Surface Defect Detection with Modified Real-Time Detector YOLOv3

In this paper, a modified YOLOv3 net has been proposed for surface defect detection. Different from other pixel-level segmenting methods, locates the regions of defects with bounding rectangles. Compared conventional detectors, operating efficiency is rather high without generating region proposals by sliding boxes. Although details are omitted in process, primary information location detects and class labels extracted accuracy. This sufficient inspection, computational improved, simultaneously. To further light structure YOLOv3, loss function optimization pruning strategy have adopted original YOLOv3. The ratio determined tradeoff between detecting accuracy efficiency. our experiments, we compared performance several state-of-the-art achieves best on six types DAGM 2007 dataset.