Infrared Dim Targets Detection Based On High-Order Statistics

Nowadays, radars are faced with heavy threats from counteraction. Passive detecting technique is an effective way to overcome this problem. Furthermore, the passive infrared detection is an important research field and can improve the survival and counter attack capacity of radar systems. Dim and small targets detection of infrared images is a key technique for the passive infrared system and has been widely investigated for decades.This thesis mainly deals with the dim and small targets detection methods of infrared images based on High-Order Statistics (HOS). According to the infrared image model, the characteristics of noise, dim targets and background are analyzed respectively, followed by an introduction to HOS, which are the basis of the later detection methods. For a single frame image, a spatial detection method based on Fourth-Order Cumulant is analyzed. Due to the frequency differences between the target signal and background signal in the same image, a novel detection method based on Bispectrum is proposed, which suppresses the noise and increases the target information simultaneously. Both of above methods using the features of a single frame result in the lower precision when the gray value of background changes greatly. So, for the image sequences, two detection methods based on Third-Order Cumulant and Bispectrum likelihood ratio are given respectively. The former views the fluctuation of grey-scale values of pixels of the corresponding images, caused by a passing target, as a non-Gaussian signal, which can be detected by calculating its Third-Order Cumulant. While the latter estimates the pixel positions where a target passed by and then determinate the target trajectory, by calculating the Bispectrum likelihood ratio in frequency domain. Simulation results on various real data indicate that the presented methods are effective and robust to detect dim and small targets of infrared image.