To overcome the shortcomings of the traditional manual detection of underwater targets in side-scan sonar (SSS) images, a real-time automatic target recognition (ATR) method is proposed in this paper. This method consists of image preprocessing, sampling, ATR by integration of the transformer module and YOLOv5s (that is, TR–YOLOv5s), and target localization. By considering the target-sparse and feature-barren characteristics of SSS images, a novel TR–YOLOv5s network and a down-sampling principle are put forward, and the attention mechanism is introduced in the method to meet the requirements of accuracy and efficiency for underwater target recognition. Experiments verified the proposed method achieved 85.6% mean average precision (mAP) and 87.8% macro-F₂ score, and brought 12.5% and 10.6% gains compared with the YOLOv5s network trained from scratch, and had the real-time recognition speed of about 0.068 s per image.

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.

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.

Sidescan sonar (SSS) creates images through interpolation of scan lines. The instability of the transducer position caused by the vessel's turning and the boost from the swells, which leads to misalignment, overlapping, and uneven spacing of the scan lines (heading distortion), is a problem that has been largely overlooked in the processing of SSS data. The traditional interpolation method tends to cause serious mosaic and overlapping texture problems in SSS, which interferes with the subsequent image analysis work. Additionally, the practice of simply cutting and discarding also tends to waste resources. To enhance data usability, this article leverages the deep convolutional neural network (DCNN) to learn the correlations between textures, transforming the issue of heading anomaly correction into one of misalignment fusion in overlapping areas and gap texture filling, providing a feasible scheme for detecting scanning line heading anomalies and filling gaps. Addressing the lack of continuity in textures repaired by DCNN in larger gaps, a continuity-guided branch network is proposed to help the main repair network consider texture continuity. Through quantitative evaluation with real sonar images as a reference and qualitative evaluation without a real image reference, the effectiveness of the proposed method in filling gaps in scan lines with varying degrees of anomalies has been validated. For regions with minor heading anomalies, the method achieves repair results comparable to traditional interpolation techniques. In the area with large anomalies, the proposed method shows improvements over the traditional optimal method, with the peak signal-to-noise ratio index increase of over 5%, the structural similarity index improvement of over 20%, and the naturalness image quality evaluator index enhancement of over 8%, greatly enhancing the data's usability.