Parallel Algorithm Design For CVA Change Detection Of Remote Sensing Imagery Based-on DBN

In the rapidly changed and developed modern society, in order to use earth resource in a sustainable way, it should be obtained the land use and land change information in time. Remote sensing change detection technology has been concerned by people, in consider of remote sensing for earth observation technology with the advantage of wide coverage, strong macroscopic, fast, multi-temporal and abundant comprehensive information.The aim of remote sensing change detection is to obtain change region and identify change categories in study area. Either change region or change categories is detected in a lot of remote sensing change detection algorithms. However, both of the two aspects are consider in change detection algorithm based on change vector analysis. This method is mainly to obtain chang magnitude image and change direction image between bi-temporal images. Change magnitude image is for change area, and change direction image for change categories.In view of change detection accuracy and efficiency, the article mainly studies CVA change detection with the auxiliary of deep learning, and completes change detection in a parallel way with GPU/CUDA. The main content includes:(1) After obtaining change magnitude image, it needs binary processing to ensure change area. Generally speaking, it is difficult to get an optimal threshold to distinguish changed and unchanged area. What’s worse, it is not rigorous for just a threshold or an equation to ensure change area. So the article takes the advantage of deep belief network to analysis change magnitude image for change area. On the other hand, we take the index images(NDVI, NDBI and NDWI) for original images to execute change direction coding, then we get change categories.(2) In order to achieve the real-time remote sensing change detection, this article designs a parallel processing model in consider of Compute Unified Device Architecture(CUDA), in allusion to CVA-based change detection algorithm. Firstly, the model takes the advantage of Geospatial Data Abstraction Library(GDAL) to realize image block reading, block operation and block saving. Secondly, it divides the procedure of CVA change detection into three sub-process: Change Magnitude Detection, Change area definition based on DBN, The design of Index Table and Change Direction Detection. Then it embeds the three sub-process of CVA-based change detection algorithm in CPU and Graphic Process Unit(GPU) by the means of CUDA C. Lastly, this article studies multi-group images of different sizes with this model to execute CVA change detection procedure.