Remote Sensing Image Road Extraction Studies

According to the road information in remote sensing image,based on understanding and mastering the road extraction classical theory, to study how effective recognition and extraction these road information. In the research process, firstly based on the support vector machine for road extraction samples, get road outline by the training samples; secondly the road roughly outline for clustering analysis, thus separating the main road information; then using2D Maximum Entropy, specific design genetic algorithm to search the optimal solution and get binary image image; finally using pattern recognition knowledge and mathematical morphology to realize complete extraction of road target. Mainly focuses on the following research work:(1)Mathematical morphology combined with the generalized Hough transform come true the road extraction. Play advantage of using mathematical morphology, the relevant operator processing two value image, and making full use of road geometrical feature, obtain better result. In other studies, imaging follow-up processing part combined with the actual situation uses mathematical morphology to undertake processing operations.(2)The fuzzy C-means (FCM) algorithm to do the research, combined with the specific problems in the design of a reasonable classification scheme. The advantage of using machine learning, support vector machine (SVM) algorithm to do the research, in the FCM classification based on the selection of effective learning samples, for machine learning classification results obtained. The experimental results show that the classification of road extraction have better effect. From the final evaluation data, this method is effective.(3)Road feature extraction as the important step of road extraction, here using2D Maximum Entropy. Because the genetic algorithm to search the optimal solution advantages, combining with the feature extraction of specific objective function, in accordance with the standard genetic algorithm to undertake the design of the algorithm, the ideal road segmentation results.(4)To get better segmentation results from above, the road feature extraction based on morphology and comprehensive use of various identification and related knowledge on road target standardization, finally get a better road network information. Experimental method based on MATLAB simulation software, according to its powerful image processing tool, completed the road extraction algorithm. The remote sensing images from Google Earth, and ultimately to achieve the remote sensing image read and road extraction. Combined with specific effects and the use of the road extraction effect, this scheme is reasonable and effective.