关键词: 卷积神经网络, 深度学习, 领带花型, 织物情感, 情感分类

Abstract:

In order to avoid disadvantages of traditional manual features and local features that cannot characterize or quantify emotional features of images comprehensively and accurately, an method of fabric images emotion classification based on convolution neural network was proposed for evaluating the necktie pattern, and this method combined manual aesthetic features. This method is helpful for clothing design, clothing selection, etc. First, emotional evaluation of necktie pattern was processed, then an emotional sample library of necktie pattern images was established. We extracted manual emotional features such as image saturation and texture combined with the pixel values of sample images as the input data of the convolution neural network. An emotion classification model of convolution neural network was built, and the network structure and parameters were determined by experiments. We used 2 000 images as training samples to train the neural network. 1 000 test samples were input to the convolution neural network, and emotional classification of the images was evaluated automatically through our models. Experiment results showed that our method achieves classification accuracy of 89.7%, which is a greater improvement than the traditional methods based on manual features, and better than other convolution neural networks models with popular network structures.

Key words: convolution neural network, keep learning, necktie pattern, fabric imate emotion, emotion classification