Defect Detection Method of Trash Trap Based on Deep Learning Semantic Segmentation

Due to the fact that the image information of trash trap is usually collected underwater, the quality of the image is low, which affects the reliability of the results of trash trap defect detection. Therefore, a trash trap defect detection method based on deep learning semantic segmentation is proposed. The threshold segmentation in semantic segmentation is used to realize the division of the target part and the background part, and the valley gray value between the corresponding wave peak in the mesh rope area and the corresponding wave peak in the mesh area is taken as the segmentation threshold value. When the gray value of the pixel in the trash net bag image is less than the threshold value, its gray value is set to 0, otherwise it is set to 255, so as to realize the segmentation of the target image. The convolution neural network in deep learning is used to recognize and detect the segmented target image. After the convolution neural network is preprocessed by Xavier initialization, the segmented target image of the trash trap is propagated forward in the convolution neural network to extract the characteristics of the image, which is normalized by softmax, and then the corresponding defects are identified by back propagation. In the comparison test results, regarding the detection of the defects of the trash trap, the design method witnesses the error rate of lower than 15.00%, the accuracy rate of more than 85.00%, the effective detection rate of more than 85.00%, and the maximum value of 95.00%, which is significantly better than the comparison method.