Research On Features Extraction And Recognition Based On Infrared Signatures Of Space Targets

Infrared features extraction and recognition design of space targets are the crucial tech-niques in the study of space targets recognition in the missile defense system(MDS).During terminal guidance,the infrared detectors are the main detecting means in tracing and recognizing space targets.Our paper aims to investigate the methods of extracting shape and micro-motion features from infrared signatures of space targets,serving as infrared recognition task in MDS.The main work of our paper focuses on the following four points:Firstly,the main infrared characteristics of space targets are analyzed.Emphasis is put on the analysis of micro-motion dynamics,thermal behaviors and geometry diff-erences of space targets,their influences to the dynamic law of infrared signals were in-vestigated,and the infrared signature model based on the radiation guides is established.Based on this,the measurement and discrimination power of main infrared features are discussed.The variable of target projection area is dependently extractable,which re-duces the extraction difficulty of shape and micro-motion features.Secondly,the requirements and methods of shape and micro-motion feature extraction are discussed.Related theory of determining and recovering shape in geometrics is introduced to explore the requirements of extracting space targets'features and to inves-tigate the methods of discrete describing and estimating shape parameters.Conclude that the symmetric space targets with half-convex surface,such as warheads and mimic decoys are feasible to be effectively estimated and recovered when the range of observation angles is beyond 0.5?for the infrared data.In discussing the joint para-meters estimation method of shape and micro-motion,the micro-motion makes the objective of joint estimation non-convex,which means there are several local minimums within the value intervals.To solve this problem,the grid method is put forward to search for the optimal solution and guarantee the reliability and accuracy of estimation results.Thirdly,the classifier of space targets based on infrared signatures is proposed.A classifier to distinguish axisymmetric and non-axisymmetric targets based on Random RNNs(R~2NNs)is present to finish the task of coarse classification of space targets recognition system,which project the history predictions into the current input space in a random weighting way,which enhances the memory ability of traditional algorithms and boost the generalization.By testing the public time series and infrared signatures,the algorithm is demonstrated that significant performance improvement in space targets classification without increasing the training complexity,compared with the traditional RNNs.Finally,put forward a joint estimation method for the shape and micro-motion parameters of space targets based on sparse decomposition representation of infrared signatures.According to the sparsity nature of shape description,the sparse decompo-sition model of infrared signatures of space targets is established,the estimation pro-blem is converted into the sparse problem.Based on the analysis of sparse problem,an iterative optimization method is proposed to jointly estimate target shape and micro-motion parameters.Experimental results with limited data demonstrate that only if the sparsity of targets'geometrical description is assumed to be reasonable,the micro-motion parameters can be effectively estimated using the proposed method.