Underwater acoustic data generation method based on generative adversarial network

Abstract

The invention belongs to the technical field of underwater acoustic target classification, and discloses an underwater acoustic data generation method based on a generative adversarial network, and the method comprises the following steps: constructing a self-adaptive stable deep convolution generative adversarial network model based on a generative adversarial network model in combination with a self-adaptive controller; adjusting training steps of a generator and a discriminator based on an adaptive controller, and accelerating a convergence process of the network; in the training process, a progressive learning strategy is adopted, learning is carried out step by step from low frequency to high frequency, the generation process is stabilized, and the sample generation quality is improved. According to the underwater acoustic data generation method based on the generative adversarial network, the adaptive stable deep convolution generative adversarial network model is constructed to generate the synthetic sample of the underwater acoustic target, so that the problem of scarcity of existing data is solved, and the network convergence speed is improved; and the classification precision of the underwater acoustic target is improved by generating a high-quality Mel-frequency cepstrum coefficient.

Description

Technical Field

[0001] The present invention relates to the technical field of underwater acoustic target classification, and in particular to a method for generating underwater acoustic data based on a generative adversarial network. Background Art

[0002] In the field of underwater acoustic target classification (UATC), commonly used methods include those based on traditional signal processing, such as wavelet transform, low frequency analysis and recording (LOFAR), and envelope modulated noise detection (DEMON). These methods can achieve good results under specific conditions, but they perform poorly in complex noise backgrounds and changing acoustic environments.

[0003] In recent years, deep learning methods have also been introduced, such as convolutional neural networks (CNNs) for underwater target classification. However, deep learning methods rely on a large amount of labeled data, especially high-quality labeled data, and the problem of data scarcity makes the model prone to overfitting. Generative adversarial networks (GANs) are used to generate underwater acoustic data, but the training process is very unstable and converges slowly, which limits its widespread application in underwater acoustic target classification. The complex underwater environment produces a large amount of sample noise, which further increases the difficulty of GAN training. Moreover, existing methods usually train directly on complex data and fail to effectively utilize the strategy of gradual learning from low frequency to high frequency, resulting in suboptimal quality of the generated results.

[0004] Therefore, there is an urgent need for an underwater acoustic data generation method based on a generative adversarial network to overcome the above technical defects. Summary of the invention

[0005] The purpose of the present invention is to provide an underwater acoustic data generation method based on a generative adversarial network, which generates synthetic samples of underwater acoustic targets by constructing an adaptive stable deep convolutional generative adversarial network model to overcome the problem of scarcity of existing data and improve the network convergence speed; and by generating high-quality Mel-frequency cepstral coefficients (MFCCs) to improve the classification accuracy of underwater acoustic targets.

[0006] To achieve the above object, the present invention provides a method for generating underwater acoustic data based on a generative adversarial network, comprising the following steps: Step S1: Based on the generative adversarial network model and combined with an adaptive controller, an adaptive stable deep convolutional generative adversarial network model is constructed; Step S2: Based on the adaptive controller, the training steps of the generator and the discriminator are adjusted to accelerate the convergence process of the network; Step S3: During the training process, a progressive learning strategy is adopted to gradually learn from low frequency to high frequency, stabilize the generation process, and improve the quality of sample generation.

[0007] Preferably, in step S1, based on the generative adversarial network model, an adaptive stable deep convolutional generative adversarial network model is constructed in combination with an adaptive controller; Among them, the adaptive stable deep convolutional generative adversarial network model includes: an adaptive controller and a progressive learning strategy.

[0008] Preferably, in step S2, based on the adaptive controller, the training steps of the generator and the discriminator are adjusted to accelerate the convergence process of the network, and the working process is as follows: Step S21, initializing the scale parameters of the generator G and the discriminator D; Step S22: During the training process, generate a random probability P; Step S23, if P meets the controller setting, when P tends to the generator, the generator network is trained; otherwise, the discriminator network is trained; Step S24: If P does not meet the controller setting, adjust the proportional parameters of G and D according to the controller; Step S25, compare the adjusted ratio with P to determine the probability of training G or D in the next iteration; Step S26, loop the above steps until the end.

[0009] Preferably, in step S3, during the training process, a progressive learning strategy is adopted to gradually learn from low frequency to high frequency, stabilize the generation process, and improve the sample generation quality. The specific implementation steps are as follows: Step S31, initializing the fuzzy processing of Mel-frequency cepstral coefficients MFCCs; Step S32, during the training process, blurring the data set and the generated MFCCs; Step S33: if in the early stage of training, allow the generator to learn without giving priority to details, and train the generator network; Step S34: if in the later stage of training, gradually reduce the blur level so that the generator focuses on generating clearer MFCCs; Step S35, loop the above steps until the end.

[0010] Therefore, the present invention adopts the above-mentioned underwater acoustic data generation method based on generative adversarial network, and the beneficial effects are as follows: 1) The present invention improves the convergence speed of the network: Compared with the traditional generative adversarial network (GAN) model, the AS-DCGAN model using an adaptive controller can significantly accelerate the convergence speed of the network and reduce the training time by about 40%; 2) The samples generated by the present invention are of higher quality: the progressive learning strategy enables the model to gradually learn from low-frequency to high-frequency features, thereby generating high-quality Mel-frequency cepstral coefficients (MFCCs). These synthetic samples can significantly improve the classification accuracy of underwater acoustic targets; 3) The data enhancement effect of the present invention is obvious: on two public datasets (DeepShip and ShipsEar), the classification accuracy is improved to 82.74% and 86.41% respectively by using synthetic samples for data enhancement, which is significantly better than the classification accuracy of the original datasets.

[0011] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is an adaptive stable deep convolutional generative adversarial network (AS-DCGAN) model framework of an underwater acoustic data generation method based on a generative adversarial network in the present invention. DETAILED DESCRIPTION

[0013] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.

[0014] The present invention provides a method for generating underwater acoustic data based on a generative adversarial network, comprising the following steps: Step S1: Based on the generative adversarial network model and combined with an adaptive controller, an adaptive stable deep convolutional generative adversarial network model is constructed.

[0015] like Figure 1 As shown, the present invention constructs an adaptive stable deep convolutional generative adversarial network (AS-DCGAN) model based on a generative adversarial network model to generate synthetic samples of underwater acoustic targets to overcome the problem of scarce existing data.

[0016] Among them, the adaptive stable deep convolutional generative adversarial network model includes two modules: adaptive controller (AC) and progressive learning (PL) strategy. The adaptive controller (AC) adjusts the training of the generator and discriminator in the adversarial network, changes the training ratio to achieve network convergence and accelerates the convergence speed; the progressive learning (PL) strategy is introduced at the generator end of the adversarial network to gradually stabilize the process of generating samples, and the steps of introducing low-frequency feature fuzzy processing to gradually weaken the fuzzy effect to ensure high-frequency features can effectively improve the quality of generated samples, and finally achieve fast, efficient and high-quality generation of underwater acoustic data.

[0017] Step S2: Based on the adaptive controller (AC), the training steps of the generator and the discriminator are adjusted to accelerate the convergence process of the network. The workflow is as follows: Step S21, initialize the scale parameters of the generator (G) and the discriminator (D).

[0018] Step S22: During the training process, a random probability P is generated.

[0019] Step S23: If P meets the controller setting and P tends to be the generator, the generator network is trained; otherwise, the discriminator network is trained.

[0020] Step S24: If P does not meet the controller setting, the proportional parameters of G and D are adjusted according to the controller.

[0021] Step S25: Compare the adjusted ratio with P to determine the probability of training G or D in the next iteration.

[0022] Step S26, loop the above steps until the end.

[0023] The above process dynamically adjusts the training ratio through an adaptive controller, effectively reducing the training time by about 40%.

[0024] Step S3: During the training process, a progressive learning (PL) strategy is adopted to gradually learn from low frequency to high frequency, stabilize the generation process, and improve the quality of sample generation. The specific implementation steps are as follows: Step S31, initializing the fuzzy processing of Mel-frequency cepstral coefficients (MFCCs).

[0025] Step S32: During the training process, the data set and the generated MFCCs are blurred.

[0026] Step S33: If in the early stages of training, allow the generator to learn without prioritizing details and train the generator network.

[0027] Step S34: If it is in the later stage of training, gradually reduce the blur level so that the generator focuses on generating clearer MFCCs.

[0028] Step S35, loop the above steps until the end.

[0029] Example This embodiment performs a performance test on a method for generating underwater acoustic data based on a generative adversarial network proposed in the present invention.

[0030] Based on the constructed adaptive stable deep convolutional generative adversarial network (AS-DCGAN) model, high-quality spectrograms can be generated for UATC. In the experimental comparison of the AS-DCGAN model by controlling the ratio of original data and generated data, the recognition accuracy of the model increases with the increase of the proportion of generated images in the training set regardless of the proportion of original data. When the proportion of generated images reaches 80%, the recognition accuracy of the model reaches 84.53% in ShipsEar and 80.49% in DeepShip, and finally reaches 86.41% in ShipsEar and 82.74% in DeepShip as the proportion of generated images increases. This result exceeds the recognition accuracy of the model when only original images are used as input data, which are 80.48% in ShipsEar and 78.38% in DeepShip respectively. It shows that by using high-quality generated images to enhance data, the proposed model can effectively improve the recognition accuracy of underwater acoustic targets. It is worth noting that when only generated images are used for training, the model still achieves 60.74% and 57.55% recognition accuracy in ShipsEar and DeepShip, respectively. This result shows that the generated images are of high quality and can be used to train the classification network, and the recognition accuracy of the model will gradually increase as the proportion of generated images gradually increases.

[0031] Therefore, the present invention adopts the above-mentioned underwater acoustic data generation method based on generative adversarial network, and constructs an adaptive stable deep convolutional generative adversarial network model to generate synthetic samples of underwater acoustic targets, so as to overcome the problem of scarcity of existing data, significantly accelerate the convergence speed of the network, and reduce the training time by about 40%; the progressive learning strategy enables the model to gradually learn from low-frequency to high-frequency features, thereby generating high-quality Mel-frequency cepstral coefficients (MFCCs), which significantly improves the classification accuracy of underwater acoustic targets; on two public datasets (DeepShip and ShipsEar), data enhancement is performed through synthetic samples, and the classification accuracy is improved to 82.74% and 86.41% respectively, which is significantly better than the classification accuracy of the original dataset.

[0032] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for generating underwater acoustic data based on a generative adversarial network, characterized in that: The following steps are involved: Step S1: Based on the generative adversarial network model and combined with an adaptive controller, an adaptive stable deep convolutional generative adversarial network model is constructed; Step S2: Based on the adaptive controller, the training steps of the generator and the discriminator are adjusted to accelerate the convergence process of the network; Step S3: During the training process, a progressive learning strategy is adopted to gradually learn from low frequency to high frequency, stabilize the generation process, and improve the quality of sample generation.

2. The method for generating underwater acoustic data based on a generative adversarial network according to claim 1, characterized in that: In step S1, based on the generative adversarial network model and combined with an adaptive controller, an adaptive stable deep convolutional generative adversarial network model is constructed; Among them, the adaptive stable deep convolutional generative adversarial network model includes: an adaptive controller and a progressive learning strategy.

3. The method for generating underwater acoustic data based on a generative adversarial network according to claim 1, characterized in that: In step S2, based on the adaptive controller, the training steps of the generator and the discriminator are adjusted to accelerate the convergence process of the network. The workflow is as follows: Step S21, initializing the scale parameters of the generator G and the discriminator D; Step S22: During the training process, generate a random probability P; Step S23: If P meets the controller setting, when P tends to the generator, the generator network is trained; Otherwise, train the discriminator network; Step S24: If P does not meet the controller setting, adjust the proportional parameters of G and D according to the controller; Step S25, compare the adjusted ratio with P to determine the probability of training G or D in the next iteration; Step S26, loop the above steps until the end.

4. The method for generating underwater acoustic data based on a generative adversarial network according to claim 1, characterized in that: In step S3, during the training process, a progressive learning strategy is adopted to gradually learn from low frequency to high frequency, stabilize the generation process, and improve the quality of sample generation. The specific implementation steps are as follows: Step S31, initializing the fuzzy processing of Mel-frequency cepstral coefficients MFCCs; Step S32, during the training process, blurring the data set and the generated MFCCs; Step S33: if in the early stage of training, allow the generator to learn without giving priority to details, and train the generator network; Step S34: if in the later stage of training, gradually reduce the blur level so that the generator focuses on generating clearer MFCCs; Step S35, loop the above steps until the end.

Images