Abstract: The change detection of multi-temporal remote sensing image is widely used in natural resource management such as survey and monitoring. According to the high construction cost and deep learning algorithms difficult of the sample library,this paper proposes multi-temporal change detection method to improve image change deep learning detection. This method take multi-temporal images as different band for information fusion,and transform the change detection task into image segmentation task, use land use vector data as label for model training and build deep learning sample library. Improve the structure of the original U-type deep learning neural network, and accelerated model training. Experimental results show that:①Multi-temporal change detection method is conducive to learn more features during model training and improving the feature extraction capability of the model,and finally getting the better prediction effect;②The recall rate and precision rate of the model is improved in a certain degree,and the whole prediction effect is obviously improved.

Key words: multi-temporal change detection, change detection of remote sensing image, U-type neural network, deep learning