Weak Target Detection Based On Multiband Optical Imaging And Chain Optimization

The weak target detection technology using optical imaging is widely applied in military and civil fields. There exist lots of degradation factors in the optical chain imagery detection, such as aberration of pptical system, electronic noise of imaging sensor, relative movements between scene and sensor, which would lead to the low signal-to-noise (SNR) and signal-to-clutter ratio (SCR) of target image. Moreover, because of long-distance imaging, target is small and lack of detail characteristic, target signal is so weak relative to background clutter and noise. All of those factors lead to the weak characteristics of target. With the increasement of demand for long-distance detection and low false alarm rate, traditional optical detection methods have been unable to meet demand. How to solve the traditional optical imaging detection problem under complicated background and effectively improve the ability of weak target detection is the key research point in this paper. Using image pre-processing technology for optimization and compensation in imaging chain, multiple means are adopted to improve the SNR of target image. And the author tries his best to design excellent algorithm for target detection, to achieve high Probability of Detection (PD) with low Probability of False Alarm (PFA), and improve the ability of weak target detection to satisfy the demand of many kinds of application.The framework of multi-mode weak target detection technology is analyzed. Based on pricinple of full-link imaging, the author analyzes the feasibility of multi-mode target detection technology. We have discussed the optical reflection and radiation character of typical target and background, which could be used for target detection and background suppression. And we give out the experimental design for multi mode weak target detection, including image sensor, experimental platform and system, indoor experiment design and so on. Finally, we figure out the discriminant rules for weak target.In order to improve imaging SNR. we try to analyze compensation mechanism based on degradation in imaging chain. The typical models of image degradation are introduced. Several typical image restoration methods are introduced and analyzed. We propose one full-reference method and one no-reference algorithm for image quality assessment, named gradient-based ripple image quality assessment approach and the algorithm combined edge width and blurring, respectively. For image restoration, we propose local constrained Richardson-Lucy deconvolution approach, introducing spatial weight matrix for local constraint which can solve the balance between ringing suppression and details preserving. And we also propose the gradient constraint regularized fast image restoration methos, which does well in edge preserving and reduce noise.In order to improve infrared imaging SNR, we also try our best to analyze the noise suppression and compensation in infrared imagery. We deduce the non-uniformity correction model. In the field of non-uniformity correction, two methods are designed, the one is based on single image, and the other need sequence images. The sequence images based approach works real time,0.02seconds for each frame with size of354x236.A lot of work has been done in weak target enhancement and multi-band image fusion. Firstly, with the help of local frequency based saliency extraction, we propose an enhancement method, which could strength the contrast between target and background. Secondly, in order to well utilize the information of multi-band images, we combine multi scale analysis and saliency extraction technologies together to design the image fusion algorithm, which can keep even enhance the details of original image.According to the features of target in image, we propose several ideas for target detection. Considering the high constrast in target area, a target detection method based on saliency detection in Fourier domain is proposed. Because those targets usually exist in region with large local difference, local similarity difference is used for target detection. In order to detect target accurately, we make use of sparse representation theory for e automatic target detection. Finally, a real-time automatic small target detection method using saliency extraction and morphological theory is designed.