Research On Surface Defect Detection Algorithm Based On Semantic Segmentation

Image semantic segmentation refers to the technique of classifying and labeling each pixel in an image.Surface defect detection is the key task of product quality assurance in the manufacturing industry.With the emergence of digital and intelligent networking in China’s manufacturing industry,the number and types of industrial products have increased day by day,and manufacturers and consumers have put forward higher requirements for product quality.In recent years,computer vision inspection methods based on deep learning have achieved good results in the industry.However,the automatic detection of surface defects is still a challenging task,and there are problems such as limited sample size,noise interference,low contrast,blurred boundaries,intra-class differences,and inter-class similarities.In this regard,the paper proposes a surface defect detection algorithm based on semantic segmentation,which can effectively solve the above problems.The main contributions are as follows:1.Aiming at the problems of noise interference,intra-class difference and interclass similarity,low contrast,and boundary blur,a SE-SDDNet is designed.The feature extraction and coding module of the model combine different levels of coarse-grained and fine-grained defect features to increase the robustness of the model to solve the problem of noise interference.After that,the feature aggregation mechanism maps the coarse-grained defect features to the hierarchical recursive aggregation module and makes full use of the contextual multi-scale information to highlight the boundary and location information of the defect to solve the problems of intra-class differences and inter-class similarity.In addition,the paper designs a global background suppression enhancement module to reduce background noise and strengthen the defect area.To make SE-SDDNet pay attention to the edge information,the boundary refinement module is designed to improve the accuracy of model segmentation.On the benchmark data set,experiments have verified its effectiveness and feasibility.2.Aiming at problems such as limited sample size,a TLMS-LDDNet is designed.Solve the problem of sample scarcity from the perspective of data domain,model domain,and algorithm domain.In the data domain,the quality and quantity of the sample data are improved from data preprocessing and data enhancement.The backbone feature extraction module is optimized in the model domain to extract more effective features while reducing the computational complexity of the model.In the algorithm domain,Transfer the prior knowledge learned in other domains to speed up the algorithm search.TLMS-LDDNet also involves a multi-head self-attention module,which is used to perceive the importance of features extracted from different layers and to model long-distance content relationships.A large number of experiments have proved that the effectiveness of the model and related strategies can improve the learning ability of the model.This paper explored and discussed the surface defect detection algorithm based on semantic segmentation.Extensive experiments have proved its effectiveness and superiority.