Deep Learning And Practice Of Slow Moving Target Detection Under Clutter Environment

In recent years,the technology of Unmanned Aerial Vehicle(UAV)and intelligent vehicle have developed rapidly,but it also brings about the problem that "black flight of UAV" or pedestrian detection is difficult.Therefore,the effective detection of slow moving target such as UAV and pedestrian become increasingly important.However,when target detection is carried out in such the environment,the moving target is often affected by the strong background clutter.And at the same time,the target has the characteristic of slow moving speed and small Radar Cross Section(RCS),resulting in the weak echo energy and easy to be "submerged" by strong ground clutter,which makes it difficult to detect the target.Therefore,this paper focuses on the problem of weak slow moving target detection in the environment of strong clutter.The target detection method based on deep learning is mainly studied,and the algorithm is verified by simulation experiment and measured data.The main research contents is shown below:1.The state of art is studied on slowing-moving weak target detection technology and deep learning method.The statistical characteristic and simulation method of Rayleigh clutter are introduced.The principle of traditional detection method in clutter environment is introduced,and the Moving Target Indicator(MTI)and Moving Target Detection(MTD)algorithm,as well as the subspace-based method is analyzed.2.Common deep learning framework is expounded and the structure of convolutional neural network is specifically analyzed,and a multi-dimensional feature fusion target detection method of convolutional neural network is proposed in this paper.Firstly,the echo signal is pre-processed to analytic the echo signal from multiple domains and select its feature of different dimensions.Then,the echo signal’s feature information on multiple domains is further extracted by combining with the convolutional neural network,and the feature fusion is carried out,and finally realize the effective recognition and detection of the target.The experimental results show that this method can detect the slow moving target in the environment of strong clutter with good accuracy and improve the detection performance.3.To solve the problem on slow moving target detection in the background of strong clutter,the target detection method of deep neural network is designed based on the theory of deep learning.A slow moving target detection method based on Deep Convolutional Neural Networks(DCNN)is designed.The Range-Doppler spectrum of the echo signal is taken as the input and sent into the deep convolutional neural network.And the residual spectrum of the echo signal can be obtained by learning the clutter feature in the echo signal and removing the target components implicitly.Then,the background cancellation is carried out with the residual spectrum to suppress the clutter,so as to detect the moving target.In addition,a target detection method of Convolutional Auto-Encode(CAE)is designed.The target signal components and clutter components in the Range-Doppler spectrum are learned by the encoder,and the hidden features of the moving target signal are extracted.Then the de-convolution layer is used as the decoder to reconstruct and recover the target signal.The network adopts the double-channel structure to extract amplitude features and phase features of target echo,and the skip connection structure is designed in the network,which connects the top and the bottom of the network to improve the recovery ability of target feature in decoder.Moreover,it mitigates the vanishing gradient problem from the deep network,improving the efficiency in end-to-end training.The experimental results show that the method based on deep neural network is effective and can achieve better detection performance than the traditional method.