TSDNet: A New Multiscale Texture Surface Defect Detection Model

Industrial defect detection methods based on deep learning can reduce the cost of traditional manual quality inspection, improve accuracy and efficiency detection, are widely used in industrial fields. Traditional computer focus features require a large amount data, which has some limitations. This paper proposes texture surface method convolutional neural network wavelet analysis: TSDNet. The approach combines analysis with patch extraction, detect locate many defects complex background; extraction random windows is proposed, quickly effectively extract defective patches; judgment strategy sliding window proposed to robustness CNN. Our achieve excellent DAGM 2007, micro-surface database KolektorSDD dataset, find location accurately. results show that background, obtain high only small training data accurately position.