Incomplete target detection of glass production based on parallel residual classification network language segmentation algorithm

The accuracy requirements of optical element defect detection are gradually improved with the development of industry, which poses a new challenge to image processing. Traditional machine learning algorithms have some limitations in the scene of complex background defect extraction and weak defect feature extraction. As an important part of artificial intelligence technology, deep convolution neural network algorithm has made great breakthroughs in image classification, target detection, image segmentation and other fields in recent years. In this paper, symmetrical convolution neural network is used to extract the features of image defects in each dimension to realize weak defect image segmentation. Through the parallel residual classification network language segmentation algorithm, the defect image data obtained in the detection process are marked at the pixel level, and the data set containing defect and background image is obtained, which is used for model verification. In the experiment, symmetrical convolutional neural network is used to extract the characteristics of each dimension of image defects. The average recall rate is 95.3% and the average accuracy rate is 91.8%. It obtains a better weak defect segmentation effect than the traditional algorithm, which lays a foundation for the application of further improving the accuracy of defect segmentation.