Optimize and deploy machine learning algorithms on embedded devices for manufacturing applications
This proposal discusses the techniques of optimizing and deploying machine learning algorithms on embedded devices for manufacturing applications; We investigate problems of printed circuit board (PCB) defects and artificial intelligence in embedded system. PCB defects detection had been an essentia...
Saved in:
| Main Author: | |
|---|---|
| Format: | Final Year Project / Dissertation / Thesis |
| Published: |
2025
|
| Subjects: | |
| Online Access: | http://eprints.utar.edu.my/6201/1/fyp_CS_2025_TMX.pdf http://eprints.utar.edu.my/6201/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | This proposal discusses the techniques of optimizing and deploying machine learning algorithms on embedded devices for manufacturing applications; We investigate problems of printed circuit board (PCB) defects and artificial intelligence in embedded system. PCB defects detection had been an essential problem to solve in manufacturing environments, whether it is quality assurance or the needs for PCB inspection had been ramping since decades ago. Fundamental limitation of human-based judgement of inspection engineers is the primary cause of faulty products including PCB defects exiting the manufacturing environment. On the other hand, artificial intelligence had been ways to enhances embedded system by enabling real-time, accurate detection and management of PCB defects through advanced pattern recognition and automated inspection methods. However, embedded system often been having limited computing power, small memory storage and relies on battery capacity. Not to say the difficulty in deploying either artificial intelligence or deep learning in embedded environments due to significant parameters size and computational complexity. In recent studies, we seen developers and researchers proposing solutions on deep learning algorithms like YOLO, EfficientNet, CNN, MobileNet etc. On the other hand, network compression and acceleration techniques such as pruning and quantization also been the focus of the studies for light-weight algorithms in embedded system. While in our studies, we primarily focusing on the deployment and fine-tuning of deep learning model which is YOLOv5 for PCB defects detection. We aim to levitate the baseline YOLOv5 into a state-of-the-art version that focus on lightweight performance, called the LW-YOLOv5, which can been deploy seamlessly into embedded systems for manufacturing applications. As we conduction evaluation experiment on our model using openly accessible datasets like PKU-Market-PCB and perform comparative studies with the latest proposed solutions. |
|---|