Robust Semi-supervised Kernel-FCM Algorithm Incorporating Local Spatial Information for Remote Sensing Image Classificat
- PDF / 2,238,679 Bytes
- 15 Pages / 547.087 x 737.008 pts Page_size
- 14 Downloads / 176 Views
RESEARCH ARTICLE
Robust Semi-supervised Kernel-FCM Algorithm Incorporating Local Spatial Information for Remote Sensing Image Classification Chengjie Zhu & Shizhi Yang & Qiang Zhao & Shengcheng Cui & Nu Wen
Received: 20 February 2013 / Accepted: 14 May 2013 # Indian Society of Remote Sensing 2013
Abstract Fuzzy c-means (FCM) algorithm is a popular method in image segmentation and image classification. However, the traditional FCM algorithm cannot achieve satisfactory classification results because remote sensing image data are not subjected to Gaussian distribution, contain some types of noise, are nonlinear, and lack labeled data. This paper presents a robust semi-supervised kernel-FCM algorithm incorporating local spatial information (RSSKFCM_S) to solve the aforementioned problems. In the proposed algorithm, insensitivity to noise is enhanced by introducing contextual spatial information. The non-Euclidean structure and the problem in nonlinearity are resolved through kernel methods. Semi-supervised learning technique is utilized to supervise the iterative process to reduce step number and improve classification accuracy. Finally, the performance of the proposed RSSKFCM_S algorithm is tested and compared with several similar approaches. Experimental results for C. Zhu : S. Yang : Q. Zhao : S. Cui : N. Wen Center of Optical Remote Sensing, Anhui Institute of Optics and Fine Mechanics, Chinese Academy of Sciences, Hefei 230031, China C. Zhu College of Electronic Engineering, Anhui University of Science and Technology, Huainan 232001, China C. Zhu (*) Huaibei, Anhui, China e-mail: [email protected]
the multispectral remote sensing image show that the RSSKFCM_S algorithm is more effective and efficient. Keywords Kernel-FCM . Remote sensing image . Image classification . Semi-supervised . Local spatial information
Introduction Remote sensing image is utilized in various applications of natural resources inventory and management and weather services. Image classification plays a key role in these applications. Many algorithms and approaches were proposed in the past to classify satellite images. These algorithms and approaches have achieved huge progress. Fuzzy c-means (FCM) algorithm is a popular method in image classification because of its structural and computational simplicity. The success of FCM is mainly attributed to the introduction of fuzziness for the belongingness of each image pixel. Some of the complications of real-image data, such as image context spatial information and nonlinear space, are not considered by standard FCM (Rafael 1997; Liew et al. 2000; Pham 2002; Chen and Zhang 2004; Tamma et al. 2011). For example, the traditional FCM algorithm does not consider the context spatial information of the image, indicating that FCM is very sensitive to noise and outliers. In addition, standard FCM utilizes Euclidean distance to compute the objective function, leading to discontented classification results. Many
J Indian Soc Remote Sens
authors proposed several approaches and algorithms to improve FC
Data Loading...
