Research On Detecting Algorithm Of Moving Dim Small Targets In IR Images With Complex Cloud Background

High maneuverability and stealthy-ability are the main developing directions of present weapon systems, such as ballistic missile, cruise missile, long-range fighter-bomber, etc, and they often emerge as moving targets in the sensed images. Since rapid reaction and far-distance detecting ability is the unique efficient way to countermine these weapons, it is of great importance to realize real-time detection of moving targets on surveillance satellites. This thesis majors in the detection of small moving targets with clouds as the main clutters, and the work here belongs to the information processing task of the IR images captured by missile-surveillance satellite. According to the characteristics of moving targets and cloud clutters, the detecting algorithm based on Median Subtraction Filter (MSF) is proposed. Experiments show that the algorithm has robust detection ability and is suitable for several typical cloud-backgrounds.This thesis is composed of four chapters that are summarized as following:The 1st chapter is an introduction to the context of the detection task and the difficulties that emerge in the detection of small targets in IR images from space-borne satellite. After a detail review of the relevant works, a principle diagram of the detection algorithms used in this thesis is shown.The 2nd chapter mainly concerns with detection algorithms based on background-suppression. Taking account of the statistic characteristics of clutters in IR images and the requirement of detection on satellite, a fusion algorithm is proposed that coalesces the median subtraction filter (MSF) and the edge-detection operator. For IR images with huge amount of data quantum, a simpler algorithm is proposed which is based on MSF and twice-threshold segmentation. At the end of this chapter, taking account of the motion characteristics of small targets in IR image sequences, a judging method based on neighborhood is discussed and used.In the 3rd chapter, performance evaluation is done on the detection algorithms based on background-suppression. With a thorough analysis of the key steps in the proposed algorithms, the relationship of the detection probability (DP) and false alarm probability (FAP) to the SNR and the parameter of detection threshold are shown for the single frame case. From the theory of binominal distribution, the relationship of the total DP and FAP for multi-frame case to those of the single frame case is shown. The analysis here offers a theory base for the selection of the threshold and the frame-number in multi-frame detection. Experiments show that the detection algorithm in 2nd chapter has robust detection ability.In the last chapter, the work in this thesis is summarized, and the key problems and direction of the future research are outlined.