Acoustic Recognition Of Dolphin Species Based On SVM

Dolphins are marine mammals with a long history of evolution,their "language"system is well developed and they communicate with each other use whistles.At present,the classification and identification of dolphin species mainly rely on visual observation,but dolphins most of the time underwater,and the underwater environment is complex,relying on visual recognition dolphins have their limitations.Many dolphins can produce a unique whistle,and the whistles spread well in the water,so we can use the whistle to classify the dolphin species.Through the identification of dolphins,the relationship between the dolphin whistle and the species can be better established,which is of great significance to further study the dolomian semantics,individual identification and protection of dolphin species.For the above reasons,we have constructed a complete system for pretreatment,contour repair and feature extraction of dolphin whistles into classification and recognition.The whole system is divided into three parts:(1)In the pre-processing part of the whistle,the collected dolphin whistle is first converted to a spectrum by a short-time Fourier transform.There is a lot of background noise in the original whistle spectrum.According to the spectral characteristics of different noise,respectively its mechanical noise filtering,binarization,noise filtering outlier,vertical line noise filtering and denoising mode harmonic filtration.(2)In the process of pretreatment,the whistle contour may be damaged,resulting in the absence of the whistle profile,which affects the feature extraction of the dolphin whistle.Therefore,this paper uses the Kalman filter algorithm to repair the whistle contour and obtain a more complete Whistle profile.And then use the local maximum detector to extract the time-frequency parameter as the eigenvector for the reconstructed whistle profile.(3)In order to obtain higher recognition rate,the grid search method,the genetic algorithm and the particle swarm optimization algorithm are used to optimize the radial basis function parameters of the dolphin whistle feature space.After the combination of the radial basis function parameters,the support vector machine classifier based on directed acyclic graphs is constructed to classify the dolphin species and achieve high recognition rate.Using the linear discriminant analysis and the classification and regression tree algorithm in the literature to identify the experimental data,and the experimental results are compared with the proposed algorithm to verify the superiority of the algorithm in dealing with a large amount of noise data.Through the above three steps,we realized the pretreatment and repair of the dolphin whistle,made the whistle contour more complete,constructed the classifier to identify the dolphin species,and improved the recognition rate of the dolphin species.