This book proposes practical application paradigms for deep neural networks, aiming to establish best practices for their real-world implementation.
This book proposes practical application paradigms for deep neural networks, aiming to establish best practices for their real-world implementation.
Over the past decade, deep neural networks have made significant progress. However, effectively applying these networks to solve various practical problems remains challenging, which has limited the widespread application of artificial intelligence. This book is the first to comprehensively address implementation paradigms for deep neural networks in practice. The authors begin by reviewing the development of artificial neural networks and provide a systematic introduction to the tasks, principles, and architectures of deep neural networks. They identify the practical limitations of deep neural networks and propose guidelines and strategies for successful implementation. The book then examines 14 representative applications in urban planning, industrial production, and transportation. For each case, the authors present a landing paradigm that effectively addresses practical challenges, supported by illustrations, background information, related work, methods, experiments, and conclusions. The experimental results validate the effectiveness of the proposed implementation approaches.
The book will benefit researchers, engineers, undergraduate and graduate students interested in artificial intelligence, deep neural networks, large models, stable diffusion models, video surveillance, smart city, intelligent manufacturing, intelligent transportation, and other related areas.
1. Introduction to Artificial Intelligence for Practical Application
2.
Fine-grained Dataset Design Paradigm for Mask-Wearing Recognition
3.
Hand-Held Action Detection Paradigm for Smoking Detection
4. Human Action
Recognition Paradigm for Person Safety Supervision in Various Practical
Scenarios
5. Person Counting Paradigm for Intelligent Video Surveillance in
Various Practical Scenarios
6. Operation Procedure Detection Paradigm for
Noncompliant Operations Detection of Oil Unloading
7. Object Measurement
Paradigm for Length Measurement of Iron Chains
8. Quality Estimation Paradigm
for Copper Scrap Granules Recycling
9. Human-in-the-Loop Learning Paradigm
for Fabric Anomaly Detection
10. Supervised Learning Paradigm for Edible Oil
Anomaly Detection
11. Unsupervised Learning Paradigm for Industrial Visual
Anomaly Detection
12. Object Identity Recognition Paradigm for Fishing Boat
Recognition
13. Image Editing Paradigm for Clothing Fashion Customization
14.
Retrieval-Augmented Generation Paradigm for Professional Knowledge
Acquisition Applications
15. Multimodal Generation Paradigm for Visualizing
Historical Artifacts
Shiguo Lian received his Ph.D. from Nanjing University of Science and Technology, China. He currently serves as Chief Scientist at the Data Science & Artificial Intelligence Research Institute and Chief Engineer of AI Technology, China Unicom. He is a member of the IEEE Multimedia Communications and Computational Intelligence Technical Committees. His research focuses on visual recognition, multimodal large models, robotics, and multimodal interactions.
Zhaoxiang Liu received his Ph.D. from the College of Information and Electrical Engineering at China Agricultural University, China. He currently serves as Director of AI Research at the Data Science & Artificial Intelligence Research Institute, China Unicom. His research interests include artificial intelligence, large language models, multimodal large models, deep learning, computer vision, and embodied AI.
