基于改进的卷积神经网络多类商品精细分类

A NOVEL CONVOLUTIONAL NEURAL NETWORK APPROACH TO IMPROVE THE MULTI-PRODUCT IMAGE BASED CLASSIFICATION IN E-COMMERCE

随着物联网的兴起、电子商务的蓬勃发展,依据图像特征对商品进行有效检索和分类具有重要应用价值.针对传统图像分类方法提取特征复杂,浅层卷积神经网络分类效果不佳的问题.本文对经典的AlexNet进行改进,优化了卷积核的尺寸,改变了各层连接,提出了一种分类效果更好的卷积神经网络结构.通过对8种商品进行测试训练,本文网络的分类准确率达到了91.2%,分类结果明显高于AlexNet的85.9%.

The e-commerce has become one of the important commercial activities, especially with the devel- opment of Internet of things. From the technical point of view, the effective retrieval and classification of product images based on the imaging features have been playing an extramely important role in support of e-commerce and related applications. However, there is a bottleneck due to the complicated feature extraction technology currently being used by the traditional image classification. Furthermore, the poor classification effect of shallow convolution neural networks is another weakness with the traditional image classification. This paper presents a novel convolu- tional neural network structure, which is based on AlexNet. The novelty is the efficiency because of the size of the convolution kernel is optimized and the connection of each layer is dynamic. An experiment based on the training and testing of 8 species of product images has been used to evaluate the performance for the proposed convolutional neural network structure. The obtained numerical results show that the classification accuracy for the novel eonvolutional neural network structure is able to achieve up to 91.2%, which is even higher than that of 85.9% by the AlexNet.