Research On Infrared Dim Small Target Detection And Tracking Algorithm In Complex Background

Infrared dim target detection and tracking in complex background is widely applied in military defense system.It is a key technology,including early warning,precision guidance,missile tracking system and marine monitoring system.With the application of anti-radiation radar and stealth aircraft,radar system is facing severe challenges.Passive infrared detection technology can avoid these problems.Usually infrared image background is heterogeneous and the target is dim.Detection and tracking of dim targets in complex background is the difficulty of infrared guidance,is also the core issue,and has attracted wide attention.In this thesis,the problems of infrared dim target detection and tracking in complex background are analyzed.The main contents are as follows:1.The background and target characteristics of infrared image are analyzed.In the detection and tracking of dim and small infrared targets,the background occupies almost all the pixels in the infrared image,and the target occupies only a few to a dozen pixels,and the local background has strong correlation with its adjacent regions.Based on the above background and target characteristics,the target with low rank background is sparse after image processing.2.Aiming at the problem of infrared dim target detection,the detection methods of weighted low rank and enhanced sparseness based on structural priori are analyzed.After image processing,the detection results can be obtained by using the low rank characteristics of the background and the sparse characteristics of the target.However,the method is approximate,and the sparse characteristics of the background edge will also affect the detection results.Considering the above shortcomings,the third chapter proposes the methods of weighting the background image and enhancing the sparse target image,which make the model more accurate.At the same time,in order to suppress the influence of background edge on detection,the structure tensor is proposed to suppress background edge,which can greatly reduce the false alarm rate.Finally,through the verification of the measured infrared data,the proposed weighted low rank and enhanced sparse detection method based on structural priori can achieve good detection results.3.Aiming at the problem of target feature extraction,the feature extraction method of infrared dim small target based on deep learning is analyzed.Once the target is detected,it is necessary to track the target in real time.At present,the target tracking method based on correlation filtering achieves good tracking performance.The premise is to extract good target features to train correlation filters.However,infrared dim and small targets only occupy a few pixels in the image,without obvious texture structure information,it is difficult to extract effective features.In the fourth chapter,a target extraction method based on deep learning is proposed.Firstly,a convolution neural network is constructed.Then,the trained convolution neural network is used to extract the features of infrared dim and small targets,and the convolution features are obtained.Finally,the convolution features are used to train the correlation filter.4.Aiming at the problem of target tracking,the tracking method of regularized correlation filtering in central bias space is analyzed.Based on the spatial regularized correlation filtering,this thesis improves it,and proposes a central offset spatial regularized correlation filtering,which can suppress the boundary effect and reduce the computational complexity.Finally,the infrared dim and small target tracking method based on deep learning and center bias space regularization is proposed.Through a lot of experiments and comparative analysis,the tracking method proposed in this thesis can achieve good tracking performance.