MD-YOLO: Surface Defect Detector for Industrial Complex Environments

Abstract

Computer vision inspection techniques are currently being effectively used in industry with positive outcomes. Due to flaws in the manufacturing process and the interference of outside factors, surface defect detection still faces numerous challenges in realistic and complex industrial settings, including wide variation in defect scales, small differences between defect imaging and background, low contrast, etc. To address these issues, this paper introduces MD-YOLO, a surface defect detection model based on YOLOv5. First, we introduce a data enhancement algorithm designed to enhance image quality by processing the images. Secondly, within the network model, we have developed a Multi-Channel Fusion Module (MCF) to enhance the network's capability in extracting image features. Finally, we add the dynamic head block (Dyhead Block) to the detection head, resulting in a target detection head with attention to perform classification and regression tasks. The experimental results show that our method achieves 78.2% mAP on the NEU-DET dataset and 98.3% mAP on the BAT-DET dataset, with a parameter count of 9.0M.This demonstrates the robustness and versatility of the method proposed in this paper.