Research On Intelligent Extraction Method Of Modulation Line Spectrum Features Of Underwater Acoustic Target

The feature extraction of underwater acoustic target radiation noise has always been a key topic in the field of underwater acoustic technology and an important prerequisite for underwater target recognition.Ship is a kind of underwater moving target.The modulation line spectrum of ship radiated noise contains a lot of effective information,which is an excellent characteristic of underwater acoustic target.Due to the complex and changeable marine environment and the rapid development of noise reduction technology,the signal-tonoise ratio of underwater acoustic signals collected by hydrophones is getting lower and lower.In this case,the characteristics of the modulation line spectrum of underwater acoustic target cannot be accurately extracted,which has a serious adverse effect on the subsequent underwater acoustic target detection and recognition tasks.To solve the above problems,this thesis combines deep learning theory and sparse representation theory,and focuses on the intelligent high-resolution extraction method of underwater acoustic target modulation line spectrum features.The main work and innovations of the thesis are as follows:A new DEMON spectrum estimation method based on convolutional network demodulation is proposed.This method uses convolution network to build filters to demodulate underwater acoustic target signals.Compared with the traditional DEMON spectrum estimation method,this method extracts clearer and more accurate modulation line spectrum features.This thesis proposes a high resolution extraction method of modulation line spectrum features based on sparse representation by analyzing the sparsity of modulation line spectrum features.A Cauchy-Gaussian model of the modulation line spectrum of underwater acoustic target is constructed by this method.And a better estimation result of modulation line spectrum is obtained by solving the model iteratively.Experimental results show that the modulation line spectrum extraction method based on sparse representation can extract the modulation line spectrum features with higher resolution and better accuracy from underwater acoustic target signals,and it is more universal under the high noise environment.A new modulation line spectrum high resolution extraction method based on soft threshold iterative network are proposed in this thesis.Combined with the DEMON spectrum estimation method based on convolution network demodulation,a complete high resolution extraction network of ship-radiated noise modulation line spectrum features is built.This network is constructed according to the soft threshold iteration formula,which has good interpretability.Compared with traditional methods,this network has faster speed and better generalization.Experimental results show that the modulation line spectrum extraction method based on deep learning can extract the target features with high resolution and improve the output accuracy,and also has clear and accurate modulation line spectrum extraction results under the high noise environment.This thesis aims to obtain clearer,more accurate and more stable modulation line spectrum features of underwater acoustic targets under the high noise environment,so as to lay a solid foundation for various underwater acoustic tasks such as target detection and recognition.