A novel method for infrared small targets detection

Small target detection is widely used in many fields, such as environmental monitoring and assessment, space remote sensing, recognition and tracking of infrared target. In this paper, aiming at the problem of over-fitting and poor generalization capability in using artificial neutral network for small target detection, a new method is presented. First, it uses the structure element to set up training samples. Then, based on support vector machine theory, it builds learning machine by selecting proper model and trains the samples. The result can be used to suppress the background of following image. Finally, an improved weighted variance local entropy method is presented to partition the image, which can get lower false alarm probability than traditional local entropy in the complex background. The simulation results show that this method is more effective for low SNR image than traditional neutral network and local entropy method.