Abstract:

The traditional multi category classification method of clothing is mainly to extract the color, texture, edge and other features of the image manually. These manual feature selection methods are cumbersome and have low classification accuracy. The depth residual network can increase the depth of the neural network to obtain a higher recognition accuracy, which is widely used in various fields. To improve the accuracy of clothing image recognition, an improved depth residual network model is proposed in this paper. The model improves the convolution layer in the residual block, adjusts the arrangement order of the batch normalization layer and the activation function layer, introduces attention mechanism and adjusts the structure of convolution kernel. The network model is tested on the standard data set Fashion-MNIST and the multi category large clothing data set（DeepFashion） provided by the Multimedia Laboratory of the Chinese University of Hong Kong. Experimental results show that the proposed network model is superior to the traditional depth residual network in the accuracy of clothing image recognition and classification.

Key words: residual network, attention mechanism, clothing classification, deep learning

摘要：

传统的服装多类别分类方法主要是人工提取图像的颜色、纹理、边缘等特征，这些人工选取特征方法过程繁琐且分类精度较低。深度残差网络可通过增加神经网络的深度获得较高的识别精度被广泛地应用于各个领域。为提高服装图像识别精度问题，提出一种改进深度残差网络模型：改进残差块中卷积层、调整批量归一化层与激活函数层中的排列顺序；引入注意力机制；调整网络卷积核结构。该网络结构在标准数据集Fashion-MNIST和香港中文大学多媒体实验室提供的多类别大型服装数据集（DeepFashion）上进行测试，实验结果表明，所提出的网络模型在服装图像识别分类精度上优于传统的深度残差网络。

关键词: 残差网络, 注意力机制, 服装分类, 深度学习