In the algorithms of change detection of remote sensing image based on semisupervised SVM, PTSVM algorithm uses paired labeling rule that once marks only one pair of unlabeled samples to label unlabeled samples, which achieves good results in the change detection of remote sensing image and to some extent, improves the detection accuracy. But PTSVM algorithm uses only spectral information of samples. Sample characteristics information includes not only the spectral information, but also spatial information. Spatial information as one of the basic characteristics of the sample is equally important, so this paper proposes the improved PTSVM algorithm combined spatial information of samples with spectral information of samples in change detection for remote sensing images. Through the experimental results analysis, the method has achieved good results in change detection of remote sensing image, and further improves the change detection accuracy.