Research On Dynamic Programming Track-before-detect Algorithm Based On Saliency For Infrared Dim And Small Target

National defense security has always been the fundamental guarantee for maintaining a country's stable development.The infrared detection system is suitable for the military field such as precise guidance and early warning system because of its all-weather,concealment and stability.Therefore,infrared image dim and small target detection technology has been the focus of scholars at home and abroad.In recent years,the saliency detection technology based on bionics has opened up a new research direction for infrared dim and small targets detection.In this paper,the saliency of infrared dim and small targets in frequency domain,space domain and time domain is studied for the problem of high false alarm rate of dim and small targets with complex background and low signal-to-noise ratio.The main research contents are as follows:Aiming at the problems existing in the existing saliency detection algorithms,such as large amount of calculation,sensitivity to noise and non-directionality,a saliency algorithm based on difference of local regional is proposed.This algorithm takes 8 neighborhood mean contrast as the template to convolute with the input image,and takes 8 neighborhood minimum contrast as the saliency output,and comprehensively considers the problems of noise,computation and direction.By improving the frequency domain tuning spectral residuals algorithm to obtain the region of interest,the final saliency is obtained by using local area difference in the region of interest.Finally,the experimental results of extracting the target location according to the visual attention mechanism show that the proposed algorithm can achieve good results in each scene.Aiming at the problem of high false alarm rate of single frame image detection technology in complex background with low signal-to-noise ratio,considering the continuity of saliency in time domain,a dynamic programming track before detect algorithm based on saliency intensity is proposed.The algorithm takes the saliency difference between frames as the value function,designs the search strategy according to the attention transfer mechanism,and updates the search range in real time by using the target moving speed.After verifying the probability distribution of the inter-frame saliency difference images,a dynamic programming track before detect algorithm based on the posterior probability density function is proposed.The value function of the algorithm is the product of the saliency likelihood ratio of the target and the cost function.Experimental results show that the performance of the proposed algorithm is better than that of the single-frame image detection algorithm and other dynamic programming track before detect algorithms.Aiming at the problem of inter-frame correlation of each potential trajectory in dynamic programming search process,a dynamic programming track before detect algorithm based on kernelized correlation filter is proposed.In this algorithm,the correlation between the target and the potential trajectory at the next moment is obtained by kernelized correlation filter and it is used as the weight of the phase value function.Aiming at the fact that the nuclear correlation filter is easy to be offset by the background edge when tracking the infrared dim and small targets,a multi-channel tracking algorithm with kernelized correlation filter based on saliency is proposed.In this algorithm,multiple features,such as local region difference saliency graph,are used as input to carry out kernelized correlation filter,and then the response value of the adaptive fusion multi-channel is obtained to the final confidence graph according to the peak side lobe ratio of each channel.Simulation results show that the proposed algorithm can achieve better detection and tracking performance.