Texture Recognition and Classification Based on Deep Learning

Texture image classification has always been a very active research topic in computer vision and pattern recognition. In this paper, based on the deep learning advanced framework-Keras, we use Convolutional Neural Networks (CNN) to classify 12 kinds of texture images. Because there are few original datasets and the quantity is not balanced, We used such as reflection enhancement, elastic transformation, random lighting and other data augmentation techniques to enhance and expand some texture images. On the one hand, it balances the number of various types of texture images. On the other hand, it enhances the generalization ability of the datasets. It plays a key role in the training of the model and improves the accuracy of the model. The final test accuracy is close to 90%, which is more advanced and convenient than the traditional texture image classification method, and the accuracy rate is higher.