Fabric Defect Detection And Recognition Based On Depth Neural Network

China is a major textile country,and the textile industry has increasingly higher requirements for the quality of fabrics.Finding and identifying fabric defects is an important factor to limit the efficiency and quality of fabric production.Traditional fabric defect detection and recognition based on artificial recognition,but this paper proposes to apply deep neural network algorithms to identify fabric defects.This method ranges from defect image processing to defect detection,recognition and classification,which can realize intelligent defect detection on the fabric production line.Informatization,automation and digitization have improved the detection efficiency of fabric defects.The actual measurement results show that the detection speed of the system is 2-3 times higher than the manual detection speed,and the detection error of the system is 4% lower than the manual error,and the detection accuracy rate is also higher than 92.5%.The main research work of this paper is as follows:1.The problem of difficult classification of fabric defects is solved.Training based on the deep learning model requires a large number of data sets,collecting fabric defect samples,and establishing a fabric defect image data set.A deep neural network classification algorithm based on a hybrid algorithm of simulated annealing algorithm and particle swarm is proposed.Make full use of the advantages of simulated annealing algorithm and particle swarm algorithm in finding the global optimal solution and improving the overall performance of the model.Combined with the deep neural network algorithm,the extracted fabric defect feature values are used as the input of the neural network to perform processing on various fabric defects.Accurate classification meets the target needs of the enterprise.2.The optimization of network model training methods and compression algorithms,combined with multi-threading,multi-task learning,and deep network feature maps have richer detailed information.By optimizing the method of model training,the convergence speed and effect of model training are improved.Using dual network training model can not only improve the performance of the existing model compression algorithm,but also reduce a lot of workload of layer-by-layer training.3.The fabric defect detection system is developed by using matlab algorithm library,python language and QT application development framework.By verifying the performance of the system,the company's testing requirements for the system are met.The actual production application shows that the detection accuracy of this system for fabrics of different colors is between 92.5% and 95%(greater than 90%),which has achieved the expected effect.