Research On Clutter Suppression Based On Deep Learning

In recent years,with the rapid development of the stealth technology,computer technology and new materials technology,radar target has the stronger ability of invisibility.Target signal with the characteristics of low signal-to-noise ratio easy to be covered in the background noise,so traditional methods of radar target detection is still facing great difficulties.At present,the radar clutter suppression technology is mainly divided into the method based on tracking and classification method.Method based on tracking mainly use the property differences between target and clutter signal and motion feature of target,achieve the goal of clutter suppression by using the method of tracking.The method has been able to achieve good results and applied to the practical radar system,however there is still a low signal-to-noise ratio effect is poor and bad real-time performance.The method based on feature classification need expert design feature extraction on the basis of experience,this kind of method is easy to restricted to man-made structure characteristics of the power of expression.In this paper,based on the time delay problem of the algorithm and the characteristics of the classification algorithm expression ability is weak,clutter suppression based on the deep learning algorithm is proposed.With the characteristics of deep learning strong learning ability,learning more effective features from the raw data.First,this paper presents a multiple classification strategy based on trajectory,the clutter to the target point of binary classification problems translated into the target track multiple classification problems,improve accuracy and efficiency in the clutter suppression.Secondly,in terms of input characteristics,based on the deep learning method,including single cycle target and target trajectory for automatic feature extraction during the week.In radar signal processing is often face a problem that target and noise ratio imbalance seriously.Based on the neural network model,we proposed deep belief network model to solve the problem of clutter suppression,it is a generation model,through the study of the historical data to obtain estimates of probability density function of target and clutter,reduces the effects of sample proportion serious imbalance.Finally,we evaluated the method in a collection of radar datasets for three days,through the forward link path strategy,using the statistical features of target and clutter differences,spatial and temporal correlation and path continuity of target signal,realizes the real-time target discrimination and effectively improve the performance of clutter suppression,namely under the guarantee of the target detection probability,raised the probability of suppressing clutter.