Related Experiment Video
Updated: Sep 3, 2025
06:37
Author Spotlight: Addressing Technical and Subjective Challenges in Measuring Classroom Attention
Published on: December 15, 2023
4.0K
An Efficient YOLO Algorithm with an Attention Mechanism for Vision-Based Defect Inspection Deployed on FPGA.
Longzhen Yu1, Jianhua Zhu1, Qian Zhao2
1College of Economics and Management, Qingdao University of Science and Technology, Qingdao 266000, China.
Micromachines
|July 27, 2022
Summary
This study introduces an efficient, FPGA-accelerated You Only Look Once (YOLO) v3 algorithm with an attention mechanism for smart defect inspection in manufacturing. The enhanced algorithm achieves 99.2% accuracy at 1.54 FPS with low power consumption.
Area of Science:
- Intelligent Manufacturing
- Computer Vision
- Artificial Intelligence
Background:
- Vision-based defect inspection is crucial for quality control in parts manufacturing within Industry 4.0.
- Current methods often rely on manual operations, highlighting the need for automated, AI-driven solutions.
- Advancements in AI and machine learning have enabled significant progress in auto-adaptive inspection systems.
Purpose of the Study:
- To propose an improved smart defect inspection algorithm for industrialization demands.
- To enhance the You Only Look Once (YOLO) v3 algorithm with an attention mechanism for focused defect detection.
- To deploy the algorithm on a Field Programmable Gate Array (FPGA) for efficient, real-time industrial application.
Main Methods:
- An attention mechanism was developed, including image preprocessing (cutting, zooming, splicing - CZS operations), YOLOv3 backbone network tailoring, and data augmentation.
- The improved YOLOv3 algorithm was converted for deployment on a PYNQ-Z2 FPGA board using the Xilinx Deep Neural Network Development Kit (DNNDK), involving pruning, quantization, and compilation.
- The system was optimized for accuracy, efficiency, and extensibility in industrial scenarios.
Main Results:
- The FPGA-accelerated YOLOv3 algorithm achieved a high inspection accuracy of 99.2%.
- The system demonstrated an efficient processing speed of 1.54 Frames per Second (FPS).
- The deployed solution exhibited low power consumption, utilizing only 10 W.
Conclusions:
- The proposed FPGA-accelerated attention-based YOLOv3 algorithm offers a highly efficient and accurate solution for smart defect inspection in industrial settings.
- The integration of FPGA provides a low-latency, cost-effective, and power-efficient platform for deep learning acceleration in manufacturing.
- The developed system meets industrial production requirements for accuracy, speed, and extensibility, paving the way for advanced quality control.
