Research On Fabric Defect Detection Method Based On Parallel Attention Mechanis

In the assembly line of industrial products,to improve the quality and production efficiency of products and grasp the quality of products,it is an important link to inspect the surface defects of industrial products.The textile industry has become one of the important components of my country’s industrial industry.In the production process of the textile industry,machine failure or improper operation of workers may occur at any time,making the products produced inevitably have defects.Most of the traditional textile defect detection methods are carried out manually.This traditional method has problems such as low detection efficiency,missed detection,and high false detection rate to a certain extent.Therefore,to ensure the controllable quality of the production process,automatic fabric surface defect detection has become a research hotspot in the textile industry,information technology,and other industries.In recent years,deep learning technology has also been applied to this.Deep learning technology originated from the artificial neural network,which can extract the characteristics of data layer by layer.It has the ability of automatic learning and has broad prospects in the application of fabric defect detection.Therefore,based on deep learning theory,we designed two network structures according to different detection difficulties in exploring the task of fabric defect detection.The main work is as follows:(1)Aiming at the problem that the small relevant features of the cloth defect area are not easy to extract,a fabric defect detection model MPANet based on the parallel attention mechanism is proposed.In this model,a multi-branch parallel feature extraction network is composed of pre-and post-attention modules.Among them,the Transformer global attention mechanism and local space and channel attention mechanisms are included in the pre-and post-attention modules to improve the sensitivity to small target features;and the multi-branch parallel feature extraction network can make full use of the target Low-dimensional morphological features to improve the network’s ability to learn high-dimensional semantic features.According to the experimental test,the F1-Socre value of the network model reaches 0.905,and the m AP value reaches 0.916.The average increase of the two indicators is 0.086 compared with that before optimization.Experimental results show that the model has better feature extraction ability for small objects.(2)Aiming at the problem that the high similarity of small cloth defects leads to wrong detection types,a fabric defect detection model FF-MPANet based on multigranularity receptive field aggregation and decoupling double-head detectors is proposed based on MPANet network.Firstly,the multi-granularity feature map aggregation module is used to fully aggregate the feature information extracted by the backbone network to enhance the richness of feature map feature information;secondly,a decoupled dual-branch detector is used to predict the category and location of defects.After testing,the Precision index and Io U index of FF-MPANet have a certain improvement compared with the MPANet network,and the F1-Score index and m AP reach 0.913 and 0.922 respectively.The experimental results show that the model has better resolution for similar small objects.and positioning capabilities.Finally,we summarize the work done and give some outlook on the subsequent development work and research directions of the fabric defect detection model.