Underwater acoustic target radiation noise identification method based on domain adaptation

Abstract

The invention discloses an underwater acoustic target radiation noise identification method based on domain adaptation. The method comprises the steps: constructing a convolutional neural network shared by a source domain and a target domain; pre-training a convolutional neural network by using a source domain label sample; respectively calculating source domain classification loss and target domain pseudo-label classification loss by utilizing the source domain label sample and the target domain sample; calculating a cross-domain multi-core maximum mean difference distance between a source domain label sample and a target domain sample in a convolutional neural network in a multi-layer manner; obtaining the total loss of the network according to the source domain classification loss, thetarget domain pseudo label classification loss and the multi-core maximum mean value difference distance; assigning a target domain sample label predicted by the network to the label-free sample through pseudo label learning, so that the label-free target domain underwater acoustic target data has the capability of supervising the training model, the category characteristics of the source domain underwater acoustic target data and the target domain underwater acoustic target data are mapped to the same mark space, and the problem that the target domain underwater acoustic target sample is notprovided with a label and is accurately identified when the data volume is small is effectively solved.

Description

technical field [0001] The invention belongs to the field of detection and recognition of underwater acoustic targets, and in particular relates to a method for recognizing radiation noise of underwater acoustic targets based on domain adaptation. Background technique [0002] In recent years, the exploration of industrial science and technology in the marine field by countries around the world has continued to deepen, and the attention paid to underwater acoustic targets in the marine field has gradually increased, and the research on accurate and efficient identification methods has become more and more important. relatively backward. Therefore, with the significant changes in the international strategic situation and the surrounding security environment, the research on underwater acoustic target recognition needs to be promoted urgently. [0003] Document CN201910661350: "Intelligent Recognition Method of Underwater Acoustic Targets Based on Big Data" first collects a l...

AI Technical Summary

Problems solved by technology

However, this method first needs to manually extract features, and the degree of automation for underwater acoustic target recognition is not high, and end-to-end recognition cannot be achieved.

At the same time, when our target recognition field data is not exactly the same as the marginal distribution or conditional distribution previously used for model training data, this method tends to cause lower recognition accuracy problems

Due to the interference of complex ocean background noise and other reasons, the classification methods proposed by many scholars in recent years cannot achieve ideal classification performance.

Images