Research On Infrared Small Target Detection Algorithm Based On Spatio-temporal Tensor

With the rapid development of modern military science and technology,various precision-guided missiles play an important role in modern warfare.Among them,the infrared detection system ensures the detection and tracking of hit targets,and is an important part of missile precision guidance.With the rapid development of modern aerospace technology,the flight speed,maneuverability and sensitivity of precision-guided missiles are getting higher and higher,so higher requirements are put forward for the performance of the target detection system.However,in practical situations,the background has a great influence on the effect of ultra-long-distance imaging,the signal-to-noise ratio of the image obtained from the infrared sensor is very low,and the number of target pixels in the image is very small,resulting in small targets in the infrared image.Detection is very difficult.However,in practical situations,the background has a great influence on the effect of ultra-longdistance imaging,the signal-to-noise ratio of the image obtained from the infrared sensor is very low,and the number of target pixels in the image is very small,resulting in small targets in the infrared image.Detection is very difficult.Aiming at the low rank exhibited by the background in infrared images and the sparseness exhibited by some strong edges,an efficient tensor completion method is proposed,which simultaneously utilizes the low rank approximation as a global guide,and uses the sparse representation as a tensor Quantitative completion of local cues.For low-rank approximations,a weighted kernel norm with an efficient weighting scheme is introduced to measure the rank of a tensor,and weighted kernel norm minimization using an adaptive weighting scheme allows different singular values to shrink with different thresholds,resulting in better close to the rank.For sparse representations with strong edges,an orthogonal dictionary learning process is integrated into the sparse coding stage to discover local patterns in tensor data more efficiently.In view of the sparsity and lack of robustness of objects in small object images,we noticed that the local structure and texture representation of small objects can be characterized by two attributes of intensity gradient.In this thesis,the target is enhanced by a balanced structure texture representation that combines local gradients and local intensities.The method makes full use of the structure and texture of the latent space in the infrared image,while suppressing the background clutter and improving the detection probability of small infrared targets.In addition,we also added an additional balance factor to deal with different scenarios and improve the adaptability of the model.Last,the infrared small target model based on spatio-temporal tensor proposed in this thesis is solved by the alternating direction multiplier method(ADMM).And compared with some other excellent algorithms,it is proved that the algorithm has better robustness through BSF,SCRG and other indicators.