Research On Geometric Highlights Feature Extraction And Recognition Method Of Underwater Target

Feature extraction and recognition of active sonar target is one of the key technologies in active sonar detection.Through acquiring underwater target highlight characteristics,the detection and recognition target is the main applied technical method at present.The highlight can effectively reflect the physical information such as shape and material of the target.According to the mechanism of highlight formation,it is divided into geometrical highlights including mirror reflection of convex smooth surface and edge prism and corner reflection,as well as elastic highlights corresponding to surface traveling wave or elastic scattering wave under specific conditions.Based on the geometrical highlight characteristics of underwater targets,this paper mainly carries out research on extraction methods and recognition of geometric highlights of targets.The research work of this paper is as follows:(1)Because there is echo mixing and reverberation interference between highlight echoes,the recognition effect of extracted features for target recognition is not good.To solve this problem,two deconvolution methods are proposed.The time-frequency transformation results are deconvoluted and time-frequency energy filtering is used to suppress reverberation interference.Finally,the highlight feature of target echo is extracted by Radon transform.The simulation and experiment results show that deconvolution algorithm combined with timefrequency energy filter can weaken the effect of highlight aliasing and suppress reverberation interference,which provides a new technical method for target recognition and classification in time-frequency domain to acquire the highlight feature of echo.(2)The back-projection algorithm is used to obtain the multi-angle target echo intensity plane mapping features of the underwater target,and the multi-azimuth target time series echo signals are projected to two-dimensional spatial coordinates.Feature mapping method(Isometric Feature Mapping,Isomap),kernel function principal component analysis method(Kernel PCA,KPCA)and kernel function principal component analysis method optimized by the cuckoo search algorithm extract the feature vector of the target,and finally extract the main The features are input into the support vector machine optimized by the cuckoo search algorithm to classify and identify the underwater target.Through the simulation and experimental data,the feasibility of the back-projection algorithm combined with the four feature extraction algorithms to classify and identify underwater targets is verified,and the classification and identification of the features extracted by various feature extraction methods in the classifier are compared.The recognition rates corresponding to each feature extraction method are,from high to low,the kernel function principal component analysis method optimized by the cuckoo search algorithm,the kernel function principal component analysis method,the principal component analysis method and the isometric feature mapping method.A posteriori weighting method is proposed to weight the extracted feature vectors,and it is verified that the method has the ability to improve the performance of target recognition,and provides a new technical method for target highlight feature extraction and recognition and classification.