Research On Classification Of Foreign Fibers In Cotton Based On Deep Learning

During production,raw cotton will be mixed with foreign fibers,which are different from cotton fibers,including polypropylene,hemp rope,plastic film and animal feather.Such foreign fibers will be mixed into the finished yarn,resulting in uneven dyeing and defects on the fabric surface,which will reduce the product grade and one of the main factors affecting the quality of production in the textile industry.Foreign fibers can be picked out on line by foreign fiber removal machine with machine learning algorithm,but the traditional machine does not realize the automatic sorting of foreign fibers,and cannot accurately know the class information of detected objects.The traditional machine learning algorithm is hard to solve the problem of classification of true and false foreign fibers in artificial feature extraction,but deep neural network does not need to extract features artificially and learn the features of images automatically,which can avoid the problem of unable to extract features.Moreover,deep neural network can be more widely adapted to the actual working state of the equipment,it is of great significance to research the new algorithm to classify foreign fibers,and then to evaluate the quality of cotton.Firstly,the research background and development status of foreign fiber in cotton about image classification are introduced.Moreover,the theoretical basis of deep learning and the composition of deep neural network are expounded,and convolutional neural network is also introduced.Secondly,a image classification method of cotton foreign fibers based on deep neural network is used.In this method,traditional deep neural networks including AlexNet,Vgg16 and GoogleNet are used to classify foreign fibers in cotton.The experimental results show that deep neural network can effectively learn the characteristics of cotton foreign fibers and achieve better classification accuracy.Then,a image classification method based on rich features of deep neural network is proposed.This method makes improvements on the basis of traditional neural network.It mainly uses convolution kernels to increase the features information from shallow layer,middle layer to deep layer,and uses point convolution to combine features.Finally,the all feathers are connected to the full connection layer to classify foreign fibers.Compared with the traditional deep neural network,the experimental results show that this method can further improve the classification accuracy of cotton foreign fiber image by deep neural network.Finally,the whole researches are summarized and the further research directions are prospected.