A Transfer Learning Method for Ship Target Recognition in Remote Sensing Image

In this paper, an effective approach of ship target recognition is proposed. This method based on the theory of transfer learning aims at using labeled ships with different imaging angles and different resolutions to help identifying unlabeled ships in a fixed angle. Since training ship samples and test ship samples are imaging in different angles, they obey different distributions. However, in traditional machine learning method, training data and test data obey the same distribution. In order to solve this problem, we proposed a method called mapped subspace alignment (MSA) which is different from other domain adaptation methods. While maximizing the difference between different categories, it first uses Isometric Feature Mapping (Isomap) to generate subspace and uses objective functions to spatial alignment and probabilistic adaptation. This paper focuses on the identification of three types of ships which are destroyers, cruisers, and aircraft carriers basing on MSA. The experimental results show that this method is better than several state-of-the-art methods.