Research On Noise Reduction Algorithm And Recognition Technology Of Ship Underwater Acoustic Radiation Signa

It is virtually significant to classify and recognize underwater acoustic targets for the ocean resource exploiting and the marine environment security.Effective target recognition of ships is extremely important for protecting the safety of China’s territorial waters.Currently,the recognition of ship radiation signals increasingly relies on front-end processing platforms,such as buoys,submarines,unmanned ships,and other means to survey the sound field environment at sea.Ship radiation signals are an important part of the marine acoustic environment,but the actual collected sound field environment not only contains various information about ship characteristics,but also contains a large amount of marine environmental noise.Therefore,it is necessary to denoise the ship radiation signals collected by the ocean front-end processing platform,extract feature quantities with classical features,and combine them to classify and recognize ship targets.This thesis proposes a denoising algorithm for ship radiation signal denoising,studies feature extraction of ship targets,and achieves effective classification and recognition of three types of ship targets.Firstly,ship target recognition requires obtaining high-quality raw data of radiation signals,so it is required to filter out the marine environmental background noise in the ship’s radiation signals.Based on Empirical Mode Decomposition(EMD)algorithm and Normalized Least Mean Square(NLMS)algorithm,a new noise reduction method is proposed for ship radiation signal noise reduction filtering.The noise reduction capability of this method is compared with the Least Mean Square(LMS)algorithm,NLMS algorithm and EMD-LMS algorithm in ship radiation signal.By analyzing the noise reduction of a plurality of measured data,the results indicate that EMD-NLMS is apparently superior to other three methods in eliminating ship radiation signal noise.Secondly,in ship target recognition,due to the high dimensionality and a large amount of feature data,directly using these data for model training can result in high computational complexity,long model training time,and poor real-time performance.Therefore,this thesis uses Mel-Frequency Cepstral Coefficients(MFCC)feature extraction to extract features from ship radiation signals and reduce the data dimension for model training in target recognition.Finally,in the classification and recognition of ship radiation signals,this paper optimizes BP neural network based on Ant Colony Optimization(ACO),and uses optimized BP neural network to design a classifier.Through simulation analysis,the convergence speed of ACO-BP algorithm based on ant colony algorithm and BP neural network is compared.The results show that the convergence speed of the former is significantly faster than that of the latter.Four kinds of noise reduction signals in the first part of the study are trained and recognized by ACO-BP algorithm.The results show that compared with the other three noise reduction algorithms,after noise reduction by EMD-NLMS algorithm,the recognition rate of classifier reaches 94%.This thesis proposes an EMD-NLMS based denoising algorithm for ship radiation signals,and extracts MFCC features from the denoised signals.Finally,the BP neural network optimized according to ant colony algorithm is applied to the identification of ship targets.Simulation experiments have shown that the ship radiation signal denoised by EMD-NLMS algorithm has a better target recognition rate,and has broad practical value in ship recognition.