ABSTRACT With the continuous development of emerging technologies such as big data and artificial intelligence, the related technologies of perception tasks on which underwater object tracking relies have made great progress. However, a significant barrier still exists in implementing high‐performance real‐time underwater object tracking on low‐power edge devices. To achieve real‐time tracking of underwater objects for edge devices, this article develops an underwater real‐time visual sensing system applied to AUV tracking. First, an underwater object tracking device is designed in this article employing a stereo binocular camera, an edge embedded NVIDIA Jetson Xavier NX and an STM32 control board. Then, after preprocessing, the input image with the effective USM algorithm, we propose a quick approach for detecting underwater objects based on SIoU‐YOLOv8n, which enables automatic object recognition and selection. At the same time, this article proposes a twin network UW‐Siam for continuous tracking of underwater objects, which achieves more accurate underwater object tracking. Finally, the algorithm is deployed to the designed real‐time underwater vision sensing system and tested in real‐world scenarios. The tracking accuracy reached 0.652, and the detection mAP reached 0.97. The results indicate that the system can rapidly detect and continuously monitor objects, performing well in real‐world scenarios with high accuracy and robustness.

This article proposes a robust peer-to-peer localization scheme using compact arrays in underwater time-varying channels. The method combines two-way ranging and the time difference of arrival (TDOA) positioning technique, utilizing three elements of the compact array to achieve accurate localization. To better adapt to time-varying channels caused by the inherent changes in the water medium, and improve localization accuracy, this article adopts a TDOA correction method based on channel feature matching using the correlation between the received signals of different elements of the compact array. This method combines line-of-sight with non-line-of-sight information of multipath signals to correct the TDOA. The specific localization steps are as follows. First, multipath features are extracted from the single-frame received signal of an array element through the Savitzky-Golay filter and dynamic time warping (DTW) algorithm. Then, based on the obtained multipath features of multiframe signals, hierarchical clustering is used to extract the channel features from the perspective of the current array element. Finally, the target position is estimated by combining two-way ranging with the TDOA values corrected by DTW from different array elements. The accuracy and stability of the proposed method are validated through simulations and pool experiments by comparing it with the traditional methods.

Due to the limited bandwidth and severe noise interference in underwater acoustic channels, underwater image transmission typically requires a coding scheme with a high compression ratio and strong robustness. However, existing semantic communication methods are predominantly designed for terrestrial wireless scenarios and do not fully consider the unique characteristics of underwater images and channel features. To address these problems, we propose an efficient semantic communication scheme of underwater images guided by physical priors. Specifically, we design a semantic encoder-decoder with a hybrid architecture of CNN and Swin Transformer for efficient semantic feature extraction and reconstruction. Meanwhile, we design an underwater physical prior fusion module, which leverages physical prior knowledge from underwater imaging to optimize the encoding process, preserving and enhancing essential semantic features. Additionally, a noise-aware training strategy is proposed to guide the network in learning the noise characteristics of measured underwater acoustic channels, improving its robustness in intricate marine conditions. Extensive experiments verify that our method outperforms existing schemes regarding transmission efficiency and image quality.

The dim shooting environment and light scattering and absorption frequently result in degraded underwater images. The images are characterized by uneven brightness, low contrast, color deterioration, and blurred details. Existing underwater image enhancement methods excel in full-reference and non-reference metrics, yet may fail to align with human visual tendencies. To make the restored images more consistent with natural visual effect, an underwater image enhancement framework named BRIUIE is proposed. BRIUIE draws inspiration from the morphology and functions of various cell layers in the vertebrate retina. Following the visual transmission mechanisms of retinal signals, image brightness is balanced by simulating the feedback and dynamic regulation processes of horizontal cells in response to illumination variation. Meanwhile, simulating the center-surround receptive fields of bipolar and ganglion cells and implementing the color opponent mechanism effectively mitigate color distortion and low contrast. The designed multi-scale feature fusion module facilitates the complementary advantage of the ON and OFF visual pathways of ganglion cells, employing a contrastive learning strategy to prevent overfitting because of simple consistency loss. Comprehensive full-/non-reference experiments demonstrate the proposed BRIUIE outperforms other SOTA methods in quantitative evaluations, while also delivering qualitative results that closely align with human visual assessment standards.

Optical remote sensing serves as a critical technology for exploring underwater environments. However, light absorption and scattering underwater significantly degrade underwater optical images, affecting the extraction and analysis of information. Underwater image enhancement (UIE) methods aim to eliminate this degradation and improve the visual quality of images. Nonetheless, the complex and dynamic underwater imaging environment, limited computing resources, and scarce training data/data pairs restrict the practical application of existing methods. To solve these problems, we propose an UIE network (UIEN) based on neuromorphic computing, which simulates the pathway of the visual system to perceive and process light information, and can use a lightweight network structure to achieve good performance through unsupervised learning. Specifically, we propose a visual perception module comprising a 2-D Duffing oscillator (2D-DO) with pixel-wise potential barrier parameters. This module can generate the stochastic resonance (SR) phenomenon to enhance the degraded image. Inspired by physics-informed learning, a dual-path neural network is employed to estimate the potential barrier parameters and solve the partial differential equation (PDE) that describes the visual perception module. Subsequently, we introduce three nonreference (NR) losses to guide the network training and improve the enhanced image’s visual quality. Extensive experiments demonstrate that the proposed method can achieve outstanding performance with less computing resource cost compared to state-of-the-art (SOTA) methods. Furthermore, we examine the generalization and versatility of the proposed method to establish its reliability across various degradation types and tasks in practical applications of optical remote sensing.

Abstract Drowning is a significant public health concern. A video drowning detection algorithm is a helpful tool for finding drowning victims. However, there are three challenges that drowning detection research typically encounters: a lack of actual drowning video data, subtle early drowning traits, and a lack of real time. In this paper, the authors propose an underwater computer vision based drowning detection device composed of embedded AI devices, camera, and waterproof case to solve the above problems. The detection device utilizes the high‐performance computing of Jetson Nano to realize real‐time detection of drowning events through the proposed drowning detection algorithm on the acquired underwater video stream. The proposed drowning detection algorithm primarily consists of two stages: in the first step, to successfully solve the interference of the surroundings and to give a trustworthy basis for video drowning detection, the YOLOv5n network is used to detect the near‐vertical human body based on the characteristics of the drowning person. In the second stage, the authors propose a lightweight drowning detection network (DDN) based on a deep Gaussian model for fast feature vector detection. The lightweight DDN is combined with the Gaussian model to detect anomaly in the high‐level semantic features, which has higher robustness and solves the lack of drowning videos. The experimental results show that the proposed drowning detection algorithm has good comprehensive performance and practical application value.

Abstract Accurate fish size measurement in breeding areas is crucial for the fishing industry. Unlike acoustic methods with high equipment cost and low measurement accuracy, current image‐based methods offer a promising alternative. However, these image‐based methods still face challenges in selecting measurement points. To address this issue and achieve precise measurements of individual fish, this paper introduces an automatic fish size measurement method based on key point detection. We established a Fish‐Keypoints dataset and utilized deep learning techniques for the detection of fish and their key points. Using a binocular camera system, we reconstruct a three‐dimensional coordinate system to measure key points at the fish's head and tail, facilitating fish length calculation. The detection model achieves an accuracy of 85.1% in key point detection. The proposed method is tested in both land and underwater environments, demonstrating a relative measurement error of approximately 7% for fish in pools. This confirms the proposed method's ability to accurately detect measurement points, offering superior accuracy compared to other methods.

In the design of low-power long-range (LoRa) modulation systems in underwater Internet of Things (IoT) applications, multipath and noise in underwater acoustic (UWA) channels need to be considered. Commonly used LoRa modulation technologies, such as chirp-MFSK and chirp-MCSK, perform data modulation in a single dimension, and cannot provide reliable anti-multipath and anti-noise performance at the same time. This article proposes a two-dimensional modulation method chirp-MFCSK for the LoRa system. This method applies the$M_{2}$-ary chirp-MCSK subsystem in the$M_{1}$subbands of the FSK dimension. In specific UWA channels, the chirp-MFCSK modulation obtains a larger processing gain under the premise of suppressing the influence of multipath effects by flexibly adjusting the orders of the two dimensions. Simulation results and field experimental data show that in different UWA channels, the bit error rate performance of the optimal-ordered chirp-MFCSK modulation is several orders of magnitude better than that of the single-dimensional modulation methods.

Underwater acoustic channel (UAC) is featured as fast time-varying characteristic, and challenges the transmission designs. To countermine the time variation effect, we propose an adaptive design for orthogonal frequency division multiplexing (OFDM) transmission systems by utilizing the long-term stability of the second-order statistics of the channel state information (CSI). We derive the analytical expression of signal-to-interference-plus-noise-ratio (SINR) at each subcarrier to reach the target error performance based on the statistical information of the CSI. Thereafter, a new adaptive coding and bit-power loading algorithm with low computational complexity is proposed to pursuit the highest achievable bit rate with fixed error rate. The validity of SINR calculations and the effectiveness of the proposed adaptive algorithm are demonstrated under various conditions, wherein both simulated and measured channels have been tested.

In sonar applications, important information such as distributions of minerals, underwater creatures has a high probability of being contained in sonar images. In many underwater applications such as underwater rescue and biometric tracking, it is necessary to send sonar images underwater for further analysis. Due to the bad conditions of underwater acoustic channel and current underwater acoustic communication technologies, sonar images very possibly suffer from several typical types of distortions. As far as we know, limited efforts have been made to gather meaningful sonar image databases and benchmark reliable objective quality model, so far. This paper develops a new objective sonar image quality predictor (SIQP), whose core is the combination of two features specific to a quality measure of sonar images. These two features, which come from statistical and structural information inspired by the characteristics of sonar images and the human visual system, reflect image quality from the global and detailed aspects. The performance comparison of the proposed metric with popular and prevailing quality evaluation models is conducted using a newly established sonar image quality database. The results of experiments show the superiority of our SIQP metric over the available quality evaluation models.

Underwater image compression is an important and essential part of an underwater image transmission system. An assessment and prediction method of effectively compressed image quality can assist the system in adjusting its compression ratio during the image compression process, thereby improving the efficiency of the image transmission system. This study first estimates the perceived quality of underwater image compression based on embedded coding compression and compressive sensing, then builds a model based on the mapping between image activity measurement (IAM) and bits per pixel and structural similarity (BPP-SSIM) curves, next obtains model parameters by linear fitting, and finally predicts the perceived quality of the image compression method based on IAM, compression ratio, and compression strategy. Experimental results show that the model can effectively fit the quality curve of underwater image compression. According to the rules of parameters in this model, the perceived quality of underwater compressed images can be estimated within a small error range. The presented method can effectively estimate the perceived quality of underwater compressed images, balance the relationship between the compression ratio and compression quality, reduce the pressure on the data cache, and thus improve the efficiency of the underwater image communication system.

Image restoration is a critical procedure for underwater images, which suffer from serious color deviation and edge blurring. Restoration can be divided into two stages: de-scattering and edge enhancement. First, we introduce a multi-scale iterative framework for underwater image de-scattering, where a convolutional neural network is used to estimate the transmission map and is followed by an adaptive bilateral filter to refine the estimated results. Since there is no available dataset to train the network, a dataset which includes 2000 underwater images is collected to obtain the synthetic data. Second, a strategy based on white balance is proposed to remove color casts of underwater images. Finally, images are converted to a special transform domain for denoising and enhancing the edge using the non-subsampled contourlet transform. Experimental results show that the proposed method significantly outperforms state-of-the-art methods both qualitatively and quantitatively.

Underwater hostile channel conditions challenge video transmission designs. The current designs often treat video coding and transmission schemes as individual modules. In this study, we develop an adaptive transceiver with channel state information (CSI) by taking into account the importance of video components and channel conditions. The design is more effective than the traditional ones. However, in practical systems, perfect CSI may not be available. Therefore, we compare the imperfect CSI case with existing schemes, and validate the effectiveness of our design through simulations and measured channels in terms of a better peak signal-to-noise ratio and a higher video structural similarity index.

One of the most critical missions of sonar is to capture deep-sea pictures to depict sea floor and various objects, and provide an immense understanding of biology and geology in deep sea. Due to the poor condition of underwater acoustic channel, the captured sonar images very possibly suffer from several typical types of distortions before finally reaching to users. Unfortunately, very limited efforts have been devoted to collecting meaningful sonar image databases and benchmark reliable objective quality predictors. In this paper, we first generate a sonar image quality database (SIQD), including 840 images. All distorted images were collected without artificially introducing any distortions beyond those occurring during compression and transmission. The subjective quality assessment was conducted for gathering mean opinion score (MOS) to represent the image quality and existence of target (EOT) which describes whether the image is useful. Based on the built SIQD database, state-of-the-art general image quality metrics were found to poorly correlate with “ground-truth” MOS. As a consequence, this paper further develops a novel full-reference local entropy backed sonar image quality predictor (LESQP). The experimental results demonstrate the superiority of our LESQP metric over the available quality measures.