Classification of hyper-spectral images with Probabilistic Fuzzy Kernel based Fuzzy C-Means clustering and Support Vector Machine

A framework of Probabilistic Fuzzy Kernel based Fuzzy C-Means (PFK-FCM) and Support Vector Machine (SVM) methodology is presented to achieve a superior performance for the hyper-spectral images classification problem. Principle Component Analysis (PCA) is employed as the preparation step to extract the principle components of the hyper-spectral image for the computation simplicity. The concept of probability is integrated to kernel design to deal with uncertainties in the data points, which could be involved in the sample selection and the classifier learning task. A new clustering method, which incorporated with the novel PFK kernel, is presented to improve the clustering results by the optimization process. PFK-FCM is designed to select the pixel samples for the learning of the SVM classifiers. By integrating SVM with PFK, PKF-SVM is also presented and its model would be trained as classifier for the pixels classification in hyper-spectral image. Extensive experiments are performed both on the Indian Pine and Salinas datasets to demonstrate that the proposed PFK based FCM and SVM method can obtain better performance than those state-of-the-art classification techniques.