Underwater mapping 2026
BUTTON TEXT Previous page Back to the menu
01 - Robust Localization of Flange Interface for LNG Tanker Loading and Unloading Under Variable Illumination a Fusion Approach of Monocular Vision and LiDAR
Authors: Mingqin Liu, Han Zhang, Jingquan Zhu, Yuming Zhang and Kun Zhu This paper addresses the challenge of accurately localizing flange interfaces in LNG tankers under extreme illumination conditions. While monocular vision fails in glare or low light, and LiDAR lacks texture information, the authors propose a vision-LiDAR fusion method with three key innovations: an illumination-adaptive framework that dynamically adjusts parameters based on grayscale evaluation, a multi-constraint flange detection strategy using physical constraints and K-means clustering, and a customized fusion pipeline combining ROI extraction, plane fitting, and 3D circle solving to compensate for individual sensor limitations.
02 - A lightweight underwater image and video enhancement method based on multi-scale feature fusion
Authors: Gaosheng Luo, Haiyang Li, Huanhuan Wang, Hengshou Sui, Xuewen Zhang, Rongjun Zhang, Bocheng Chen, and Zhe Jiang Underwater image quality is reduced by light absorption and scattering, affecting AUVs and marine monitoring. The authors propose UIVE, a lightweight enhancement algorithm for underwater visuals. Key features include using residual blocks instead of batch normalization, multi- scale connections for detail preservation, and an adaptive brightness correction module.
Click on the octopus to return to the top of the page
03 - A Review of Learning-Based Sonar Signal Processing Using Neural Network
Authors: W. S. C. Rodrigo, Udara S. P. R. Arachchige, C. Perera, C. Gunarathna, W. H. Y. N. Samarasinghe, S. L. Mallawathanthri, D. D. P. P. Jayawardhana, Z. I. Coonghe, G. K. W. P. S. G. Kumbura, A. H. M. K. P. Abeysinghe, S. J. M. S. Theekshana, D. M. C. Ronali, D. I. U. Rupasinghe, M. D. Wijesinghe This paper reviews learning-based sonar signal processing techniques, primarily neural networks, as a data-driven alternative to conventional methods that are limited by handcrafted features and model assumptions in complex underwater environments. The review categorizes existing studies into feature-based deep learning, end-to-end learning from raw acoustic data, and hydrophone-array processing. It also covers general applications like sonar imaging and target classification.
04 - Underwater 3D target detection: Semi-analytic Monte Carlo model and UAV-based scanning lidar system
Authors: Xinke Hao, Yan He, Huixin He, Deliang Lv, Yingjie Ruan, Hui Qi, Guangxiu Xu, and Junwu Tang A semi-Monte Carlo model (MCT) was developed to simulate lidar detection of underwater targets, considering interactions with the target, surface waves, and stratified water. This model was validated by a UAV-based linear scanning oceanic lidar system (SOL) and achieved less than 5% error within 50m depth. Field experiments confirmed SOL's consistent target localization. The MCT model was then used to analyze SOL's detection capabilities for different water types (Jerlov II and 3C), introducing an extended detection range concept to assess how horizontal scanning resolution changes with target depth, offering guidance for optimizing detection and efficiency.
05 - An explainable context-adaptive fusion and expert-in-the-loop evaluation framework for underwater sonar image classification
Authors: Kamal Basha, Anukul Kiran, Athira Nambiar, Suresh Rajendran, and Sooraj K Ambat Sonar image interpretation is critical for identifying submerged objects in underwater exploration. However, conventional deep learning models often fail to capture sonar-specific features such as acoustic shadows and highlights, and they typically operate as black boxes, limiting both performance and trust. To address these challenges, the authors propose an explainable sonar image classification system that fuses specialized classifiers for shadows, highlights, and general features through a context-adaptive fusion mechanism. The system further incorporates expert-in-the-loop evaluation via the Augmented QUality Assessment for eXplainability (AQUA- X) framework to ensure interpretability and trust.
06 - Analysis of Underwater Single-Photon LiDAR Signals: A Comprehensive Study on Multi-Parameter Coupling Effects
Authors: Ceyuan Wang, Shijie Liu, Shouzheng Zhu, Wenhang Yang, Chenhui Hu, Yuwei Chen, Chunlai Li, and Jianyu Wang Underwater laser signal attenuation poses challenges for conventional detection, but single-photon LiDAR (SPL) with high sensitivity offers a promising solution. Prior studies mainly examined isolated parameters, leaving the coupled effects of environmental and system factors underexplored. This research developed a 532 nm underwater SPL system to systematically investigate multi- parameter coupling in laboratory water tanks, varying turbidity, detection distances, laser energy levels, integration times, and target types.
07 - Comparison of Conventional, Rake, and Sonar-Based Biophysical Habitat Measurements in a Shallow Ontario River
Authors: Karl A. Lamothe, Jason Barnucz, D.  Andrew,   R. Drake Accurate habitat data is vital for managing and restoring freshwater species, but fine-scale sampling is labor- intensive. This study in Ontario, Canada, compared sonar- derived habitat measurements (depth, macrophyte volume, substrate) with conventional point-based methods in a shallow river. Both methods showed nearly all areas were <2m deep, though conventional depth readings were higher at 88% of sites. Depth correlations were strong, but substrate and macrophyte measurements showed weak alignment. Differences stemmed from measurement scale and inherent errors in both methods. The optimal approach depends on the precision required for management decisions.
08 - SonarKAN:Sonar Kolmogorov-Arnold Network for Disentangling Passive Sonar Signatures
Authors: Xiaoliang Chen Data-driven neural models for passive sonar can be accurate but hard to interpret. The proposed Physics- aligned Sonar Kolmogorov–Arnold Network (SonarKAN) aligns networks with the passive sonar equation's structure. It separates transmission loss from source signatures and uses a physics-informed ridge projection for faster convergence. Simulations show improved accuracy and robustness to label noise compared to a multilayer perceptron (MLP).
09 - Underwater SLAM and Calibration with a 3D Profi ling Sonar
Authors: António Ferreira, José Almeida, Aníbal Matos, and Eduardo Silva High-resolution underwater mapping is essential for the sustainable blue economy, supporting offshore energy expansion, marine habitat protection, and resource monitoring. This work presents a pose-graph Simultaneous Localization and Mapping (SLAM) and calibration framework specifically designed for 3D profiling sonars such as the Coda Octopus Echoscope 3D. It employs a probabilistic scan-matching method for direct scan-to-scan 3D sonar registration, enabling accurate trajectory and map estimation even under degraded dead-reckoning conditions.
10 - Volumetric Path Planning and Visualization for ROV-Based Forward-Looking Sonar Scanning of 3D Water Areas
Authors: Yu-Cheng Chou, and Wei-Shan Chang Remotely operated vehicles (ROVs) equipped with multibeam forward-looking sonar are widely used for underwater object searches in low-visibility environments. Achieving complete three-dimensional (3D) scan coverage within a limited mission duration is challenging due to sonar beam geometry and vehicle motion constraints. This study presents a deterministic, geometry- driven framework for volumetric path planning of forward-looking sonar scans in predefined circular and rectangular underwater volumes. The approach constructs layered planar scan trajectories by explicitly incorporating sonar detection range, horizontal and vertical beamwidths, and scan volume geometry.
12 - A Comparison of Detection Methods for Identifying the Presence of Active Sonar in Long-Term Passive Acoustic Data: A Case Study Within a Scottish Marine Protected Area
Authors: Benedict L. Dell, Ellen L. White, Denise Risch, Jonathan M. Bull, Paul R. White & Nienke C. F. van Geel Standardized methods for detecting and classifying active sonar are lacking, which affects ecological research using Passive Acoustic Monitoring (PAM) data. This chapter compares two classification pipelines for identifying military sonar in 20-minute files using deep learning and Gradient Boosted Random Forest.
13 - The Principle and Application of Underwater Sonar Systems
Authors: Siyi Wang, Yiduo Wang, and Lang Yin As underwater detection needs increase in marine research and engineering, sonar systems have become a key research focus. Enhancing their precision, efficiency, and adaptability is urgent. The article outlines different sonar types, detailing single-beam's lower cost and limited precision versus multi-beam's higher cost but better accuracy for complex seabeds. It summarizes current sonar systems, highlighting their deficiencies for future development.
11 - Spiking transformer with learnable threshold mechanism for underwater image dehazing to aid vision-based navigation
Authors: Vidya Sudevan, Rizwana Kausar, Sajid Javed, Hamad Karki, Giulia De Masi, and Jorge Dias snnTrans-DHZ is a lightweight spiking transformer for underwater image dehazing that uses a learnable threshold membrane potential and only 0.57 million parameters to boost clarity with low power through temporal SNN dynamics. Static images are converted to time-dependent RGB-LAB sequences and processed via three modules: K estimator, background light estimator, and soft image reconstruction, trained end-to-end with surrogate-gradient backpropagation through time.
14 - Embedded very-low-frequency underwater acoustic acquisition technology and experimental study for underwater explosion Monitoring
Authors: Xuexu Li, Zhen Song, Xinliang Pang, Fan Yang, Yu Lu, Yunfen Chang, Hao Yin, and Yunping Liu This study presents an embedded acquisition device designed to capture very-low-frequency (VLF) underwater acoustic signals from underwater explosions, addressing the lack of available off-the-shelf solutions. It effectively records and exports data using the FATFS file system. Successful experiments demonstrated the device's ability to acquire initial shock waves and long-range VLF signals, confirming its efficacy for VLF monitoring and underwater acoustic research, thus providing essential support for developing VLF monitoring equipment.
15 - Development and Testing of an In Situ Observation Device for Seafl oor Boreholes
Authors: Haodong Deng, Jianping Zhou, Xiaotao Gai, Chunhui Tao, and Bin Sui Mid-ocean ridge hydrothermal systems exchange heat and matter between seawater and lithosphere, yet high- precision sub-surface monitoring remains difficult. This study introduces a versatile in-situ penetration probe for multi-parameter vent monitoring, deployable post-drilling in boreholes or via ROVs. It features a titanium conical tip and chamber with tri-axial accelerometers plus dual temperature-pressure sensors. Design was optimized using SST k-w turbulence modeling and finite element analysis, then validated by lab vibration tests and shallow- water trials, providing a scalable, cost-effective framework for 3D sub-surface observation and resource exploration.
16 - AUV Path Planning Method for Underwater Moving Target Search Based on a Target-Position-Controlled Mutation Strategy Genetic Algorithm
Authors: Qiuying Wang, Shuo Zhang, and Yunfeng Han his paper proposes a genetic algorithm (GA)-based path planning method for autonomous underwater vehicles (AUVs) searching for stochastically moving underwater targets. To overcome limitations of conventional GAs—such as oversimplified target motion models and inflexible mutation—the method introduces a target- position-controlled mutation strategy. It first models stochastic target motion using a Markov process with grid- based spatial discretization to obtain the target’s position probability distribution. Second, it simulates sonar signal fluctuations on this grid to build a probabilistic detection model. These components are integrated into the GA to optimize the AUV search path.
17 - Global Path Planning for UUVs in Nearshore Environments Using an IAPF-RRT* Method
Authors: Xiaojing Fan, Fang Kong, Zhenhao Cui, and Yinjing Guo Nearshore environments present significant challenges for global path planning in unmanned underwater vehicles (UUVs) due to complex obstacles and dynamic disturbances. To address these challenges, this paper introduces an improved artificial potential field-guided method for efficient and robust path planning in coastal environments. The approach integrates an enhanced artificial potential field into sampling and node expansion to improve goal-directed exploration and obstacle avoidance. Additionally, a target-biased sampling strategy, an adaptive attraction mechanism, and a NURBS-based refinement scheme are incorporated to accelerate convergence and enhance trajectory smoothness.
18 - A Survey of Underwater Degraded Image Restoration
Author: Zheyuan Zheng Underwater images are vital for marine exploration and robot operations but suffer degradation like color shift, blurring, and low contrast from light absorption, scattering, and refraction. This survey examines their negative effects on visual tasks, categorizes restoration methods into physics-based, non-physics-based, and deep learning-based approaches, analyzes pros/cons and integration, and reviews datasets plus quality frameworks. It builds a structured knowledge base to aid researchers and promote faster deployment in marine engineering and underwater exploration.
19 - ACWMA: An Adaptive Cooperative WMA for 3D Path Planning of UUVs in Complex Marine Environment
Authors: Jingyi Bai, Yong Liu, and Xiaoyu Li Underwater images are vital for marine exploration and robot operations but suffer degradation like color shift, blurring, and low contrast from light absorption, scattering, and refraction. This survey examines their negative effects on visual tasks, categorizes restoration methods into physics-based, non-physics-based, and deep learning-based approaches, analyzes pros/cons and integration, and reviews datasets plus quality frameworks. It builds a structured knowledge base to aid researchers and promote faster deployment in marine engineering and underwater exploration.
20 - An Integrated Framework for Safe and Efficient AUV Navigation: Synergizing Enhanced Path Planning, Curvature-Adaptive Tracking, and Information-Driven 3D Exploration
Authors: Mingming Xiao, Yuliang Wen, Jiaheng Li, Naiyao Liang, and Dan Xiang This paper introduces an integrated framework for autonomous underwater vehicles (AUVs) in complex, unknown environments, combining enhanced path planning, curvature-adaptive trajectory tracking, and sonar-constrained 3D exploration. The path planner incorporates safety margins, 3D obstacle avoidance, and online replanning, while the tracking module uses B-spline optimization and speed control for smooth trajectories. The exploration strategy employs frontier clustering, information gain evaluation, and TSP optimization.
21 - Quasi-Static In Situ Deep Learning for Forward-Looking Sonar Target Detection in Complex Underwater Environments
Authors: Yixuan Chen, Zhenqing Ding, Yu Feng, Jiale He, Ziqin Xie, Tinggang Xiong, Kai Chen, and Qi Gao Forward-looking sonar (FLS) target detection for AUVs is challenged by acoustic distortions, environmental variability, and scarce fine-annotated data—limiting standard deep learning. This work introduces a quasi-static in situ learning paradigm for underwater acoustic target detection (UATD), combining lightweight deep learning with physics-informed scene priors: echo-intensity–based probability weighting and acoustic attenuation compensation. These enable pixel-wise sonar image enhancement and score-level fusion of statistical descriptors with neural predictions, allowing dynamic adaptation to real-time conditions.
22 - VK-RRT*: A Multi-Constraint Coupling Approach to Energy- Efficient AUV Path Planning in Three-Dimensional Ocean Current Environments
Authors: Ziming Chen, Jinjin Yan, Huiling Zhang, and Xiuyan Peng This paper introduces an energy-efficient velocity- kinodynamic RRT* algorithm to resolve multi-constraint coupling issues, energy estimation inaccuracies, and dynamic infeasibility in autonomous underwater vehicle (AUV) path planning. The approach integrates velocity–energy, velocity–curvature, and velocity–safe- distance models, enhanced by potential-field-guided sampling, adaptive step-length expansion, kinodynamic constraints, and a multi-objective cost function.
23 - A New Calibration Method Based on Gravity-Assisted Navigation
Authors: Zhibo Zhou, Jianfeng Wu, Huabing Wu, Lintao Liu, Xinghui Liang, Hubiao Wang, Guocheng Wang, Junjian Lang, Zhimin Shi, and Xiaolong Guan Gravity-assisted navigation (GANT) enables autonomous, passive, and long-endurance navigation for underwater vehicles by correcting inertial navigation systems (INS) using gravity matching. This study experimentally validated GANT using shipborne INS data, gravity measurements, and a satellite-derived gravity database, employing normal time-frequency transform (NTFT) for error correction and Kalman filtering for matching algorithms.
24 - Ultra-Fast Object Detection for Side-Scan Sonar Images via Target Presence Awareness
Authors: Guoqing Xie, Guang Pan, Ju He, Hu Xu, and Yang Yu Side-scan sonar (SSS) imaging is vital for underwater perception in AUVs, but real-time object detection is hindered by sparse targets and limited computational resources. Existing dense inference methods suffer from redundancy and deployment limitations. We propose a lightweight, ultra-fast SSS object detection framework using target presence awareness. The method employs a coarse-to-fine approach: a target presence analysis (TPA) module filters out irrelevant patches using spatial- frequency convolution to capture structural and contextual cues efficiently, followed by an Object Forward Detection (OFD) module with an AttnConv-enhanced detector for fine-grained analysis.
25 - A Lightweight Forward-Looking Sonar Sensing Framework for Embedded Target Detection in Resource-Constrained Underwater Systems
Authors: Hong Peng, Chaolin Yang, Chen He, Wei Ye, and Renyou Yang This study introduces a lightweight forward-looking sonar (FLS) target detection framework for resource-constrained underwater systems. It addresses key challenges, weak target echoes, clutter-induced occlusion, and strict embedded hardware limits, via three innovations: (1) FPN- Mix, a compact backbone with Conv-Mix for efficient contextual aggregation; (2) a target-aware dynamic weighting loss that prioritizes hard samples (e.g., weak echoes, ambiguous boundaries); and (3) multi-level knowledge distillation from an enhanced teacher to the student detector.
26 - HydroAir: An Air-Propelled Surface Vehicle for Autonomous Navigation and 3D Reconstruction in Shallow and Obstacle-Rich Aquatic Environments
Authors: Leonardo de Mello Honório, Vinícius Ferreira Vidal, Iago Zanuti Biundini, Rodolfo Almeida Machado, Felippe Fernandes, and Murillo Ferreira dos Santos HydroAir is a novel air-propelled Unmanned Surface Vehicle (USV)designed for shallow, obstacle-rich waters like lakes and dams. Its aerial propulsion overcomes submerged obstacles and shallows where traditional propellers fail, offering reliable maneuverability in manual/autonomous modes. Serving as a mobile sensing platform, it acquires bathymetric data fused with UAV aerial reconstructions into unified 3D models. Real-world tests at a Brazilian hydroelectric dam confirmed operational feasibility and stability, making it promising for environmental monitoring and mapping.
27 - An underwater dual-modal denoising detection network for illumination fluctuation and dense occlusion scenes
Authors: Chao Zhang, Xingkun Li, Shuang Wu, Zhongfeng Zhang, Xiangyang Cao, Tiantian Xia, and Wenlong Yu Underwater object detection faces challenges due to light scattering and absorption, which degrade RGB images and reduce detection accuracy in occluded scenes. The proposed UW-DualDet network integrates depth information to mitigate RGB degradation by introducing two key modules: the Underwater Denoise Feature Fusion (UDFF) module, which suppresses noise and fuses RGB and depth data to handle extreme illumination, and the Underwater Feature Enhancement Module (UFEM), which enhances feature representation using multi-scale convolutions. Additionally, a Gaussian-filter-enhanced is designed to reduce false detections.
28 - Multi-Source Sensor Fusion Localization Method for Autonomous Underwater Vehicles Based on Deep Learning
Authors: Xin Pan, Guoli Feng, Haiyan Zeng, and Qunhong Tian This study introduces a deep-learning-based multi-source sensor fusion framework to enhance the underwater localization accuracy of Autonomous Underwater Vehicles (AUVs) during long-duration missions. The framework integrates high-frequency data from INS and DVL for continuous position propagation, while low- frequency corrections from USBL and Sonar mitigate error accumulation. Three models: DNN, LSTM, and BSsMSLNN, were evaluated against EKF and SVIn2, with BSsMSLNN demonstrating superior performance in trajectory fitting, RMSE, and R². The approach offers a viable solution for high-precision AUV navigation in GPS-denied underwater environments.
29 - E2E-AUD: An End-to-End Adaptive Underwater Detection Framework Integrating Physical Priors and Frequency-Adaptive Learning
Authors: Wenhao Zhou, Junbao Zeng, Shuo Li, and Yuexing Zhang Underwater detection is crucial for Autonomous Underwater Vehicles (AUVs) but has challenges like poor image quality and complex objects. Current methods often separate image enhancement and detection, which can lead to poor performance with irregular targets. To improve this, the End-to-End Adaptive Underwater Detection framework (E2E-AUD) combines image enhancement and detection in one system. It uses linear deformable convolution to better handle varying targets and Haar wavelet downsampling to keep important details. Tests on the DUO and URPC datasets show that E2E-AUD surpasses other detectors and maintains a fast processing speed with low computational needs.
30 - A Spatial Distribution Probability-Guided Detection Framework for Underwater Sonar Imagery
Authors: Dayu Jia, Yan Huang, Jianan Qiao, Zhenyu Wang, Hao Feng, and Jiancheng Yu Underwater target detection using side-scan sonar is important for defense and economic reasons but faces challenges like few targets, high costs, and difficulties in analyzing data. To solve these problems, especially for small-object detection, we propose a Spatial Distribution Probability-Guided Detection Framework for Unmanned Underwater Vehicles (UUVs). This framework uses a pre- trained model to create maps that help detect targets accurately with little data. Experiments showed that this system greatly improves detection performance and can effectively locate and cluster targets.
31 - Underwater Robot Object Detection Algorithm Based on YOLOv11
Authors: Yongqing Shi, Wei Chen, Duo Wan and Lu Han Underwater environments suffer from scattering, absorption, and uneven lighting, complicating target detection despite the ocean's vast coverage. This study introduces a degradation-aware YOLOv11s-based framework to enhance underwater robotic object detection. It integrates three key improvements: (1) Spatial Channel Reconstruction Convolutions (SCConv) with SRU- CRU to reduce redundancy and better capture target features, (2) a Shuffle Attention module to strengthen fine- grained feature representation, and (3) Focaler-IoU for more stable bounding-box regression. The enhanced model achieves an mAP@0.5 of 88.4%, outperforming the baseline YOLOv11s by 3.2% while meeting real-time demands for underwater applications.
32 - Visual enhancement and 3D representation for underwater scenes: a review
Authors: Guoxi Huang, Haoran Wang, Brett Seymour, Evan Kovacs, John Ellerbroc, Dave Blackham & Nantheera Anantrasirichai Underwater visual enhancement (UVE) and 3D reconstruction face major challenges from complex aquatic imaging conditions. This paper provides a comprehensive review, introducing fundamental physical models, surveying methods from non-learning approaches to data-driven techniques like NeRF and 3D Gaussian Splatting, evaluating state-of-the-art algorithms quantitatively and qualitatively on benchmarks, and highlighting future research directions.