This paper proposes a novel semisupervised support vector machine classifier (Formula presented.) based on active learning (AL) and context information to solve the problem where the number of labeled samples is insufficient. Firstly, a new semisupervised learning method is designed using AL to select unlabeled samples as the semilabled samples, then the context information is exploited to further expand the selected samples and relabel them, along with the labeled samples train (Formula presented.) classifier. Next, a new query function is designed to enhance the reliability of the classification results by using the Euclidean distance between the samples. Finally, in order to enhance the robustness of the proposed algorithm, a fusion method is designed. Several experiments on change detection are performed by considering some real remote sensing images. The results show that the proposed algorithm in comparison with other algorithms can significantly improve the detection accuracy and achieve a fast convergence in addition to verify the effectiveness of the fusion method developed in this paper.
- active learning
- context information
- remote sensing
- semisupervised support vector machine