E-commerce Item Identification Based on Improved SqueezeNet

Abstract In order to improve the recognition rate of product images in e-commerce recommendation scenes, we proposed a high-performance improved SqueezeNet convolutional neural network, which uses a Fire Module structure containing two large convolution kernels to reduce the complexity of the model while fully extracting features, and adding a pooling layer behind each Fire Module to effectively filter key features of the classification. Experiments show that the algorithm in this paper has a better application in image recognition in e-commerce scenes, and the recognition accuracy is improved by 1%.