Underwater acoustic target recognition method based on deep convolutional generative adversarial network

Abstract

The invention relates to an underwater acoustic target recognition method based on a deep convolution generative adversarial network, and belongs to the field of underwater acoustic target recognition. The method comprises the following steps: constructing a generation model and a discrimination model; normalizing the original underwater acoustic signals with the category information and the original underwater acoustic signals without the category information, and framing; setting generation model parameters; setting hyper-parameters; constructing a convolutional neural network in the discrimination model; taking the signal data as the input of a convolutional neural network, and calculating and outputting the signal data through the network to obtain a classification result; and countingclassification errors, returning the errors by using a BP algorithm, and updating the weight parameters of the network in the three steps, including the weight between the convolution kernel and thefull connection layer, until the iteration frequency is reached. The method has the advantages that the extracted features completely depend on the data, parameters related to the data do not need tobe set manually, the extracted features are effective to a certain extent for the data, and the data can be effectively utilized to mine distribution information existing in the data.

Description

[0001] Field [0002] The invention relates to an underwater acoustic target recognition method based on a deep convolution generation confrontation network, belonging to the field of underwater acoustic target recognition. Background technique [0003] Due to the development of technology and the needs of practical engineering applications, underwater acoustic target recognition faces two important problems: 1. Since manual calibration of a large number of acoustic target data is a costly task, generally only a few of the same samples can be Some of them are manually calibrated, so for the acoustic samples that have not been calibrated during the collection process, based on the consideration of making full use of the acoustic signal samples, it is necessary to propose an identification scheme for unknown targets; 2. The underwater acoustic targets are diverse, and for a certain target or It is difficult for the features extracted from the dataset to include all possible obje...

AI Technical Summary

Problems solved by technology

It can solve the disadvantages of cumbersome feature extraction procedures in existing methods, relying on human experience, and difficulty in making full and effective use of non-category information data, and can improve the accuracy of classification and recognition compared with the baseline model

Images