PO-YOLOv5: A defect detection model for solenoid connector based on YOLOv5

Area of Science:

• Manufacturing Engineering
• Computer Vision
• Automotive Safety Systems

Background:

• Solenoid connectors are vital components in electronic stability systems.
• Accurate defect detection in solenoid connectors is challenging due to similar visual characteristics between faulty and non-faulty parts.
• Defect detection is critical for ensuring safe driving performance.

Purpose of the Study:

• To develop an accurate defect detection model for solenoid connectors.
• To address the limitations of existing methods in identifying subtle defects.
• To enhance the reliability and safety of electronic stability systems.

Main Methods:

• Proposed a novel defect detection model named PO-YOLOv5.
• Incorporated an additional prediction head to capture larger-scale defect features.
• Introduced dynamic convolution with a multidimensional attention mechanism to improve detection accuracy and reduce inference time.

Main Results:

• The PO-YOLOv5 model achieved a mean Average Precision (mAP) of approximately 90.1%.
• Demonstrated a 3% improvement in precision compared to the original YOLOv5 model.
• Validated superior performance against state-of-the-art object detection methods on the same dataset.

Conclusions:

• PO-YOLOv5 offers enhanced accuracy and efficiency for solenoid connector defect detection.
• The model's improvements contribute to more reliable automotive safety systems.
• The proposed method effectively overcomes challenges in inspecting visually similar defects.