Research On Clothing Style Classification Algorithm Based On Capsule Network

Along with the popularity of the Internet and the rapid development of e-commerce,the sales mode of the apparel industry,which is one of the traditional consumer industries in China,has gradually shifted from offline to online.The development of clothing style classification technology as a basic technology in clothing retrieval and the application of deep learning in the clothing field is crucial.The existing clothing style retrieval methods with graph retrieval are mostly similarity retrieval,which cannot achieve accurate style retrieval.The traditional clothing image style classification has the disadvantages of high cost,low efficiency and low accuracy.The method based on convolutional neural network has insufficient performance in clothing style classification because it cannot utilize the spatial location relationship between clothing image features.For this reason,this paper is designed to implement the following work:1.The capsule network is improved to optimize the model computation and the number of parameters of the clothing style classification model.The number of layers in the traditional capsule network is small and the feature extraction ability is limited.In the process of feature extraction in the convolutional neural network,the number of layers is deepened and asymmetric convolution is introduced to optimize the model computation and the number of parameters while the network is deepened.The number of layers of asymmetric convolution deepened network can effectively reduce the model computation and number of parameters to10.02 MFLOPs and 19.13 M.2.An attention mechanism is added to the improved capsule network to design and implement a clothing style classification algorithm based on the improved capsule network.The attention mechanism is added to the convolutional network layer and the capsule network layer to pay attention to the channel information of clothing images.The experimental results show that the accuracy of the capsule network model with the attention mechanism is improved to 71.19% for the classification of clothing style.3.In order to further improve the accuracy of the network model for clothing style classification,this paper optimizes the clothing style classification algorithm based on an improved capsule network.The residual blocks in Res Net are introduced in the network model,jump-connected downsampling is designed to reduce the loss of feature information during convolution,and dual attention network for scene segmentation(DANet)is introduced in order to adaptively aggregate the correlation between global and local features,capture remote dependencies,and improve feature representation.In the experimental section,the accuracy of the capsule network-based clothing style classification algorithm before and after optimization and the accuracy of the algorithm for clothing image style classification in a single clothing category are compared,and the experimental results show the effectiveness of the optimized method with an accuracy increase from 71.19% to 72.72%.