Enhanced YOLOv5s-Based Algorithm for Industrial Part Detection.

Area of Science:

• Computer Vision
• Machine Learning
• Industrial Automation

Background:

• Accurate recognition and localization of industrial targets are critical in complex manufacturing settings.
• Existing object detection methods may struggle with multi-scale features, occlusions, and cluttered backgrounds.

Purpose of the Study:

• To improve the precision and accuracy of object detection for industrial targets.
• To enhance the robustness of industrial target recognition and localization algorithms.

Main Methods:

• An improved YOLOv5 algorithm incorporating Crossing Bidirectional Feature Pyramid (CBiFPN) for effective multi-scale feature fusion.
• Integration of an attention mechanism (C3_CA) to augment feature expression capabilities.
• Introduction of an Edge Detection Head (EDH) to handle occluded objects and cluttered scenes by leveraging edge information.

Main Results:

• The improved YOLOv5s algorithm demonstrated enhanced performance on the modified ITODD dataset.
• Key metrics such as mAP@0.5, mAP@0.5:0.95, precision, and recall showed significant improvements compared to the original YOLOv5s.

Conclusions:

• The proposed method effectively fuses multi-scale and multi-level features, improving detection for objects of various sizes.
• The integration of attention mechanisms and edge detection enhances feature representation and robustness in challenging industrial environments.