Research On Clothing Attribute Label Recognition Technology Based On Multi-task Deep Learning

With the rapid development of Internet technology,the scale of image data has increased exponentially.Image classification is an important means to obtain effective information in images.In the field of clothing,most of the existing methods are based on the whole clothing processing and recognition and ignore the details of the characteristics of parts.With the arrival of "AI+ fashion",artificial intelligence technology is closely combined with business scenes to quickly and accurately identify and classify clothing,which has become a hot topic in academia and industry.For clothing label attribute recognition,this paper designs and realizes an attribute tag identification method based on the in-depth study,using data from Ali electricity network clothing image database,based on the idea of multitasking joint learning recognition model is established,based on the deep learning of neural network prediction model is setup,the weight of the identification model of integrated training,complete clothing attribute prediction,thus accomplishes the accurate identification and classification of apparel attribute.This paper has done the following research on the existing basis:(1)Based on SE(Squeeze-and-Excitation)module for the Inception and Inception-Res Net network was improved,the optimized network,in the case of adding less amount of calculation,the performance of ascension,the late based on improved network modeling,obtained the better recognition effect.(2)To solve the problem of low recognition accuracy of single-channel network,a length recognition model based on two-channel convolutional neural network is proposed,which increases the generalization ability of the model and improves the recognition accuracy.(3)In the process of constructing the two-channel length recognition model,a classification strategy of clothing image tiled and non-tiled data was proposed,which realized the classification of clothing attribute length tiled and non-tiled data,and provided a strong support for the establishment of the two-channel recognition model.(4)Considering that different model structures have different extraction of network features,this paper integrates multiple models in a weighted form to obtain better prediction effect than a single model.The experimental results show that the method can effectively realize the recognition and classification of clothing attribute tags,improve the classification accuracy,effectively solve the problem of difficult recognition and classification of clothing attribute tags,and realize the efficient recognition and classification of local attributes of clothing images.