A Weighted Kernel-based Possibilistic Fuzzy Clustering With Its Applications in Aerospace Image Segmentation

While possibilistic and fuzzy C-means clustering is one of the essential soft clustering algorithms in machine learning, its effectiveness is limited to complex geometric shapes and nonlinear separability data. We propose a weighted kernel-based possibilistic and fuzzy clustering algorithm (WKPFCA) to solve this problem. The proposed WKPFCA considers the contributions of different features to each cluster in the kernel clustering process, which reduces the influence of irrelevant (bad) features and increases the good ones. Compared with the existing hard and soft kernel-based clustering algorithms, the proposed WKPFCA is more robust and can generate more stable cluster centers. Experiments are carried out to verify UCI real data sets and aerospace images, and the proposed WKPFCA has certain superiority, compared with some classical and state-of-art algorithms,. This paper is beneficial to the application of image segmentation.