Research On Techniques Of Detection And Discrimination Of Point Target In IR Image

It is a great challenge for modern defense technology to detect, track and recognize a target at far range from the detector . Point target detection and recognition are the key technique to improve working distance of defense system, and widely used in long-distance infrared early warning system, infrared imaging guidance , etc. In this dissertation, we have comprehensively studied the detection and recognition techniques for point targets under complex background and deep space background, and some valuable results are achieved.When infrared warning system detect moving objects in low altitude, detection performance will decrease significantly due to the cloud appeared in the image. In this paper, we researched the temporal profiles of target and cloud on a detector respectively, and build four temporal models. Based on these models, a temporal detection algorithm is proposed. According to theoretical analysis and experiment results, this new algorithm can effectively detect the point target in different cloud backgrounds.With the development of detector technology, detector's size and sampling rate is increasing, so the complexity of detection algorithm is increased fast. It is important to develop fast detection algorithm to meet the requirements on computation ability and memory of real system. In order to make KKV detect target in deep space in real time, Multistage Velocity Matching algorithm (MSVM) is presented. Compared with other detection algorithm, MSVM have higher detection probability with less computation requirement. To decrease the computation furtherly, a Modified Multistage Velocity Matching algorithm(MMSVM) is presented, which is faster and less computationally than Multistage Velocity Matching algorithm.Point target recognition is the most difficult task in anti-missile system. We have studied the temperature changing properties of space object, and analyzed the difference between target and decoys on infrared feature. A Recognition method based on modified time-delay neural network is designed, which use space object's temperature and luminance change frequency as recognition feature. By simulation with different condition of target and decoys, the results are prospective and satisfied.