Remote Sensing object recognition based on transfer learning

The deviation of an object´s real data distribution from the known training data distribution would lead to low reliability of object recognition. To tackle this problem for Remote Sensing (RS) images, we propose a novel object recognition method based on transfer learning. The feature vectors of an object are first extracted by a joint Local Binary Pattern (LBP). The transfer learning is then employed to find the common parameter set among feature spaces of the object under different distributions. Through extensive experiments, it has been shown that a significant improvement on the accuracy is has been brought by the proposed novel method.