Study On Feature Extraction And Classification Methods Of Fish Species Echoes

Scientific assessnent of fishery resources is of sreat importance for the sustainable develop-ment of fisheries.Compared with biological sampling,the acoustic evaluation of fishery resources has the advantages of quickness,large sampling rate and no interference with the natural state of the group.There are many kinds of signal feature extraction and classification methods,and their performance varies with the application environment.In order to comprehensively evaluate the performance of related methods in feature extraction and classification of fish species echoes,this paper fuses the Doppler information of fish school echoes to improve the existing classical fish school echoes simulation model based on the possible relative motion characteristics of fish and ship-borne sonar in practice.The improved model is more in line with the actual echo data,which provides a simulation basis for subsequent analysis and evaluation of the performance of relevant methods.The fish species classification framework consists of two basic components:feature extraction from observational data and classification of features based on classification methods.For the feature extraction methods of fish species echoes,because of the complexity and variability of the internal structure of fish body and the aggregation pattern of fish species,it is usually difficult to extract useful information hidden deeply in the fish species echoes,which results in fewer features and lower species resolution.On the basis of comprehensive analysis of common methods for extracting echo features,convolution neural network is exploited to extract the features of fish species echoes.This method can extract more complex and deep-seated features hidden in the fish species echoes through convolution cascade network structure.For the classification methods of fish species echoes,the current research work seldom involves the applicability analysis of specific classification methods for specific features,and the evaluation of performance improvement effect brought by the fusion of different features and different learning methods.This paper applies the classical machine learning classification methods to fish species echoes classification.On this basis,the multi-model Stacking integrated learning classification method is improved according to the features of fish species echoes,and the performance of fish species classification is improved.Finally,in order to further verify the validity of the feature extraction and classification meth-ods of fish species echoes,the tank experiment is earied out.Through the analysis of echo data of three typical fish species,the experimental results are in good agreement with the simulation.