Research On Underwater Target Recognition Based On Hilbert-Huang Transform And Extreme Learning Machine

The underwater target recognition technology has always been a hot focus in the research field of underwater acoustic signal analysis. The research of this problem is valuable in both military field and civilian field. Pattern recognition technology contains the underwater target recognition, which has developed according to the feature extraction technology and the design of classifier. The main task of feature extraction is how to extract from the signal and select the effective characteristics represent the right category; Classifier design is to study a variety of classification algorithms, using the feature vectors obtained in the extraction process trained to recognition.This paper research the feature extraction and the classifier design for ship radiated noise signal, which focused on Hilbert-Huang transform theory and made a thorough study of feature extraction ship noise signals. Classifier design focused on extreme machine learning algorithm proposed in recent years and optimized its network structure based on the theory of compressed sensing. The main contents are as follows:1. This paper introduced the basic theory about the problem of underwater target recognition, and introduced the role of each part consisting of an underwater target recognition system. Highlighting the performance of a variety of feature extraction algorithms to identify their advantages and disadvantages, as well as existing classification algorithms.2. For outstanding ability of Hilbert-Huang Transform (HHT) in dealing with nonlinear and non-stationary signals, we applied it to underwater target recognition. Meanwhile, this part focused on the Hilbert-Huang main ideas and transform algorithm. By comparing, using a more favorable water acoustic signal feature extraction algorithm to extract the instantaneous frequency characteristics of the signal, marginal spectrum characteristic features such as classification features.3. Introduce the main ideas and theories of extreme learning machine, especially the machine learning algorithms. Due to extreme machine learning fast train speed may be caused huge network structure, in order to solve the problem, this part used compressed sensing techniques to optimize and simplify the network structure without reducing the classification accuracy of the premise.4. Introduce the ideas and theories of compressed sensing algorithm, and used it to optimize the weights and nodes of the hidden layer of extreme learning neural network.Design experiments to verify the extent of compressed sensing algorithm to optimize the network structure and use the optimized network to classify the ship noise signals, to obtain better classification results.In this paper, the noise signal of the ship as an object, research goals is to enhance the performance of target identification based on passive sonar. The results are expected to be applied to the field of underwater target recognition,marine areas and noise signal analysis algorithms for the improvement of extreme learning machine.