Radar Signal Sorting And Target Recognition Based On Machine Learning

With the advancement of machine learning and radar technology,the application of machine learning in the field of radar has become more and more extensive and deeper.From radar scanning,signal acquisition and processing,one-dimensional range imaging,SAR image recognition,ISAR image recognition,radar tracking and guidance,and so on,are development with machine learning technology.Machine learning in the radar signal sorting technology mainly includes signal separation,determination of pulse parameters,the formation of a single radar pulse sequence,and then classify the radar target and the degree of threat.The denoising and radar target model identification is included in the one-dimensional range image recognition process.In the radar working mode,the current fiery deep learning can also be used for identification to do a corresponding defense and interference measure.This paper focuses on the clustering technology and target recognition technology in machine learning to meet the needs of China's electronic confrontation and other fields.A new nonlinear manifold algorithm is proposed for the radar signal sorting.The algorithm is based on the topological structure of nonlinear manifolds,transforms the signal from the time domain to the frequency domain,and clusters according to the geometric information between the data.Firstly,the local point density and cross-point distance are used to determine the set of sample points that are intersection points;then,the cosine angle information between the sample vector s of the sample points is used to linearly cluster the points near the intersection point;The points away from the sample points of intersection are clustered using the minimum distance method and merged with the points near the intersection point by category,and the radar signal sorting is finally completed.Then a radar target recognition method based on online adaptive dictionary learning is proposed for one-dimensional distance image recognition.Because the target aircraft continuously produces changes in attitude during the flight,the one-dimensional range image signals acquired in different time periods have different characteristics,and the256-dimension collected by the signal sample of the one-dimensional range image has only a few tens of dimensions as the target.Others dimensions reflected back are the noise signals.For noise problems and sparse signals,this paper turns this problem into a signal sparse problem,and online dictionary learning just solves the denoising problem.The sample reconstruction via iterative online dictionary learning and OMP algorithm solves the signal sparse problem.We first initialize samples with fixed dictionary and OMP algorithm,and then the first item in the dictionary learning formula l₂ paradigm changed into l₁ paradigm,to reduce the influence of divorce points to construct the dictionary.Finally,iteratively using OMP and dictionary learning method to find the final signal reconstruction representation and classify it with RankSvm.Finally,the characteristics of radar modulation and work pattern recognition based on quadratic classification are studied.Since the modulation method and the parameter range of the radiation source jointly determine the category to which the radiation source belongs,this paper first identifies the modulation method and then divides the different work modes under the same modulation method to determine the radar radiation source.Before recognition accuracy is not high by only using radar description words.This article proposes a method of the characteristics of radar modulation and radar parameters identifying together.First,extract the seven characteristics of the pulse modulation and improve one of the features.Using the fuzzy membership instead of the radar modulation feature as the input sample is to prevent the characteristics value to change to zero when normalization.Then,a deep belief network is used to identify the 21-dimensional fuzzy membership values.Finally,the radar parameters in the same modulation method are identified by using the data field method to obtain several working modes of the radar radiation source,and finally,the radar radiation source category is determined by the modulation method and radar parameters.