Classification of Fabric Defects Based on Deep Adaptive Transfer Learning

This paper proposes a method based on deep adoptive transfer learning for cloth defects detection. In this paper, collected cloth images are cut, marked to create the datasets, and the image classification method is used to determine the defects of the cloth. The experimental results show that the proposed method has outstanding performance on classifying the cloth defects. The classification accuracy of 95.53% and 93.82% is achieved on the InceptionV3 and DenseNet121 transfer learning models respectively. By feature learning heat map of samples, it verifies that the models have strong learning ability and generalization ability for the characteristics of the cloth defects.