Cloud-Edge Collaborative Defect Detection Based on Efficient Yolo Networks and Incremental Learning.

Area of Science:

• Industrial Engineering
• Computer Science
• Artificial Intelligence

Background:

• Deep learning defect detection models struggle with expanding to new categories and achieving real-time performance on resource-constrained edge devices.
• Existing lightweight models often have insufficient detection accuracy for industrial applications.

Purpose of the Study:

• To present a novel lightweight deep learning model, SGRS-YoloV5n, for enhanced defect detection on edge devices.
• To develop a cloud-edge collaborative system with incremental learning for improved accuracy and adaptability.

Main Methods:

• Integration of four modules (SCDown, GhostConv, RepNCSPELAN4, ScalSeq) into the YoloV5 architecture to create SGRS-YoloV5n.
• Construction of a cloud-edge collaborative system for tiered defect inspection.
• Implementation of an incremental learning mechanism for adaptive learning of new defect categories.

Main Results:

• SGRS-YoloV5n demonstrates superior detection accuracy and real-time performance compared to existing lightweight models.
• The model significantly enhances feature extraction and computational efficiency while reducing model size and load.
• The cloud-edge system effectively improves overall detection accuracy and efficiency.

Conclusions:

• SGRS-YoloV5n is a valuable and stable solution for real-time defect detection in resource-constrained industrial environments.
• The proposed cloud-edge collaborative system with incremental learning offers a novel approach to efficient and accurate defect detection.