Classification Of Remote Sensing Image Based On Spatial Information Kernelized Fuzzy C-Means Algorithm

The development of remote sensing technology has played an important role for people to discover and observe the earth. Satellite remote sensing images provide abundant observational data to earth observation. In order to utilize remote sensing data effectively, and transform the spectral information of remote sensing image to the classification information of users, it is needed to analyze and interpret the remote sensing image effectively.People still use manual method in the respect of classification mapping now. This method not only requires a lot of human and material resources, but also consumes a lot of time. The classification method of remote sensing image based on computer technology, is a significant improvement to the traditional methods. There are two major types of remote sensing image classification:supervised and unsupervised classification. Unsupervised classification method has become a research hotspot in the field of remote sensing image classification. This paper focuses on unsupervised classification algorithm based on fuzzy theory, and has completed the following works:1. Expound the basic principle of remote sensing image classification, introduce the principle and method of the accuracy evaluation of remote sensing image classification, summarize the common methods of remote sensing image classification.2. In this paper, a new unsupervised classification method is given:fuzzy C-means clustering (FCM) algorithm. The FCM algorithm is sensitive to the noise in the process of classification. Aiming at the shortcoming, a fuzzy C-means clustering algorithm based on spatial information (SFCM) is given. The SFCM algorithm can make use of spatial information of the image, thereby achieving the purpose of accurate classification for the remote sensing image containing noise. Experimental results show the effectiveness of the SFCM algorithm.3. The nonlinear processing ability of the FCM algorithm is limited, however, most of the remote sensing image classification is the nonlinear processing problem. Aiming at this problem, a fuzzy C-means clustering algorithm based on kernel function is given in this paper:kernelized fuzzy C-means clustering (KFCM) algorithm. With the kernel method, the input data is mapped implicitly into a high-dimensional feature space where the nonlinear pattern now appears linear in the KFCM algorithm, thereby achieving the purpose of improving the classification accuracy of remote sensing image.4. Aiming at the above two shortcomings of the FCM algorithm, an improved algorithm is proposed in this paper:spatial information kernelized fuzzy C-means clustering (SKFCM) algorithm. A classification experiment on remote sensing image containing noise is accomplished by KFCM, SFCM and SKFCM algorithm, and the experiment results indicate that SKFCM algorithm not only has better noise immunity, but also has higher classification accuracy.