Classification of Ethnic Dress Based on ResNet

Deep learning is widely used to process picture information. Chinese ethnic culture is colorful, and ethnic costume is one of the most symbolic representatives. In order to solve the problem of time-consuming and laborious manual classification, this paper uses transfer learning to put forward a method of ethnic costume classification based on ResNet. The convolution neural network is used to extract the features of different ethnic costumes to realize the classification. In the experiment, we first used ResNet, AlexNet, LeNet and GoogLeNet for training, and found that ResNet had the best effect, but the initial recognition accuracy was not high, the primary reason was insufficient data, so we first expanded the data set, and established a national costume data set with about 10,000 pictures. 70% of them are used as the training set and 30% as the test set. After image preprocessing, it is cut to the size of 300
^*
300
^*
3. The learning rate decay strategy is then introduced along with a deeper ResNet network—ResNext that is stacked from Inception modules. The final experimental results show that the recognition accuracy of ResNet18 is the highest, and its accuracy rate, recall rate and f1-score are about 0.6. This experiment lays a good foundation for the information processing of national costumes, and also provides support for the inheritance and protection of national culture.