Research On Key Techniques Of Multibeam Fish-finding Sonar

With the enormous potential of marine resources in the economy continue to be attached great importance by peoples,marine fishery resources have been developed as exploited and renewable valuable resources for peoples.With activity of human is extended to the depths of oceans and the rapid development of science and technology,the pace of development and utilization of marine fishery resources are also rapid.The mainland coastline length of china is more than 18,000 kilometers and the marine area is extensive;inland lakes and rivers are huge,and fishery resources are rich.How to use acoustic measurement data of multibeam fish-finding sonar effectively and obtain fishery resources information reliably,which are hot and difficult issues are concerned widely by peoples.The thesis combines research status and development trend of fish-finding technology and fish-finding sonar in both home and abroad,and focuses on the four main aspects mainly in research work,included acoustic scattering modelling method of single fish and fish school,fish school detection technology of multi beam fish-finding sonar,feature extraction method of multi-source information and therealization of classification algorithm.The main part of the thesis can be divided into following:Firstly,acoustic scattering modeling method of single fish and fish school with swim bladder have been studied.The target strength of fish is a key parameter for converting estimating the integrated value of echoes to the absolute number of fish,and the model estimation method is one of the effective way to estimate the target strength.Firstly,in order to solve the problem that Kirchhoff model can't modeling correctly at large tilt angles of fish,the method for modelling the target strength of fish with swim bladder is proposed based on two-dimensional conformal mapping method,the target strength estimated by the model are equal to measured results almost,and stability is very high,the existed problem in KRM is resolved effectively.The selection of fish samples is random,which indicates that the model is not specific to particular fishes,and can be applied to modeling acoustic scattering for all fishes with swim bladder.Finally,the traditional fish school acoustic scattering model is extended,fish school scattering model contained Doppler frequency drift is established,which combines fish target strength estimated by FMM model,equivalent directional parameter of transducer,waveform of echo and beam scattering model.The narrowband and broadband acoustic scattering signal are simulated,which show that randomness of echo is lower and distribution of envelope is more stable of broadband acoustic scattering signal due to big time bandwidth product and good space resolution of broadband coding signal.Secondly,fish school detection technology using multibeam fish-finding sonar has been studied.Firstly,against to the problem of traditional Janus configuration phased array can transmit four peripheral beams only,the five-beam fish-finding sonar phased method is proposed,construction of phased method is simple and implement is easy,the correction of proposed method is proved through testing directivity of actual phased array.Secondly,in order to solve the problems of traditional echo integration,a method for obtaining information of fish school based on the idea of system identification is proposed,in which fish school is regarded as filtering system in acoustic field,the depth,size,abundance and other information of fish school are obtained through estimate impulse response of fish school.Finally,in order to solve the problem of migrated fish school swimming speed detection,the characteristics and shortcomings of incoherent and coherent method are analyzed and the basic principles of broadband coded pulse pairs method is studied,low auto-correlation side lobes binary coding sequence is introduced into the velocity measurement of fish school at first time,and broadband low auto-correlation side lobes binary coded pulse pairs of fish school swimming velocity detection method is proposed,simulation results show that the proposed method can measure velocity of swimming fish school effectively.Then,the acoustic scattering multi-source information feature extraction method and their capabilities of classification have been studied.Features extraction using wavelet packages transform,centroid,discrete cosine transform are studied,and wavelet packages coefficient singular value features,wavelet packages subband energy features,temporal centroid features,spectral centroid features and discrete cosine transform coefficient features are extracted,and classification capabilities of which are evaluated through Fisher discriminant method.The above features extracted through five kinds of features extraction method are composed five groups' features vector,and classification capabilities of each feature were compared using SOM neural network classifier.The data processing results shows that discrete cosine transform coefficient features has highest classification accuracy,so it can characterize the differences between different fishes.Then,partial features are combined to compare and analyze capabilities of classification,combined features may cause appearing redundant features,and some features extracted through different method are complementary.Finally,classification capabilities of features extracted using acoustic scattering signal of fish with different bandwidth are analyzed,and the results showed that if the signal bandwidth greater,extracted features can characterize acoustic scattering characterization of fish more fully,which is beneficial to study fish classification.Finally,fish classification method based on multi-azimuth acoustic scattering data fusion using composite kernel SVM has been studied.To maximize advantages of many features extraction methods in fish classification and improve capabilities of classification,based on analyzing the traditional single kernel SVM classifier principle theoretically,multi-azimuth acoustic scattering data fusion fish classification method using composite kernel SVM is proposed.The composite kernel is used to combine the different feature vector information instead of single kernel,the SVM is used to achieve fish classification based on multi-azimuth acoustic scattering data and classifying fish based on multi-azimuth acoustic scattering data with composite kernel SVM.This thesis discusses the optimal parameters search method in classification algorithms based on cross-validation,and on this basis,proposed classification method is verified combining experimental data.The results show that the fish classification used by composite kernel SVM can get higher classification accuracy than traditional single kernel form.The test results also indicate that if different feature vectors are directly combined a vector for classification,its classification accuracy does not necessarily higher than the classification accuracy obtained by either one kind of feature vector information,andthe composite kernel SVM treatment can effectively solve this problem;Fish classification method based on multi-azimuth acoustic scattering data fusion using composite kernel SVM can get higher classification accuracy than using a single azimuth acoustic scattering data,which verify the effectiveness of fish classification method based on multi-azimuth acoustic scattering data fusion using composite kernel SVM.