Research On Clothing Classification Algorithm Based On Convolutional Neural Network

Image classification algorithm has high scientific research value,and it is the research focus of computer science in recent years.With the rapid development of deep learning,image classification technology based on convolutional neural networks is being widely used in daily life.In the task of clothing image classification,due to the variety of clothing,easy occlusion,easy wrinkle and easy deformation,it becomes a challenge to achieve accurate clothing image classification.Therefore,it is of great significance to combine algorithms based on convolutional neural networks with clothing image classification tasks.In recent years,due to domestic and foreign scholars' attention to the field of clothing imagery and continuous in-depth research on convolutional neural networks,the application of convolutional neural networks in clothing image tasks has become increasingly mature.By combining computer classification processing technology with clothing images,the accuracy of tasks such as attribute prediction and category classification of clothing images has been continuously improved.However,due to the unique characteristics of clothing,there is still a broad room for improvement in the algorithm research of convolutional neural networks in the field of clothing images.The main task of this paper is to apply the Deep Fashion data set proposed by He Yuming's team to propose an algorithm for clothing classification and feature prediction based on convolutional neural networks,and to optimize it on this basis.Aiming at the differences between clothing images and other types of images due to the unique attributes of clothing,this paper use clothing key tag prediction branch,and use the key location tag points in the clothing data set to apply network reinforcement training to clothing images.Then,the key position information extracted by the clothing key mark point prediction branch is fused with the global features generated by the basic convolutional neural network to build the vgg16-based clothing classification algorithm structure for subsequent clothing classification research.In the process of extracting image features with convolutional neural networks,shallow convolutional neural networks are easier to extract relevant information of clothing images,but their ability to extract semantic information is weak.As the number of convolutional layers continues to increase,the convolutional neural network structure extracts more semantic information from images,but some image information will be lost.Therefore,this paper proposes to optimize the vgg16 clothing classification algorithm by using the fusion residual feature method to fuse the low-level image information with the high-level semantic information to extract more global clothing image feature information.In addition,due to the differences in imagefeatures extracted by convolution kernels of different scales,multi-scale convolution kernels can extract image features more comprehensively.Therefore,this paper proposes to use multiconvolution kernels to extract features to optimize vgg16 clothing classification algorithm,The purpose is to improve the performance of clothing classification algorithms.In addition,focusing on the key areas in the image can improve the performance of the algorithm.Therefore,this article explains the role of the attention mechanism.And proposed to use the attention mechanism based on the space domain and the attention mechanism based on the channel domain to optimize the clothing classification algorithm.Aiming at the attention mechanism based on the channel domain,this paper proposes the use of multiple pooling fusion methods to optimize the channel attention model,the multi-scale kernel convolution method to optimize the channel attention model,and the adaptive scale convolution kernel method to optimize access Attention model,and use the optimized model to improve the vgg16-based clothing classification algorithm,and finally complete the goal of improving the performance of the algorithm and improving the accuracy of clothing classification.