Generative Adversarial Networks (GANs) have emerged as a promising approach to address the data scarcity issue in intelligent underwater object recognition. However, existing advanced GANs still suffer from the challenge of discriminator overfitting when constrained by limited sonar data. To address this challenge, this work proposes a Dual Augmentation method for GANs (DA-GAN) to stabilize training under limited data, thereby synthesizing high-fidelity and multi-view sonar images. DA-GAN incorporates two core modules: Adaptive Adversarial Example Augmentation (AAEA) and Random Differentiable Augmentation (RDA). The AAEA module generates adversarial examples by applying imperceptible perturbations to generated sonar images using Fast Gradient Sign Method (FGSM), Basic Iterative Method (BIM), and Projected Gradient Descent attack (PGD), effectively deceiving the discriminator. Adversarial examples strategically replace real samples according to the degree of discriminator overfitting. Concurrently, the RDA module applies random differentiable augmentations to both real and fake sonar samples, further mitigating the overfitting. Furthermore, leveraging the local smoothness of the latent space, we design three methods—nonlinear perturbation, sparse sign perturbation, and style-mixing to perturb the latent codes, generating multi-view sonar images. Experiment results demonstrate that DA-GAN effectively mitigates discriminator overfitting, generating sonar images with superior quality compared to StyleGAN2 on the SCTD dataset, with the FID improvement from 175.341 to 87.187. On the KLSG dataset, DA-GAN achieves an FID score of 58.067. Furthermore, after augmenting the SCTD dataset, the classification model (e.g., ResNet-34) achieves a 12.50% improvement in global accuracy, while the detection model (e.g., YOLOv5s) shows a 6.93% increase in mAP0.5:0.95.

Abstract Side‐scan sonar (SSS) detection is a key method in underwater environmental security and subsea resource development. However, many detection approaches primarily concentrate on tracking the evolution path of optical image object detection tasks when using acoustic images, resulting in complex structures and limited versatility. To tackle this issue, we introduce a pioneering dual‐domain multi‐frequency network (D 2 MFNet) meticulously crafted to harness the distinct characteristics of SSS image detection. In D 2 MFNet, a novel method for optimizing and improving the detection sensitivity in different frequency ranges called multi‐frequency combined attention mechanism (MFCAM) is proposed. This mechanism amplifies the relevance of dual‐domain features across different channels and spaces. Moreover, we introduce a dual‐domain feature pyramid network (D 2 FPN) significantly augments the depth and breadth of feature information in underwater small datasets. The methods offer plug‐and‐play functionality with substantial performance enhancements. Extensive experiments are conducted to validate the efficacy of the proposed techniques, and the results showcase their state‐of‐the‐art performance. MFCAM improves the mAP by 16.9% in the KLSG dataset and 15.5% in the SCTD dataset. The mAP of D 2 FPN was improved by 8.4% in the KLSG dataset and by 9.8% in the SCTD dataset. The code and models will be publicly available at https://dagshub.com/estrellaww00/D2MFNet .

Sub-bottom profiler (SBP) Sonar data are always polluted by noise that leads to the wrong interpretation. To deal with this situation, a method combining the prior information of SBP sonar data and the variational mode decomposition (VMD) framework is proposed attempting to recover clean data from its corrupted version. First, to deal with the unknown mode number of VMD method, a mode number determination criterion which takes the frequency distribution characteristic into consideration has been proposed. Then, the sea bottom line is extracted using the enhancement filtering and the threshold extraction method. The echoes above the sea bottom are from the water column and mainly contain noise components. To determine which modes are caused by the noise, we obtained and analyzed these echoes to achieve the noise prior frequency feature. After that, the noise modes can be determined and removed using its prior frequency information. Both simulated and real experiments have been used to test the effectiveness of the proposed method, and good performance has been achieved.

The acquirement of seafloor small target information is one of the most important tasks of a side-scan sonar (SSS) survey. Thus, SSS small target detection becomes fundamental work for SSS applications which holds vital importance for marine engineering, maritime military and so on. However, existing methods cannot take the prior shadow information into consideration well, which would easily miss small targets. In this paper, a novel SSS small target detection method considering shadow characteristics is proposed. First, we give a detailed analysis of the SSS imaging theory as well as the prior information about the characteristics of shadows. Then, considering the prior information of the SSS short-shadow, the second partial derivative of the Gaussian function is specifically introduced for the construction of a weighted item. After that, incorporating the weighted item with thel21-norm,l1-norm, and low-rank constraints on the noise, the target, as well as the background, respectively, a weighted sparse detection model is proposed. To further take the long-shadows into consideration, a long-shadow detection method and its corresponding target detection method are proposed. By combining the two detection results, we get the comprehensive detection result. Experiments based on SSS images in different scenarios proved the validity of the proposed method.

Sediment classification based on Chirp sonar data is very important in support of marine science and engineering. The traditional adopted spectral-ratio (SR) method is widely applied for its theory of simplicity and easy employment. However, the performance of SR method is heavily degraded by spectrum vibrations introduced by overlapping reflections. To solve this problem, a reflection signal decomposition method is proposed in this paper, which decomposes the overlapping reflection into separate reflection sub-signals to avoid the spectrum vibration. Firstly, we re-derive the overlapping reflection expression from the convolution model and give the initial model for the decomposition of the Chirp sonar signal. Then, introducing the spectrum smooth prior into consideration and incorporating it with the fidelity and the bandwidth terms, the decomposition model with constraints is proposed. After that, an iterative algorithm is introduced to solve the model optimization problem. The reflection sub-signal can be well obtained. Finally, through a least square linear fitting for the logarithm of the spectral ratio (the log-SR) of reflection sub-signals, quality factor (Q) which is applied for sediment classification can finally be achieved. Both real and simulated experiments have been conducted to verify the proposed method and some meaningful discussions are also drawn.

The use of side-scan sonar (SSS) images is limited by radiometric distortion, which is affected by factors such as angle response, beam pattern, and incident energy. Traditional methods have overlooked the impact of sediment-dependent angle response, degrading performance in complex SSS images. In this study, we propose a method for correcting radiometric distortion on SSS images based on the variational Retinex framework and sediment characteristics. First, we analyze the imaging theory of SSS and the angle responses of various sediments. We introduce a mixed Gaussian model to describe the feature of SSS backscatter strength (BS) distribution. After that, a radiometric distortion correction model, incorporating the sediment BS information constraint, low-rank constraint on the incident illumination component, and an anisotropic total variation (ATV) constraint on the albedo, is presented. BS information is obtained through sediment classification based on mixed Gaussian model optimization. A minimization optimization method is used to solve the model, and an SSS image with corrected radiometric distortion is obtained. Experiments with different SSS images demonstrate the effectiveness of the proposed method.

Radiometric distortion caused by the time-varying gain (TVG), beam patterns, angular responses, and sonar altitude variations, highly degrades the quality of side-scan sonar (SSS) images. Thus, radiometric distortion correction becomes a fundamental step for SSS image processing which holds vital importance for geomorphic applications. However, existing methods cannot take the prior information of the acoustic illumination component as well as the feature of seafloor into consideration well, which would easily cause damage to the image and also always be powerless for residual stripe noise. In this paper, a novel radiometric correction method is proposed. First, we give a detailed analysis of the SSS imaging theory based on the Lambert’s law as well as the prior knowledge about the characteristics of SSS images. Then, incorporating the prior of the SSS imaging process, the low-rank constraint is specifically introduced for the illumination component, while the anisotropic total variation (ATV) constraint is used to constraint the albedo component, combining other constraints, a decomposition model is proposed to correct the radiometric distortion based on the SSS imaging theory. And an alternative minimization method has been adopted to solve the proposed model effectively. Experiments proved the validity of the proposed method.