To enable side-scan sonar (SSS) to be used for large-scale seabed sediment detection, classification, and composition analysis, this article breaks through the limitations of past superficial use of echo characteristics and proposes a hidden feature extraction method that combines imaging mechanisms with seabed reflection models. By using the penetration characteristics of sound waves, weak echo width (WEW) features were extracted from the water column region of SSS echoes. By using the Lambertian model and the concept of denoising diffusion probabilistic models (DDPMs), seabed topography and sediment reflection features were, furthermore, successfully decomposed from SSS echoes, and a method to extract features highly correlated with sediment composition from these components is also presented. The effectiveness of these hidden features was demonstrated through sediment classification and analysis experiments in the Bohai Bay of China. Compared to using only echo features, the hidden features extracted in this article improved classification accuracy by over 20% on common classifiers. Based on the hidden features and sediment distribution, the causes of seabed composition near Caofeidian were explained. Additionally, through feature importance and sediment response experiments, the response mechanisms of different hidden features to various sediment types were scientifically interpreted. This study demonstrates the significant application value of SSS in large-scale seabed sediment detection, classification, and analysis for geoscientific tasks.

To overcome the challenges of limited samples, difficult acquisition, under-representation, and labeling in utilizing sonar images and deep learning for target detection, recognition, and segmentation tasks for full-class underwater targets, we propose the Seg2Sonar network based on SPADE. This network generates images through segmentation maps, thus eliminating the need for sample annotation. Additionally, we incorporate the Skip-Layer channel-wise Excitation (SLE) module into the SPADE network to enhance feature extraction ability with minimal training samples. To improve the realism of generated images, we introduce the Focal Frequency Loss (FFL) module, and propose the Elasticity loss (EL) strategy to improve the random combination capability of the network, considering the characteristics of low resolution and severe distortion of sonar images. Furthermore, we propose a weight adjustment (WA) strategy that tackles the challenge of low and unbalanced feature representation with few samples by taking into account the unbalanced distribution of features using prior information. hese four improvements enable efficient sample augmentation of sonar images with limited samples. Building upon the improved Seg2Sonar network, we propose an underwater full-class target augmentation strategy. Based on the imaging characteristics of sonar images, we classify underwater full-class targets into four categories: texture level, group level, shape level, and intensity level. We provide corresponding augmentation strategies by leveraging similar features among sonar target images or adding external radar/optical features to supplement the diversity of features. Our experimental results demonstrate the efficacy of our proposed method in achieving sample augmentation of underwater full-class targets with minimal samples (less than 10) or even zero samples. The approach achieves about 90% accuracy in detection, recognition, and segmentation for all types of targets through deep learning methods. Our findings provide a promising solution for efficient sample augmentation of underwater full-class targets with limited samples.

To meet the demands of marine scientific research and ocean engineering construction for cost-effective, high-resolution, and large-scale acquisition of seafloor topography, this study proposes an unsupervised terrain reconstruction technique for side-scan sonar (SSS). Unlike conventional methods, this technique does not rely on externally measured depths for initial terrain and ground truth; instead, it constructs these elements based on its own imaging mechanism. Through the development of a rigorous seafloor reflection model, we have designed an unsupervised depth convolutional neural network (DCNN) inversion framework. This framework utilizes the depth of the sea bottom line as the initial terrain and employs the height calculated from the target’s shadow as the ground truth. The proposed network not only extracts absolute topographical information but also captures relative changes in beam pattern and seabed substrate properties. The experimental results demonstrate the method’s excellent accuracy in terrain inversion for areas with water depths ranging from 10 to 14 m, achieving a vertical accuracy exceeding 0.17 m. Moreover, the method exhibits robust resilience to residual radiometric distortions, variations in seabed substrate, noise, and shadows. This innovative approach provides a reliable methodology for acquiring and broadly applying high-resolution seafloor terrain, potentially advancing various fields of marine research and engineering.

In order to solve the problems of small samples, acquisition difficulties, under-representation and labeling difficulties in object detection, recognition and segmentation tasks for underwater all-category targets based on sonar images and deep learning methods. we propose a side-scan sonar full-class image sample augmentation method suitable for multi-task scenarios. Based on the superior image generation ability of the diffusion model, we use transfer learning to fine-tune the optical pre-trained model to build a side-scan sonar image generation model. Then, for the object detection task and semantic segmentation task, we use the image content and target shape as guidance information to guide the generation results of the diffusion model respectively. Meanwhile, proposed a mask synthesis method for SSS waterfall image generation based on the working principle of side-scan sonar. The synthesized mask images are used to guide the generation of side-scan sonar waterfall images. Finally, the underwater object detection and segmentation models are trained on the generated data. The experiment results show that training a model with generated data can be effective in improving accuracy.

In underwater perception and maritime surveys, due to the scarcity of training data and perturbation of speckle noise, the detection performance of underwater objects in side-scan sonar (SSS) images is limited. To address these problems, we proposed a noise feature disentanglement YOLO (NFD-YOLO) by combining noise-agnostic features learning and attention mechanism. Firstly, we rethink the speckle noise by treating it as the domain shift between the training dataset and real-measured SSS images and build a domain generalization-based (DG-based) underwater object detection framework. Then, we extend YOLOv5 with a feature manipulation module, a noise-agnostic subnetwork, and an auxiliary noise-biased subnetwork for noise features disentanglement, more biases toward noise-agnostic features and less reliance on noise-biased features in underwater object detection, respectively. Finally, the ACmix attention module is introduced for a more powerful learning capacity and attention to the object areas based on a small dataset. According to the experiment results, the proposed NFD-YOLO achieved 75.1% mean average precision (mAP) in the test domain, which increased by 7.5% than YOLOv5, and 75.7% ± 0.4% mAP and 77.5% ± 1.6% mAP for different speckle noise distributions and transfer directions, respectively, which verified its generalization ability and robustness for speckle noise. Therefore, the proposed method can mitigate the effects of speckle noise and provides a new thought to address the speckle noise in underwater object detection with a small dataset, which is of significance and benefits for underwater perception and maritime surveys.

To solve the shortage of training samples when using deep learning to detect shipwrecks, a comprehensive sample augmentation method is proposed. The method fully considers the imaging mechanism and environment of side-scan sonar (SSS), such as acoustic emission and reception, waterbody, target reflection, and seafloor background, and generates diverse and representative shipwreck samples from five aspects of target diversity, target texture, imaging resolution, equipment and environmental noises, and background through a series of novel sample augmentation methods. Under the condition of zero real SSS samples, a detection model of YOLOv5s was established with these amplified samples and achieved a mean average precision (MAP) better than 96% for real SSS data detection.