Remote Sensing Image Segmentation Based On Fuzzy Theory

Remote sensing images in civil, geological prospecting, farmland planning, military reconnaissance and precision strikes have important role. Remote sensing image segmentation is a key step of understanding remote sensing image; it is a leap image processing, computer vision, pattern recognition, neural network integrated multi-disciplinary research project, has broad application prospects, in recent years has been widely attention. Remote sensing image has the following main features: gray-scale , fuzzy boundaries and susceptible to noise, etc. It is precisely because of the complexity of remote sensing image, so far no unified segmentation algorithm. The current international has made a variety of image segmentation algorithm, the analysis of remote sensing image segmentation algorithm based on the international status and trends, the paper focuses on the microcanonical annealing algorithm, fuzzy entropy, fuzzy Kohonen clustering network and the EM algorithm. While the spatial information and remote sensing image pixel information is also taken into account in remote sensing image segmentation, experimental results show that the algorithm has better segmentation effect and strong anti-noise performance.Remote sensing image local information is divided into space information and the pixel information. So-called partial information is introduced first, the image space information and the pixel information to reshape the image pixel histogram information, enhance the pixels between class scatter and within the compact class. Remote sensing images are vulnerable to noise, so the image segmentation of remote sensing has to consider the noise factor at the same time. This article process remote sensing image space information and the pixel information at the same time, the anti-noise performance achieved satisfactory results.A one-dimensional fuzzy entropy threshold segmentation algorithm based on microcanonical annealing algorithm is one-dimensional fuzzy entropy threshold segmentation algorithm and microcanonical annealing algorithm combined. Microcanonical annealing algorithm is a global optimization method, with the advantage of fast convergence. The microcanonical annealing algorithm and one-dimensional fuzzy entropy threshold segmentation algorithm can effectively improve the combination of one-dimensional fuzzy entropy threshold segmentation algorithm accuracy and calculation speed.A new image segmentation algorithm (WEFK) based on a combination of EM algorithm and FKCN algorithm, the first to ask a pixel statistics, probability of each pixel, then through EM algorithm in various gray levels for each the probability and probability normalization, will be normalized the Gaussian probability as FCM fuzzy membership. Finally, it is combined with fuzzy Kohonen clustering network, and finally form this WEFK algorithm, this algorithm has better segmentation accuracy.