A damage detection algorithm for aluminum workpiece based on improved segmentation and decision network

Aiming at the problems of low accuracy and detection efficiency of traditional artificial detection methods, this paper proposes a detection algorithm of slight apparent damage for aluminum workpiece based on improved SD (Segmentation and Decision) network. Firstly, we carry out dataset augmentation and make dataset by LabelMe. Secondly, we filter the noise by the image preprocessing and improve the image contrast by CLAHE (Contrast Limited Adaptive Histogram Equalization). Then, the channel attention mechanism is introduced into the decision network to improve the ability of feature extraction for small defects. Finally, we optimize the loss function of the segmentation network to improve the effect of defect segmentation. The experimental results show that this algorithm can effectively detect the defects such as pits and scratches, and it is practical for detecting the surface-defect of Stadge.