Research On Radar Clutter Suppression Method Based On Machine Learning

Nowadays,as the requirements of radar systems for detecting moving targets continue to increase,it is particularly important to suppress the clutter in the echo received by the radar.However,traditional clutter suppression technology mostly distinguish and suppress clutter and target with different Doppler frequency characteristics,which leads to more residual clutter points in the suppression result.At the same time,in a strong clutter environment,the target is lost Therefore,a new method of clutter suppression is urgently needed.Machine learning has made rapid progress in the last ten years,and it has good application prospects in autonomous driving,aerospace and industrial fields.This technology can learn the characteristics of data adaptively according to the algorithm,without the need for human adjustment according to the environment.In recent years,the application of machine learning to radar signal processing has gradually become popular.This paper will combine machine learning methods to propose two clutter suppression methods based on machine learning.The first is the clutter suppression method based on the probability undirected graph model.This method mainly takes each data point in the Doppler-distance spectrum data stream of the echo as a node in the undirected graph,and analyzed the correlation between the target and the clutter in time and space.Clutter has randomness and isolation in time and space,while the target shows a trajectory in speed and distance.Therefore,based on this prior knowledge,a probabilistic undirected graph model is established,and the largest clique and corresponding energy function are designed.,And finally use the ICM algorithm to solve the probabilistic undirected graph model;On the other hand,considering that the target exhibits a sparsity in space on the range-Doppler spectrum,and the clutter matrix exhibits a low-rank property,the joint sparse low-rank constraint model is added,and an improved probability undirected graph model is proposed,Use RPCA+ICM algorithm to solve.The two methods before and after the improvement have good suppression effects,but the model has the problems that the weight of the undirected graph cannot be adaptive,and the model parameters need to be adjusted manually.The second is the clutter suppression method based on graph convolutional neural network.This method is aimed at the problems in the probabilistic undirected graph model.Considering that in the graph structure,the weight of the target node and the neighboring node has a certain relationship,the construction of the undirected graph is handed over to the graph convolutional neural network for learning.A deep learning framework,including feature extraction network,adjacency matrix construction network and GCN graph convolution network,combined with data to train the network,extract features,so as to achieve the suppression of the clutter in the radar received echo.The clutter suppression method has a good suppression effect,and at the same time,it adaptively learns the weights between nodes in the undirected graph according to the network,and has good robustness.