Multi-Deep Features Fusion Algorithm for Clothing Image Recognition

In order to improve recognition accuracy of clothing image and fully exploit the multi-category features extracted from global area, main part, to part areas. This paper utilizes the target detection technology and deep residual network (ResNet) to extract comprehensive clothing features, aims at focusing on clothing itself in the process of extraction procedure, and proposes multi-deep features fusion algorithm for clothing image recognition. First, the YOLOv3 model extracts the global area, main part of clothing and part areas of the image, which forms three category area images, so as to weaken the influence of background and other interference factors. After that, the three category images were sent respectively to improved ResNet for feature extraction, which has been trained beforehand. The ResNet model is improved through optimizing the convolution layer in the residual block, and adjusting the batch normalized layer and the order of the activation function layer, also adjusting the network convolution kernel structure. Finally, the multi-category fusion features were obtained by combining the overall features of the clothing image from the global area, the main part, to the part areas. The experimental results show that the proposed algorithm eliminates the influence of interference factors, makes the recognition process focus on clothing itself, greatly improves the accuracy of the clothing image recognition, and better than the traditional deep residual network based methods.