Research On Fabric Classification Algorithm Based On Deep Learning

With the development of the textile industry,the variety of textile fabrics is increasing.Thanks to the research and development of new textile fibers and new blended fabrics,which provide us with more functional and more textured fabrics.Nevertheless,with the extensive use of blended fabrics,the difficulty and challenge of fabric component testing has increased.Some unscrupulous merchants falsify in labeling fabric components,which lead to the situation where consumers do not have the corresponding identification ability,thus resulting in damage to their legitimate rights and interests.Therefore,there is an urgent need to find an efficient and convenient method to classify fabric components.With the increasing application of convolutional neural networks in various subject areas,its good image feature extraction and classification capabilities have been widely recognized by the academic community.In order to solve the problems existing in the traditional methods of classifying textile fabric components,we propose a main component classification method for textile fabrics based on deep convolutional neural network,which uses the pictures of pure textile fabric and blended textile fabric with main component contents more than 50% shot after amplifying 100-200 times as study objects.The main work of this paper can be divided into the following three parts:(1)The main component classification algorithm of fabric based on convolutional neural network is proposed.Firstly,collect the images of the textile fabric,create the image labels,and establish the database of fabric image.Secondly,set up the network model by using the dilated convolution and the depthwise separable convolution,then put the preprocessed images into the network for training to obtain the well trained network model.Finally,unclassified textile fabric images are taken into the well trained convolutional neural network model to obtain the main component classification result of the textile fabric.Experiments on 22485 images of cotton,polyester,acrylic,wool and tencel illustrate the accuracy of classification of the five fabrics is 96.53%.(2)Verifying the algorithm proposed in this paper.Firstly,experiments are conducted to prove the effectiveness of image preprocessing operation and convolutional neural network structure in the proposed algorithm.Compared with other convolutional neural network models,the time spent for training is significantly shortened,and the size of the network and the amount of calculation are reduced under the premise of ensuring classification accuracy.Finally,the Grad-CAM algorithm is used to visually analyze the image feature areas of the network models.The analysis results demonstrate the network classification is mainly based on the fabric fiber characteristics of the image,rather than the fabric texture information.(3)Using the fabric classification algorithm to construct the fabric main component detection system.The system adopts the B/S structure,in which the client uploads the fabric image to the server,and the server uses the algorithm to return detection result to the client in real time.Compared with the traditional textile fabric component detection,this system has the superiority in regard of environmental protection,simple and quick,no damage to the fabric and less requirements for testing equipment and testing personnel.