Classification of Fabric Patterns Image Based on Improved Log-AlexNet

摘要

Fabrics are filled with aspects of our lives. Facing the massive fabric image data, how to help people quickly and efficiently classify these images becomes urgent. Aiming at the problem of low classification efficiency and accuracy, this paper proposes a binary hash classification framework based on AlexNet. By improving the activation function Log-ReLU and adding hidden layer to learn binary hash coding and other optimized network parameters, the proposed framework (Log-AlexNet) extracts rich abstract features and improves the classification efficiency and precision of fabric patterns. Compared with the traditional image feature extraction and classification methods, the experimental results show that the improved Log-AlexNet for classification of cloth fabric patterns is feasible and much better than the traditional state-of-the-art methods.