Research And Application Of Garment Label Recognition Based On Deep Learning

The label information such as garment length and collar type plays an important role in the sales process of e-commerce platform,but the traditional way of labeling by hand is time-consuming and laborious,and easy to produce errors.To solve this problem,this paper applies computer vision and deep learning technology to garment label recognition,and proposes an improved Inception-v4 clothing label recognition model based on the group's clothing cloud trading platform,which is applied to the process of garment shelving.The garment merchant only needs to upload garment images,and the system can automatically identify garment labels including garment length and collar type,and then store them in the database after the merchant proofreading,which improves the accuracy of garment labels and saves manpower at the same time.Based on the in-depth study of deep learning related techniques and image processing related algorithms,an improved Inception-v4 clothing label recognition model is proposed.Firstly,in the recognition model,the convergence speed is accelerated by introducing the residual connection to the Inception module in the case of deepening the network.Based on this,the apparel labels are divided into length class and design class,and a length class label recognition model and a design class recognition model are designed respectively.By sharing hard parameters and reusing convolution and pooling operations,the computational effort is greatly reduced while reducing the risk of overfitting and ensuring accuracy.Then by using Soft Label and loss function based on inter-class similarity,the label association information in the dataset is fully explored and the performance of the model is improved.It was also evaluated on the Fashion AI dataset,and the experimental results showed that the method described in this paper has a faster convergence speed while achieving good performance,and the accuracy rate was improved by 4.1% compared with the basic algorithm.Finally,an automatic garment label labeling system is developed,which is an application of the garment label recognition model to simplify the process of garment shelving.The automatic labeling system implemented in this paper provides accurate label information for garments in the sales platform and solves the problem of tedious and subjective manual labeling in the shelf process.In addition,the Soft Label proposed in this paper with a loss function based on inter-class similarity can also be used for the task of identifying age,height,and other tasks where incremental relationships exist.