Research On Fish Identification Based On Wavelet Packet Transform

Fish stocks are an important part of marine resources,so it is necessary to develop and protect them reasonably and effectively.Consequently,the ability of fast fish identification is needed urgently.In this paper,a series of studies on fish identification by active acoustic methods are carried out in order to get more stable and efficient fish identification methods.In this paper,based on the extensive analysis of relevant literature,some discriminating features by wavelet packet transform and the corresponding comprehensive methods are proposed to identify different fish.The echo signals of three kinds of common fish are obtained by immobilized fish experiment,and the recognition effect of each characteristic quantity is analyzed and compared.Then,in order to approximate the motion state of the fish in real environment,a free swimming fish experiment is carried out.The species classification,size and motion status of the fish are identified and classified respectively,which lay a foundation for fish identification in real environment.The main results are as follows:(1)In the fixed fish recognition,the wavelet packet band energy and the information entropy of wavelet packet are more effective;(2)Extracting the low-frequency characteristic quantity of the echo signal of the fixed fish and the feature selection of the 5 wavelet packet features are helpful to improve the recognition rate;(3)The singular value of the wavelet packet coefficient matrix and the energy entropy of wavelet packet are used to identify three kinds of free swimming fish;(4)The information entropy of wavelet packet can get better recognition results in the size and motion posture recognition of free swimming fish.(5)The recognition results of SVM are better than those of BP neural network for both fixed and free swimming fish identification.