基于卷积神经网络的服装识别分类模型研究

Research on Model of Clothing Identification and Class ification Based on Conlolutional Neural Network

针对传统服装图像识别分类效果差、识别精度低的问题,在深度残差网络(ResNet)的基础上,提出基于改进深度残差块的卷积神经网络服装识别分类方法.通过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并引入注意力机制,调整网络卷积核结构,构建了基于改进ResNet的服装识别分类模型.同时将改进后的深度残差网络分别在Fashion-MNIST和DeepFashion两个数据集中进行测试,并将其与原始的ResNet与经典的深度学习网络测试结果进行对比,验证了该方法的有效性.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ResNet网络与经典的深度学习网络,识别准确率达到88%以上,分类识别效果更好,识别精度更高,可用于实际服装识别分类.

For the problem of poor classification effect and low identification accuracy,this paper presents a classification method based on improved deep residual network(ResNet).An improved ResNet-based classification model for clothing recognition is constructed by improving the convolutional layer in the residual block,adjusting the arrangement order in the batch normalized layer and the activation function layer,and introducing the attention mechanism and adjusting the network convolutional nuclear structure.Finally,the proposed method is validated by testing the improved deep residual network in both Fashion-MNIST and Deep-Fashion datasets respectively,and comparing the original ResNet with the classical deep learning network test results.The experiment shows that the proposed network model is better than ResNet network and classic deep learning network in clothing image recognition classification accuracy,with more than 88%,which has better classification recognition effect and higher identification accuracy.It can be used for actual clothing recognition classification.