deep learning in solving real-world problems in the areas of vision, signals, and networks. The contents of this book are divided into three parts. In the first part, AI vision applications in plant disease diagnostics, PM2.5 concentration estimation, surface defect detection, and ship plate identification, are featured. The second part introduces deep learning applications in signal processing; such as time series classification, broad-learning based signal modulation recognition, and graph neural network (GNN) based modulation recognition. Finally, the last section of the book reports on graph embedding applications and GNN in AI for networks; such as an end-to-end graph embedding method for dispute detection, an autonomous System-GNN architecture to infer the relationship between Apache software, a Ponzi scheme detection framework to identify and detect Ponzi schemes, and a GNN application to predict molecular biological activities.
Author(s): Qi Xuan, Yun Xiang, Dongwei Xu
Year: 2023
Language: English
Pages: 309
Contents
Preface
About the Editors
Introduction
Part I: Vision Applications
Part II: Signal Applications
Part III: Network Applications
Part I: Vision Applications
1. Vision-Based Particulate Matter Estimation
1. Introduction
2. Related Work
2.1. Vision-based air quality estimation
2.2. Air quality dataset
3. Methodology
3.1. Haze image processing
3.2. Haze-relevant features extraction
3.3. Data processing
3.4. Model architecture
4. Experiment
4.1. Devices
4.2. Experiment setup and data collection
4.3. Evaluation metrics and inference performance
5. Conclusion
References
2. Automatic Ship Plate Recognition Using Deep Learning Techniques
1. Introduction
2. Related Work
2.1. General object detection
2.2. Text detection
3. Dataset
3.1. Data collecting
3.2. Data labeling
4. Method
4.1. Data preprocessing
4.2. Ship license plate detection
4.3. Text recognition
5. Experiment
5.1. System overview
5.2. Ship detection server
5.3. Ship recognition server
5.4. Evaluation metrics
5.5. Evaluation and performance
6. Conclusion
References
3. Generative Adversial Network Enhanced Bearing Roller Defect Detection and Segmentation
1. Introduction
2. Related Work
2.1. Object detection network
2.2. Generative adversarial network
2.3. The evaluation indicators
3. Dataset
3.1. Analysis of bearing rollers defects
3.2. Dataset collection and analysis
4. Method
4.1. DCGAN network
4.2. Image classification network
4.3. Fine-grained defect segmentation network
5. Experimental Results
5.1. Experimental setup
5.2. Binary classification networks
5.3. Data augmentation
5.4. Fine-grained detection
6. Conclusion
References
4. Application of Deep Learning in Crop Stress
1. Introduction
2. A Mainstream Network of Deep Learning in the Field of Plants
2.1. Convolutional neural network
2.2. Autoencoder
2.3. Object detection network
2.3.1. Two-stage methods
2.3.2. One-stage Methods
3. Deep Learning-Based Crop Stress Researches
3.1. Existing literature
3.2. Biotic crop stress
3.3. Abiotic crop stress
4. Summary and trend
References
Part II: Signal Applications
5. A Mixed Pruning Method for Signal Modulation Recognition Based on Convolutional Neural Network
1. Introduction
2. Related Work
2.1. Signal modulation recognition
2.2. Convolutional neural network (ResNet56)
2.3. Network pruning
2.3.1. Low-rank decomposition
2.3.2. Weight pruning
2.3.3. Filter and channel pruning
2.3.4. Layer pruning
3. Methods
3.1. Basic mixed pruning method framework
3.2. Notations
3.3. Filter pruning part
3.4. Layer pruning part
4. Experiment
4.1. Datasets
4.2. Baselines
4.3. Evaluation metrics
4.4. Results and analysis
4.4.1. Results at the same pruning rate
4.4.2. Result of high pruning rate
4.4.3. Analysis
5. Conclusion
References
6. Broad Learning System Based on Gramian Angular Field for Time Series Classification
1. Introduction
2. Preliminary
2.1. Feedforward neural network
2.1.1. Single-layer feedforward neural networks
2.1.2. RVFL neural network
2.2. Sparse auto-encoder
2.3. Singular value decomposition
3. Method
3.1. Broad learning system
3.2. Gramian angular field
3.3. GAF–BLS
3.3.1. Feature matrix generation
3.3.2. Broad learning network training
4. Experiment and Discussion
5. Conclusion
References
7. Denoising of Radio Modulation Signal Based on Deep Learning
1. Introduction
2. Preliminary
2.1. GAN
2.2. CGAN
2.3. DCGAN
3. Methods
3.1. SDGAN
3.2. Generator and a discriminator design
3.3. Loss function design
4. Experiment
4.1. Dataset and setting
4.2. Evaluation metrics
4.3. Experimental results and discussion
4.3.1. Experiment on Gaussian noise
4.3.2. Experiment on superimposed signal
5. Conclusion
References
8. A Graph Neural Network Modulation Recognition Framework Based on Local Limited Penetrable Visibility Graph
1. Introduction
2. Related Work
2.1. Modulation recognition
2.2. Visibility graph
2.3. Graph neural network
3. Methods
3.1. Time-domain feature expansion
3.2. Local limited penetrable visibility graph
3.3. Graph neural networks
4. Experiment
4.1. Datasets
4.2. Baselines
4.3. The experimental settings
4.4. Results and discussion
5. Conclusion
References
Part III: Network Applications
9. Study of Autonomous System Business Types Based on Graph Neural Networks
1. Introduction
2. Related Work
2.1. AS classification
2.2. AS relationship
3. Datasets
3.1. BGP paths
3.2. S2S relationship and IXP list
3.3. AS classification ground-truth dataset
3.4. AS relationship experimental dataset
4. Features Description and Analysis
4.1. Basic feature
4.2. Features Analysis
4.2.1. Degree and transit degree
4.2.2. Distance to clique
4.2.3. Assign VP
4.2.4. Common neighbor ratio
4.2.5. Distance to VP
4.2.6. Node hierarchy
4.2.7. AS type
5. Methodology
5.1. Aggregate surrounding information
5.2. Problem description
5.3. Model framework
6. Evaluation
6.1. Baseline methods
6.2. Evaluation metrics
6.3. AS classification result
6.4. Feature importance analysis
7. Conclusion and Future Work
References
10. Social Media Opinions Analysis
1. Introduction
2. Related Work
2.1. Graph embedding
3. Toutiao Dataset
3.1. Data description
3.2. Label
3.3. Pre-processing
3.4. Dataset analysis
4. Method
4.1. Walking strategy
4.2. Feature calculation
5. Experiment
5.1. Baseline
5.2. Performance Comparison
6. Conclusion
References
11. Ethereum’s Ponzi Scheme Detection Work Based on Graph Ideas
1. Introduction
2. Ponzi Scheme Definition and Detection
2.1. Ponzi scheme in traditional finance
2.2. Ponzi schemes on Ethereum
2.3. Ponzi scheme detection on Ethereum
2.3.1. Detect in the initial state
2.3.2. Detect in the intermediate state
2.3.3. Detect in the final state
3. Feature of Ethereum Data
3.1. Contract code feature
3.2. Transaction feature
3.2.1. Transaction feature analysis
3.3. Network features
4. Method
4.1. Data collection and preprocessing
4.2. Model
5. Experiments
5.1. Data
5.2. Comparison methods
5.3. Experimental settings
5.4. Evaluation metrics
5.5. Results and discussions
5.5.1. Performance on manual features
5.5.2. Performance on network features
5.5.3. Performance on ALL features
6. Conclusion
References
12. Research on Prediction of Molecular Biological Activity Based on Graph Convolution
1. Introduction
2. Method
2.1. Establishment of graphs
2.2. Graph convolution based on edge attention
2.3. Attention graph convolution based on multi-feature fusion
2.3.1. Multi-Feature Fusion Scheme
2.4. Datasets
3. Experimental Results
3.1. Data processing
3.2. Experimental setup
3.3. Algorithm performance analysis
4. Discussions and Optimization
4.1. Problem analysis
4.2. Model optimization
4.2.1. Focal Loss
4.2.2. Gradient Harmonizing Mechanism (GHM)
4.3. Experimental setup
4.4. Algorithm performance analysis
5. Conclusion
References
Index
