The Study Of High-Resolution Images Change Detection Method Based On Uncertainty Reasoning Model

Change detection method was generally used to find change areas in a region using a set of single-spectral or multi-spectral remotely sensed image data, by means of techniques such as image processing and pattern recognition. It was wildly used in many fields including environment, agriculture, water conservancy and military.Most of the present change detection methods aim at land cover applications using low-resolution images, and generally can not meet the needs of high-precision analysis. The most serious problems for change detection on high-resolution images are the sensitivity to registration and the complexity of terrain. Consequently, traditional change detection methods whether based on pixel level or based on classification can not produce enough accuracy. After investigations in high-resolution remotely sensed images, we produced a change detection method based on uncertainty reasoning model. The key techniques of this method were researched in this paper.By studying the theories of uncertainty reasoning and artificial intelligence, we produced a reasoning model - Constrained Bayesian Networks, with which we can express logical knowledge conveniently, and then gave a theoretic base of our change detection method. In addition, taking spatial information into account, two statistical features were presented: local matching degree and local max average difference. Determinations of change are made using spectral and spatial features together to reduce the false alarm rate brought by miss-registration.A change detection system based on uncertainty reasoning model was presented here. Working with this system, some features were calculated first, and judgments of change were made by uncertainty reasoning. The judging matching was designed using uncertainty reasoning model. Finally, the detection result was displayed after visualization processing. The experimental result shows that this method produces obviously higher accuracy than traditional change detection methods based on pixel level.