Research On Change Detection Technology Of High Resolution Images Based On Visual Attention Mechanism

With the rapid development of space remote sensing technology, the spatial resolution, temporal resolution and spectral resolution of earth observation data are continuously improving, which leads to a wider application in both civil and military fields. For example, we can achieve the information of land cover change, extension of the urban, change of military deployment in a certain place by short or long time repeated observing. However, it is hard to completely and accurately analyze more abundant earth’s surface and space information by traditional image processing algorithms. The advantage of high resolution images is reflected under this situation. However, the interpretation problem brought by some of its characteristics such as large amount of data or complex details of texture needs to be deeply researched. This article mainly focuses on the technology of change detection for multi-temporal high spatial resolution remote sense images and achieving the high precision extraction of real change target by using the saliency computing model which is constituted by visual attention mechanism of human.The article illustrates the conception of change detection and the general procedures of the algorithm firstly, and then analyzes features of high spatial resolution remote sense images. As for the problems of pixel-level detection algorithm such as the mutual-independence of the computing methods for multiple difference images and the losing information of single method, applying multi-resolution decomposition technology to various difference images is able to generate a fused difference image which effectively solves the problem about comprehensive utilization of the cross information by different acquisitions of the multi-temporal difference images, which lay the foundation of the judgment of the subsequent change target.This article also analyzes the technical difficulties of change detection of high spatial resolution images, and raises an algorithm of determining the change region based on visual attention mechanism. Aiming at the inconformity of the spectral reflection of the same area caused by various shooting time, different solar zenith angles and illumination intensity of multi-temporal resolution remote sense images, the article attempts to imitate human to acquire the information during cha nge detection. Based on the research of attention mechanism of human, the technical advantage about the saliency computing model in judging real change regions is analyzed and applied to the new change detection algorithm.In the final part of this thesis, a high accuracy extraction of target in change region is fulfilled by using features of high resolution remote sense image along with pixel-saliency adjustment method. To solve the problem of the incomplete detection of a certain target caused by the improvement of spatial resolution which leads to a different degree of change of multi-temporal images in the internal of the same target, a redistribution of pixel color and the calculation of saliency values of pixels are carried out by combining visual attention mechanism and theory of image segmentation. Then the difference of saliency between the target and background is enlarged by the color smooth technique, and a more complete acquisition of change target in difference images is achieved. Finally the result of change detection is upgraded from the region level to target level.