Small-Size Surface Defect Detection Algorithm for Quartz Wafers Based on YOLO-DDTBlock

The quartz wafer surface defect detection method uses object detection algorithm to classify and locate the defects on the wafer surface. Due to the small size and large number of surface defects of quartz wafer, the conventional target detection model is not so effective. This paper proposes a DDTBlock (Double Detection Block) module to optimize the detection. The DDTBlock module performs super resolution reconstruction of regions with low confidence, detects the reconstructed image and fuses the results. Since the DDTBlock module increases the amount of computation, this paper optimizes the backbone network to reduce the prediction time. Taking YOLOv4 as an example, the average detection time decreases from 31ms per frame to 22ms per frame, and the mAP increases from 10.1 to 21.8.