Research On Passive Sonar Signal Recognition Of Underwater Moving Target

Underwater moving target recognition technology has been an important research content in underwater acoustic signal processing technology. The research of underwater moving object recognition technology has very important value on the national defense construction, meanwhile it also has very important application value in commercial and civilian areas. Therefore, it received widespread concern around the world, and it is a hot topic in the related fields. This paper focuses on the passive sonar signal recognition technology, the signal feature extraction and classifier design is the major job, and through the experimental simulation, the proposed classification method are verified.Feature extraction and selection is the first step in the process of target recognition, effective and reliable extraction of features is a prerequisite to ensure the accuracy of classification. At first, this paper deeply discusses the basic concept and principle of the wavelet transform and wavelet packet transform and introduce the wavelet packet analysis technology to extract the energy characteristics of underwater moving target passive sonar signals, then explored the principle of empirical mode decomposition algorithm,and bring the concept of mean absolute value in the underwater moving target passive sonar signals’s feature extraction. On the basis of the empirical mode decomposition,a method about calculating the intrinsic Mode Function’s mean absolute value which was obtained after decomposition as the signal’s feature is proposed.And from the perspective of data fusion, combining the features of fusion technology, this paper construct a new classification feature vector.Classifier design is the second part of target identification process, classifier design will affect the final quality of the recognition performance. This paper introduces the principle of FCM algorithm and neural network, analyzes the advantages and disadvantages of the FCM algorithm and generalized regression neural network (GRNN). Because of the the performance about FCM algorithm depends on the random initial clustering center, it is unsupervised algorithm, easy to fall into local extremum; and the function of hidden layer nodes of the GRNN is gauss function which owes a global approximation ability. According to these features, a algorithm of combination FCM and GRNN is proposed, and a new classifier is formed, the new classifier is used to realize the recognition of the feature vectors. The experimental results show that the target recognition correct rate is relatively high about the combination algorithm based on FCM and GRNN, thus the research purpose in this paper has been achieved.