Textile Quality Inspection System Based On Deep Learning

With the development of the "Made in China 2025" plan,China's industrial manufacturing industry is becoming increasingly intelligent.However,there are still some key technologies to be studied.Textile quality inspection has always been a weak link in the process of automatic production in the textile industry.A large number of quality inspection work still relies on manpower.But the labor-dependent quality inspection task has large human resource costs and unstable quality conditions.According to the problems above,this paper aims to design a textile quality inspection system based on deep learning methods,the main contents are listed below:1.The image collecting system is designed and a textile image data set is built.Using this data set,how the color space of the input image effect the performance of the deep learning model is discussed through experiments.According to the conclusion,a color weighting module is proposed for integrating multiple color space input images.2.According to the characteristics of textile surface and defects,a specialized deep learning model structure was designed for textile product quality evaluation.The attention mechanism module based on traditional visual features was used to weight the convolution feature graph.Combined with the traditional image features,the attention module emphasizes the important areas and effective features in the textile image,and readjust the response of the feature map.A multidirectional LSTM module was designed for processing feature maps.Aiming at the continuous occurrence of most textile defects,LSTM module was used to extract sequence features from images,and the model's ability to recognize defects was strengthened by using context information to reduce misjudgment.A deep learning model for textile quality inspection was designed based on the above methods,and a series of experiments proved the effectiveness of the above methods.3.With the deep learning model algorithm mentioned above as the core,a textile quality inspection system combined with field mechanical equipment was designed.The system includes a client and a server.The client is located at the production site and is responsible for image acquisition,transmission and giving of control signals,while the server is responsible for the calculation of large-scale deep learning model.According to the above design,a quality evaluation system was built and tested on the production site.