Low Snr Infrared Target Single-frame Detection Technology

Infrared small targets detection and tracking in complex environment which aims at the point target of long distance and the plane target of short distance is a key technology in IR guidance system and IR alarm system. A large number of low-altitude feature and characteristics of the small and dim targets—long distance, low SNR and strong clutter make the detection and tracking even harder. So each step of detection and tracking of the small and dim targets is studied in this paper and it stresses on the application of wavelet transforms in small target detection and tracking.First, the noise source of infrared image is studied and the noise model is established. And on this basis the targets model is established. Then several typical image preprocessing technology are studied, such as high-pass filtering, median filtering, Tophat transformation in Mathematical Morphology and wiener filtering. The goal of preprocessing is to promote SNR and restrain the background of infrared images. Then the merits and demerits of preprocessing technology are analyzed by simulating with MATLAB.Second, wavelet transforms is introduced in image processing in this paper. Wavelet transforms has good time-frequency localized characteristic so that it can achieve a better balance in the protection of signal feature and restraint of noise. This is an incomparable advantage that the above-mentioned preprocessing algorithms don't have. Based on the basic theory, the application of wavelet transforms in image preprocessing is analyzed, and the analyse stresses on several problems in practical application which need to be attention. According to the de-correlation and energy compaction character of wavelet transforms, an improved de-noising method is proposed against the demerit of the WaveShrink method. It can avoid unreasonable threshold due to the error estimation of noise. Simulation results show the improved method can preserve the shape character of small target, suppress the background, and work efficiently in de-noising.The detection and localization of small target is studied in the last part of this paper. Two typical methods—threshold segmentation and edge detection are introduced. And their deficiencies of application in small target detection are studied. Then the research focus on the application of wavelet transforms in target detection and localization. The wavelet transforms of signal has strong correlation in each scale but the noise doesn't have. The wavelet transforms of edge points have great correlation in each scale and the edge points are also localized well in each scale but the noise concentrated just in the small scale mainly. According to this feature, a new multi-scale wavelet energy cross method is proposed. And then the multi-scale Gaussian wavelet operator for edge detection is constructed. At last, adaptive thresholding segmentation method is used for detection and localization of target, and then the centroid of targets is calculated. Simulation results show that the candidate targets getted through detection is clearly positioning and the method works efficiently in sigle-frame detection.