Research On Clothing Key Points Location And Attribute Prediction Method Based On Convolutional Neural Network

At present,due to the booming development of e-commerce,clothing visual analysis technology which uses deep learning method to process massive clothing data has been widely concerned.Among them,clothing key point positioning and attribute prediction have a wide range of practical value in the field of clothing retrieval,personalized recommendation,clothing posture estimation and virtual fitting,etc.,and become a research hotspot in the field of clothing visual analysis.The positioning accuracy of key points of clothing is easily affected by factors such as shooting angle,model pose and clothing deformation,etc.However,there are many kinds of clothing attribute labels and the difference between different attributes is small,which makes the accuracy of clothing classification prediction need to be improved.Based on the existing research,the following work has been done in this paper:In order to solve the key point positioning problem of clothing,a convolutional neural network model based on Recurrent Criss-Cross Attention(RCCA)is proposed.In view of the limitation that the convolution operation can only extract local features,the model takes full account of the position relationship between the key points of clothing,and adds the non-local information statistics part after the feature extraction operation,so as to obtain the global relationship of clothing features.In this paper,the RCCA network is used to construct the non local structure of clothing features.Based on the calculation of the correlation between the current features and other features,a non local response is obtained by weighting,and the residual operation is performed with the original input features,so as to capture the spatial connection between the key points of clothing.Experiments show that the algorithm reduces the normalized error of key point location.Based on the above research,a network model of clothing classification and attribute prediction based on mixed attention mechanism is proposed to achieve better effect of clothing classification and attribute prediction.The model uses the location information of key points to generate heat maps to help the branch network of spatial attention learn the spatial characteristics of clothing,and uses the local cross-channel interaction strategy to generate channel attention to capture the interaction information between convolutional channels.The features obtained from the two kinds of attention branching networks are fused before classification and attribute prediction.Experiments show that the algorithm has a certain degree of improvement on clothing classification and attribute prediction.In this paper,the key points of clothing positioning and clothing classification and attribute prediction are studied.The improved convolutional neural network-based clothing key point location and attribute prediction method improves the performance of the algorithm model,can promote the research work of clothing retrieval,virtual fitting,fashion matching and other applications,and has a certain promotion effect on the development of clothing visual analysis field.