Research On Fuzzy C-means Clustering Method For Remote Sensing Image

Remote sensing technology is one of the most rapidly developed and widely applied scientific technologies in recent years.Acquiring and transmitting remote sensing images and extracting information from remote sensing images is the main research aspect of remote sensing.However,the noise is inevitable due to the influence of external interference factors in the process of obtaining transmission,.When using remote sensing image to classify images and obtain image information,these noises may affect the effect of image classification to an indeterminate degree.Fuzzy c-means clustering is a widely used classification in remote sensing image classification method,its characteristic is that the clustering results is uncertain,which only provides membership degree matrix as the clustering results.The membership matrix is the probability about the image of the classification of each pixel.If getting the exact classification about each pixel to the membership matrix,the membership matrix needs to be processed,which is called clustering post-processing.This paper embarks from the fuzzy c-means clustering post-processing method,studies the fuzzy c-means clustering post-processing method combined with the image noise reduction method to improve the clustering post-processing method,makes post-processed image quality improved obviously.The main research results are as follows:(1)This paper studies the fuzzy c-means clustering and post-processing method deeply,and explores the membership matrix on the basis of relevant features.In this paper,image processing is carried out on the image of the subordinate degree of each classification,and finally the idea of new membership matrix is synthesized.This idea has guided the following three methods of improving post-processing.(2)Based on the traditional maximum membership method,this paper proposes three methods of post-treatment with anti-noise performance.After numerous tests,it has been demonstrated that the noise of the three methods after processing is suppressed,the classification accuracy and signal-to-noise ratio are improved to varying degrees,and the increase in accuracy varies from 3% to 20% depending onthe type of image.The three post-processing methods proposed in this paper are median filter post-processing method,wiener filtering reprocessing method and mixed filter post-processing method.In addition,this study improves the processing flexibility of a kind of neighborhood weighted membership and reprocessing methods.(3)In this paper,five kinds of post-processing methods,including the maximum membership method,are tested and compared in five different remote sensing images.Then,by analyzing the application of these five methods to five different remote sensing images,the paper provides some suggestions on how to use the post-processing methods of various remote sensing images in the post-processing of clustering.