Target-oriented Image Feature Extraction And Change Detection Method

Remote sensing technology is a comprehensive subject. In recent decades, with the development of space remote sensing technology, the remote sensing image registration and recognition has been widely used. To improve the speed and precision of image registration is an important means to ensure the quality of image fusion and recognition. How to accurately identify the building in remote sensing images is an important topic in target recognition.Registration and recognition of SAR and panchromatic image mainly include the following two aspects:(1) SAR image carry speckle noise. Because of the difference of imaging mechanism, pixel manifestations difference between SAR image and panchromatic image is great. In order to improve the accuracy of registration, this paper firstly extract Ratio gradient feature and use the cross entropy as a measure of the cumulative distribution of the image matching, effectively improve the accuracy loss error caused by the noise and calculation error. Use the root mean square error and the correlation coefficient as the objective evaluation criteria. The experimental results show that, compared with the traditional registration algorithm based on mutual information, this method has higher registration accuracy. In view of the large amount of data calculation and the long time of registration in SAR and panchromatic image registration algorithm, on the basis of Powell algorithm, we put forward a kind of parallel optimization method, accelerating the speed of registration. Assign the large number of independent data calculation part in serial algorithm to multiple threads to execute, making big block data computing in parallel, thus speeding up the processing speed. The experimental results show that the parallel processing method make registration time decreased obviously with the increase of the number of threads.(2) In order to achieve high accuracy about building target identification, this paper design object-oriented building target recognition method. First of all, according to contour, texture and the relation between space positions of the building target, we can get building knowledge. According to the guidance of building knowledge, the structure of the underlying characteristics such as line, shape, texture could be extracted. Using the structures of recognition method, makes the realization of the underlying features combined with the knowledge. Finally, mark the buildings identified result. The experiment results shows that the building recognition method in this paper gains higher recognition rate in contrast with most of the developed methods..