Detection of Small Target in Infrared Image Based on KFCM and LS-SVM

Aim at the problem of small targets detection in complex infrared background, a method based on the kernel fuzzy clustering and least squares support vector machine (LS-SVM) background predication is proposed. First, partition the training samples by a nearest-neighbor clustering method to get the clustering number and the initial clustering centers. These clustering centers are further processed using kernel fuzzy C-means (KFCM) method. Then, the tuning parameters of the fuzzy model are estimated by LS-SVM. Further these tuning parameters are used to predict the background of infrared images. The prediction image subtracted from the source infrared image gives the residual image. Finally, a threshold selection method based on recursive maximum between-cluster absolute difference is presented to separate the real small target from the residual image. Experimental results are given and they are compared with the results of fuzzy C-means (FCM) detection method. They show that the proposed method has better detection performance.