A Study Of Algorithms For Complex Background Suppression And Small Target Detection

Infrared Surveillance System (IRSS) is used for target detection, recognition and trackion which employ electro-optical detectors receiving infrared emission passively. IRSS has advantages of high resolution, invisibility, fine anti-jamming ability, small size, lightweight and high maneuverability. For these advantages, IRSS becomes an important part of modern information confrontation system, and obtains more attention in recent years. As the speed of missiles and planes increase, IRSS is required to meet the demand of detecting long distance targets to ensure enough pre-alarm time. However, it is difficult to separate targets from background because of its small size, usually occupying only several pixels, no shape and texture information and so on. Therefore, how small and dim targets can be reliably detected in complex background has become a hot research issue and has important theoretical and practical value.Aiming at the problem of small and dim target detection in complex sky background, the characteristics of target and background in infrared images are analyzed in this dissertation. Singular value decomposition (SVD) theory, ant colony system (ACS), multi-scale geometry analysis, bilateral filter and non-linear filter are deeply studied and used for infrared dim and small target detection. This study mainly focuses on background suppression algorithm for infrared small target and multi-frame infrared small target detection algorithm.1. Aiming at the case that infrared images contain highlighted and large area background, two background suppression algorithms are proposed. According to the singular value with invariance principle in the algebra and geometry, a background suppression algorithm based on SVD and ACS is proposed to suppress complex sky background. The experimental results show that the algorithm can restrain the complex and various background structures, improve the SNR and contrast of image, but it has heavy computation. So a background suppression algorithm based on multi-scale SVD is developed to make it real-time implemention. According to the stationary wavelet characteristics with multi-scale and translation invariance, the algorithm suppress the background based on singularity value decomposition which adjusted the stationary wavelet sub-bands coefficients with truncated eigenvalue. Experimental results demonstrate that the algorithm can suppress the background and enhance the target signal effectively. 2. In order to suppress strong undulant background with complex texture effectively, two background suppression algorithms are developed based on new bilateral filtering. Through analyzing the relationship between each pixel with gray level, geometric structure and local neighborhood pixel, an improved bilateral filter algorithm is proposed, which can separate targets from background with prediction of the background image. By the advantage of the nonsubsampled contourlet transform (NSCT) characteristics with multi-scale, multi-directional and translation invariance. A new infrared dim target background suppression algorithm based on bilateral filter is proposed which adjusted the NSCT sub-bands coefficients with local neighborhood pixel in spatial distances, geometric structure and intensity. Experimental results show these algorithms can restrain complex background for surface-sky detection system effectively.3. Two dim and small target detection algorithms are proposed based on the sequence image. Using inter-frame correlation and motion continuity of target, as well as the constructed gray correlation coefficient, a dim target detection algorithm based on Kalman filter is proposed. Since the actual target usually has high maneuverability, which means it has strong non-linear motion, so a unsecured particle filter for sequence detection algorithm is developed. Experimental results demonstrate that new algorithms can detect moving target under different motion model accurately, reduce the number of false targets and missed targets, and improve the detection capability of dim and small target in image sequences.4. The performance of dim and small target detection algorithm is also studied in this dissertation. The performance of dim and small target detection algorithm is associated with many factors. As image quality is concerned, SNR, contrast ratio etc are involved as evaluation indicators. As for detection requirement, detection probability, false alarm probability etc are involved. In this study, the relationship between the false alarm probability and detection threshold is studied deeply, so as the relationship between detection probability and SNR, the detection probability and the false alarm probability. The ROC curves for performance evaluation in infrared dim and small target detection are also given. Simulation experiments evaluate the performances of the proposed algorithms using the ROC curves, which proves that the algorithms proposed are more efficient compared with the classical algorithms.