| Infrared target detection is a key technology in many applications,including maritime surveillance systems,missile tracking and interception systems,and forest early warning systems.Due to the long imaging distance,the target size is small,the contrast is low,and no prior knowledge of the texture information and target structure is available.At the same time,due to the limitations of infrared imaging itself,infrared targets are often drowned by low clutter and noise.Signal-to-noise ratio in a complex background,so the detection of infrared target is still a challenging problem.This thesis focuses on the difficulty of detecting small infrared targets in complex environments.We propose an infrared target detection algorithm that overcomes the above problems.This thesis proposes an infrared target sensing method based on saliency filtering regularization and non-local low rank.The newly proposed detection framework can be customized as a non-convex optimization problem,where joint target visualization can be performed in a low-dimensional discriminant manifold.Sexual learning.We combine and reconstruct similar patches to better generalize the non-local spatial low rank constraints.At the same time,in order to highlight target saliency learning and suppress noise and clutter interference,we propose entropy-based saliency filtering regularization.item.The saliency filtering regularization preserves the context information of the target and the surrounding area,and avoids the approximate deviation of the low-rank matrix.Experimental evaluation of real infrared images shows that compared with some of the latest methods,the method is more accurate and more robust in different complex scenarios.In this thesis,we propose a new infrared small target detection method that takes into account both the structural prior knowledge of the target and the background autocorrelation.First,a tensor model is constructed for the high-dimensional structural features of infrared sequence images.Secondly,inspired by low-rank background and morphological operators,a new target detection method based on low-rank tensor completion and circular Top-Hat regularization is proposed.This method uses a ring-shaped Top-Hat operator to appropriately utilize the local prior structure of the target area to reduce the effects of noise and clutter,and extends the weighted Schatten-p norm to the tensor model,thereby making full use of low-rank tensor Multi-dimensional structural features.Finally,a unified optimization framework is proposed,and the proposed tensor model is optimized and solved using the alternating direction multiplier method(ADMM).In addition,compared with related methods,this method can not only improve the signal-to-noise ratio gain(SCRG)and background suppression factor(BSF),but also provide a more powerful detection model under the condition of low false alarm rate.Finally,we summarize the work of this thesis,discuss the improvement of the proposed algorithm,and look forward to the future work of infrared target detection algorithms. |
