A Domain Adaptation Deep Transfer Method for Image Classification

The deep learning models have recently shown outstanding performance in many computer vision applications. However, this superior performance requires a very large number of annotated image samples, pre-venting application to problems with limited training data. To overcome this limitation, we propose a Do-main Adaptation Deep Transfer Model (DADTM) in this paper. The DADTM improves the classical transfer models by the proposed domain invariance value metric and a domain invariance reconstruction, increasing the model transferability and enhancing the classification performance. The comparative experiments are performed to evaluate the DADTM-based classification algorithm. The results show that the proposed mod-el and algorithm outperform the traditional methods.