Fine Classification Method of Product Image Based on Multi-Level Convolutional Neural Networks

To improve the classification accuracy of e-commerce product images, a classification method for multi-category product images is proposed. The classification method is based on Convolutional Neural Networks, imitating human shopping habits and combining the features of product images, summarizing the images into multiple levels of parent and child categories, and adopting multiple classifiers to convolve the neural network from bottom to top The attention of the network is put on the global features of the parent categories and the local features of the child categories. The classification is trained layer by layer, and the weighted calculation is performed to obtain the final classification result.