Study On Infrared Dim Small Target Detection And Tracking Under Complex Sky Background

The detection and tracking of remote target is a key technique in the sky early warning systems and the weapon systems.Due to the long imaging distance,the remote targets are dim and small in the infrared image.Moreover,the sky background clutter is usually heavy and complex.Therefore,the detection and tracking of dim small target under complex clutter is a challenging problem.The main reason why it is difficult to detect and track dim small targets under heavy complex sky background is that the existing algorithms have limited representation ability for signal spatial structure and temporal structure.In this dissertation,several advanced methods for sky clutter suppression,dim small target detection and tracking are presented by improving the representation ability of signal spatial and temporal structure.(1)In order to reduce the disturbance of heavy sky background clutter,the fractal clutter dictionary sparse representation based clutter suppression method is proposed.The sky background clutter can be accurately represented by fractal random field,so the fractal theory is introduced for constructing clutter over-complate dictionary.By the sparse representation on fractal clutter dictionary,the clutter component of the infrared image can be estimated and perceived.By subtracting the estimated clutter component from the infrared image,the clutter is effectively suppressed.(2)For improving the accuracy and robustness of infrared dim small target detection,the high-dimension vector field structre based small target detection method is presented.The gradient vector field is utilized to raise the data dimension of infrared image,and the structure difference between dim small target and clutter in the vector field is obvious.The target shows negative source structure and the clutter edge shows direction coherence structure in the gradient vector field.The multi-scale flux density is presented to represent and measure the negative source structure,and the gradient direction diversity is introduced to represent and distinguish clutter edge structure.By combing the information from both multi-scale flux density and gradient direction diversity,the proposed dim small target detection method can recognize the small target even when the clutter disturbance is heavy.(3)For representing the target spatial structure more effectively and efficiently,the generalized Gaussian target dictionary sparse representation based target structure parameter extraction method is proposed.The target dictionary is generated by generalized Gaussian intensity model,and the atoms of generalized Gaussian dictionary contains clear structure information of target.Therefore,the structure parameter of target can be extracted by the sparse representation on generalized Gaussian target dictionary.The structure parameter not only can represent the target structure accurately,but also can reflect the temporal change of target structure more clearly.(4)In order to improve the tracking stability for appearance changing target,a Bayes inference based small target tracking method is proposed.According to the short term observations,the short term target structure parameter distribution is established,and according to the long term observations,the prior distribution of target parameter distribution is built.By simultaneously considering the short term and long term information,the Bayessian inference based method can presented the spatial-temporal structure of target.Therefore,the appearance of target can be predicted and the tracking method can be more robust to appearance changing target.(5)In order to observe and analyze the performance of different signal representation methods,the graphical infrared small target detection and tracking software is developed.By using this software,the performance of different infrared small target processing methods can be compared and analyzed in visual and quantitative ways.The signal representation methods presented in this dissertation can effectively represent the structure of target and clutter,therefore the performance of infrared small target processing algorithms are enhanced.Extensive experiments shown that the proposed methods are both accurate and reliable for cltter suppression,dim small target detection and tracking under heavy complex sky clutter.