underwater mapping

Underwater mapping 2025 - Part B

Click on the octopus to return to the top of the page

01 - Ship trajectory prediction via a transformer-based model by considering spatial-temporal dependency

Authors: Xinqiang Chen, Peiyang Wu, Yuzhen Wu, Loay Aboud, Octavian Postolache, Zichuang Wang The article discusses the increasing importance of ship trajectory prediction in global maritime trade, emphasizing the challenges posed by environmental complexities, multi-factor influences, and data quality issues in Automatic Identification System (AIS) data. It presents a novel ship trajectory prediction model based on the Crossformer architecture, which incorporates innovative components like Dimension-Segment-Wise embedding and a Hierarchical Encoder-Decoder structure. The model’s effectiveness is validated through experiments on public AIS datasets, showcasing its superior performance against traditional models.

08 - Submarine Topography Classification Using ConDenseNet with Label Smoothing Regularization

Authors: Jingyan Zhang, Kongwen Zhang, and Jiangtao Liu This article discusses the necessity of classifying submarine topography and geomorphology for various applications in marine resource exploitation and ocean engineering. It critiques traditional visual interpretation methods for their inefficiency and subjectivity while highlighting the challenges faced by current machine learning models when integrating multi-source data. The authors propose a new model, utilizing pruning-enhanced ConDenseNet with label smoothing regularization, which significantly improves classification accuracy and reduces computational complexity compared to leading models.

11- Multi-Branch Towed Array System: Systematic Analysis of Modeling Methods, Environmental Responses and Mechanical Properties in Fracture Analysis

Authors: Jin Yan, Kefan Yang, Shengqing Zeng, Keqi Yang, Dapeng Zhang, and Keqiang Zhu The article discusses the critical role of multi-branch towed array systems in subsea information collection, emphasizing their significance in offshore construction and pipeline laying. It highlights the difficulties faced by these systems due to complex underwater conditions, including the risk of cable collisions and torsion interactions. Such challenges can lead to damaged towing systems and alterations in their formation, which are influenced by the specific sea state. The paper also draws a comparison between these towing systems and trawl nets to better understand their mechanical responses post-damage.

14 - Online Multi-AUV Trajectory Planning for Underwater Sweep Video Sensing in Unknown and Uneven Seafloor Environments

Authors: Talal S. Almuzaini, and Andrey V. Savkin This article discusses advancements in the navigation of Autonomous Underwater Vehicles (AUVs) for effective underwater sensing and monitoring. It presents a novel algorithm that tackles the challenges posed by uneven seafloor terrains and visibility constraints due to occlusions and camera limitations. The solution incorporates a trajectory planning algorithm that enhances coverage objectives while adhering to kinematic and safety constraints. The effectiveness of the proposed approach is demonstrated through simulations that show improved performance compared to traditional strategies.

17 - Adaptive Exposure Optimization for Underwater Optical Camera Communication via Multimodal Feature Learning and Real-to- Sim Channel Emulation

Authors: Jiongnan Lou, Xun Zhang, Haifei Shen, Yiqian Qian, Zhan Wang, Hongda Chen, Zefeng Wang, and Lianxin Hu Underwater Optical Camera Communication (UOCC) is a new method for secure data exchange in autonomous underwater vehicles. Its effectiveness relies on exposure time and International Standards Organization (ISO) sensitivity, affected by underwater conditions. Traditional systems struggle with changing conditions. A new framework combines imaging, environmental data, and a model for better prediction accuracy and real-time adaptation. This results in improved signal quality and potential for practical underwater communication use cases.

15 - Geostatistical uncertainty maps for real-world efficient AUV data collection

Authors: Ana F. Duarte, Lucrezia Bernacchi, Renato Mendes, João Borges de Sousa, and Leonardo Azevedo The article discusses a geostatistical modeling workflow designed to enhance the trajectory planning of Autonomous Underwater Vehicles (AUVs) for oceanographic surveys. It centers on two main objectives: predicting ocean temperature and spatial uncertainty, which informs optimized navigation paths, and facilitating the efficient assimilation of collected data into ocean models. An illustrative case study conducted off the coast of Western Portugal demonstrates the workflow’s efficacy, revealing improved data prediction capabilities and model reliability.

16 - RECAD: Retinex-Based Efficient Channel Attention with Dark Area Detection for Underwater Images Enhancement

Authors: Tianchi Zhang, Qiang Liu, Hongwei Qin, and Xing Liu The article focuses on improving visual target detection for Autonomous Underwater Vehicles (AUVs) by proposing a new image enhancement method named RECAD. This method addresses common problems found in weakly illuminated underwater images, such as blurring and color distortion. While existing deep learning methods struggle with feature extraction and high computational costs, RECAD utilizes Retinex theory and incorporates mechanisms for dark region detection and channel attention to enhance performance. Extensive testing shows that RECAD significantly outperforms existing models.

18 - Next-generation underwater localization: Artificial Intelligence- based and energy-aware approaches

Authors: Mainul Islam Chowdhury, Quoc Viet Phung, Iftekhar Ahmed, Walid K.Hasan, Daryoush Habibi This study highlights the problems associated with large propagation delays, the lack of GPS, node mobility, and limited acoustic link capacity that complicate underwater monitoring applications. Traditional localization methods’ limitations are discussed, particularly their sensitivity to noise and communication constraints. The article also emphasizes the importance of energy efficiency due to difficulties in replacing and recharging batteries in underwater environments. It provides a comprehensive review of how artificial intelligence (AI), including deep learning and machine learning, can enhance localization accuracy and robustness while integrating energy-saving strategies.

19 - A classification method of underwater target radiated noise signals based on enhanced images and convolutional neural networks

Authors: Yuan Muye, Lei Zhufeng, Dang Yinuo, and Shi Zanrong The article discusses the challenges of identifying underwater vehicles due to advancements in technology and the inadequacy of traditional classification methods based on signal characteristics. It presents a novel approach using multiple convolutional neural network (CNN) recognition methods enhanced by Gramian Angular Field (GAF) images. The paper details the process of converting underwater target noise signals into GAF images, which are then utilized to train various CNN models. The results indicate that the VGG-16 model demonstrated superior classification accuracy of 80.67%.

20 - RG-SAPF: A Scheme for Cooperative Escorting of Underwater Moving Target by Multi-AUV Formation Systems Based on Rigidity Graph and Safe Artificial Potential Field

Authors: Wen Pang, Daqi Zhu, Mingzhi Chen, & Wentao Xu This paper discusses how to escort a moving underwater target, like a human-occupied vehicle, using multiple autonomous underwater vehicles (AUVs) in difficult ocean settings. It introduces the “Reward-Guided State- Action Prediction and Filtering (RG-SAPF)” scheme for formation control and path planning, ensuring effective target escorting in complex environments.

02 - Underwater Digital Twin of Yuba Island: Mapping a Large-Scale Marine Ecosystem through Photogrammetry and 3D Modeling - Published by the International Journal on Advanced Science Engineering Information Technology

Author: Christoph Gerigk The project described in this paper aimed to construct a detailed digital twin of the underwater environment surrounding Yuba Island in the Red Sea, Saudi Arabia, mimicking the diving experience through precise 3D mapping of both underwater and above-water landscapes. Completed in six months, the model includes accurately rendered textures and colors, with a depth accuracy of ±0.10 m. It encompasses a diverse marine ecosystem, documenting over 100 local fish species, and culminates in various digital deliverables, including VR experiences and software applications.

03 - Underwater Mapping in Shallow Coastal Waters Using MBES and Photogrammetry: Applications in Archaeology and Marine Habitat Monitoring - Published by the International Journal on Advanced Science Engineering Information Technology

Authors: Saso Poglajen, & Mojca Poklar This study investigates the synergistic use of Multibeam Echosounder (MBES) and Structure-from-Motion (SfM) photogrammetry to enhance underwater mapping in shallow coastal environments, particularly within Slovenian waters. The research includes two case studies: mapping Posidonia oceanica seagrass meadows and analyzing the submerged archaeological site of Fizine. The integration of MBES’s spatial precision with the visual detail provided by SfM improves habitat classification and documentation of underwater features by capturing both broad structures and intricate details.

06 - Application and Challenges of Photogrammetry in Underwater Excavations: Case Studies from the Jeju Sinchang-ri and Gunsan Seonyudo Sites, Korea

Author: Jeong Heon Photogrammetry is increasingly used but faces challenges in underwater contexts due to strong currents and low visibility. This study compares its application at two Korean sites: Jeju Sinchang-ri—where clear water enabled high- resolution mapping using a Double Grid method and Metashape, yielding 3D models, orthophotos, and artifact distribution maps—and Gunsan Seonyudo—where thick sediment and poor visibility necessitated experimental adaptations like tide-timed surveys and underwater lighting to produce trench plans and stratigraphic cross- sections distinguishing artifact-bearing layers. The findings underscore the need to tailor photogrammetric methods to local underwater conditions, supporting more effective documentation as shipwreck discoveries increase.

07 - Sensor Synergy in Bathymetric Mapping: Integrating Optical, LiDAR, and Echosounder Data Using Machine Learning

Authors: Emre Gülher, and Ugur Alganci This study integrates multi-sensor datasets to improve bathymetric mapping in coastal and inland waters, leveraging sensor strengths via data fusion and machine learning for better spatial coverage, resolution, and accuracy.

09 - Deformable USV and Lightweight ROV Collaboration for Underwater Object Detection in Complex Harbor Environments: From Acoustic Survey to Optical Verification

Authors: Yonghang Li, Mingming Wen, Peng Wan, Zelin Mu, Dongqiang Wu, Jiale Chen, Haoyi Zhou, Shi Zhang, and Huiqiang Yao This study proposes a collaborative system integrating a deformable unmanned surface vehicle (USV) with a lightweight remotely operated vehicle (ROV). The USV, equipped with side-scan sonar (SSS) and multi-beam echo sounder (MBES), enables rapid large-area seabed mapping and object localization, while the ROV, with optical cameras, forward-looking sonar (FLS), and a manipulator, confirms and disposes of identified objects. Field trials in the South China Sea demonstrated the system's effectiveness.

10 - LGMMFusion: A LiDAR-guided multi-modal fusion framework for enhanced 3D object detection

Authors: Haixing Cheng, Chengyong Liu, Wenzhe Gu, Yuyi Wu, Mengye Zhao, Wentao Liu, Naibang Wang LGMMfusion is a novel LiDAR-guided multimodal fusion framework for autonomous driving object detection, designed to improve small-object detection by addressing limitations of sparse LiDAR points and low-resolution image features. Unlike conventional methods that fuse LiDAR and camera features only at the detection head, LGMMfusion leverages LiDAR depth priors to guide the generation of higher-quality image Bird’s Eye View (BEV) features before fusion. It enables early spatial interaction between point clouds and pixels via multi-head, multi- scale self-attention and adaptive cross-attention mechanisms, thereby ensuring better alignment

12 - Detection of Submerged Targets Beyond Eyes' Observation Using Satellite LiDAR and Multispectral Data

Authors: Hao Liu, Jian Yang, Yue Ma, Yao Li, and Xiao Hua Wang Underwater image quality is degraded by light absorption and scattering, hindering applications in marine robotics and archaeology. Existing enhancement methods often fail to recover fine details without artifacts. This work proposes a hybrid U-Net-Transformer (UTR) architecture that combines local feature extraction with global context modeling. Its key innovation is the Recurrent Multi-Scale Feature Modulation (R-MSFM) mechanism, a gated, multi- scale iterative refinement strategy in the decoder that preserves spatial fidelity during upsampling.

13 - Underwater Image Enhancement with a Hybrid U-Net- Transformer and Recurrent Multi-Scale Modulation-

Authors: Zaiming Geng, Jiabin Huang, Xiaotian Wang, Yu Zhang, Xinnan Fan, and Pengfei Shi This research introduces a novel underwater image enhancement framework for autonomous underwater vehicles (AUVs) to improve marine ecosystem observation and conservation. The proposed method combines a Denoising Diffusion Probabilistic Model (DDPM) for image restoration with an Attention-Enhanced Convolutional Blocks (AECB) augmented Transformer backbone. AECB modules use dual channel and spatial attention to boost visual quality. The framework is designed to be computationally efficient for onboard AUV deployment. Experiments show it outperforms existing transformer- diffusion methods, with the dual attention mechanism significantly improving image clarity and detail.

21 - DM-AECB: a diffusion and attention-enhanced convolutional block for underwater image restoration in autonomous marine systems

Authors: Naveen Kumar Tiwari, Abhishek Bajpai, Shashank Yadav, Anas Bilal, Abdulbasit A. Darem, Raheem Sarwar, and Jaibir Singh This paper discus This study presents a novel underwater image enhancement framework combining a Denoising Diffusion Probabilistic Model (DDPM) with Attention- Enhanced Convolutional Blocks (AECB) and a Transformer backbone for biodiversity monitoring and habitat mapping. The AECB modules employ dual channel and spatial attention mechanisms to enhance image quality, while a lightweight architecture with skip-sampling optimization enables efficient deployment on AUVs and underwater drones.

22 - Explainable underwater target recognition models: principles, methods, and applications

Authors: Tianyang Xu, Hongjian Jia, and Jixing Qin This paper systematically reviews explainable models for underwater target recognition, covering core concepts, methods, and research progress in sonar imaging, signal analysis, and autonomous navigation. It highlights future directions, such as causal reasoning, cross-modal collaboration, and physical knowledge integration, to guide the development of safe and reliable underwater intelligent systems.

26 - Applying Deep Learning to Bathymetric LiDAR Point Cloud Data for Classifying Submerged Environments

Authors: Nabila Tabassum, Henri Giudici, Vimala Nunavath, and Ivar Oveland Subsea environments are crucial for biodiversity, climate regulation, and human activities like fishing and resource extraction, necessitating accurate mapping for sustainable management. Airborne LiDAR bathymetry (ALB) offers high-resolution underwater data but generates large, complex datasets that challenge efficient analysis. This study applies deep learning models, Long Short-Term Memory (LSTM) and Bidirectional LSTM (BiLSTM), to multi- class classification of ALB waveform data. A preprocessing pipeline extracted and labeled waveform peaks into five classes: sea surface, water, vegetation, seabed, and noise. Experiments on two datasets showed high classification accuracy:

27 - A Fourier Neural Operator-enhanced parabolic equation framework for highly efficient underwater acoustic field prediction

Authors: Xinshuo Zheng, Haofeng Xia, Hanzhuo Wang, Chi Zhang, Jiaxi Duan, Jingyi Liu and Shuai Tang This work proposes an efficient solution method for the parabolic equation model based on the Fourier Neural Operator, which leverages grid-independent O(n log n) spectral convolution for rapid global feature extraction and robust correlations between physical field parameters and sound pressure fields. The model incorporates a continuous mapping that allows self-adjustment and adaptability to diverse marine scenarios, including non- uniform sound speed profiles, broadband sources, and sloped bathymetry.

28 - Mapping River Bed Topography in Whitewater Rapids Using Bathymetric LiDAR

Authors: Jan Rhomberg-Kauer, Theresa Himmelsbach, Florian Pöppl, Lucas Dammert, Martin Pfennigbauer, Gottfried Mandlburger Bathymetric LiDAR struggles to detect riverbeds in turbulent whitewater rapids of mountain rivers. This study introduces a novel full-waveform processing method that uses curve fitting to model and subtract the water column signal, isolates residual energy, and applies peak detection to recover previously undetected bottom returns. Validated across three rivers using total station ground truthing, the method reduces median vertical errors from 20.6 cm to 9.2 cm and from 40.4 cm to 22.5 cm, significantly filling data gaps in whitewater zones.

29 - A Review of Solid-State LiDAR Principles and Metasurface-Based LiDAR Sensors

Authors: Elif Demirbas, Braden Boucher, Matthew Baker, Joshua Andrews, William Cruz, Sara Mueller, and Samuel Serna-Otalvaro LiDAR is a promising technology for autonomous vehicles, with solid-state alternatives like microelectromechanical systems (MEMS) and optical phased arrays (OPAs) offering compact and robust solutions compared to traditional mechanical LiDAR systems with rotating mirrors. Two- dimensional optical metasurfaces enable beam steering by shifting the phase of incoming light; static metasurfaces are suitable for fixed beam directions, while dynamic (tunable) metasurfaces are essential for real-time beam scanning. Metasurface-based LiDAR provides advantages such as flat optics design, robustness, and absence of moving parts.

30 - Trajectory tracking control of autonomous underwater vehicles: a review from classical methods to AI-based approaches

Authors: Aly M. Eissa, Samer A. Mohamed, Mohammed Ibrahim Awad, Hossam E. Abd El Munim, Diaa Emad Autonomous Underwater Vehicles (AUVs) are crucial for tasks like seabed mapping and search-and-rescue operations, yet face challenges in trajectory tracking due to complex dynamics and uncertainties. This review categorizes tracking control strategies into three groups: classical controllers like PID; intelligent controllers including Fuzzy Logic and Reinforcement Learning; and hybrid approaches like Adaptive Neuro-Fuzzy systems. A meta- analysis of 120 studies shows performance trends, emphasizing future research in adaptive fault-tolerant control and hybrid learning..

04 - Scene-adaptive Underwater Image Enhancement With State- Space Model

Authors: Zhu Kou, Yu Wang, Ziying Li Underwater images suffer from degradation due to light absorption, scattering, and uneven illumination, complicating visualization tasks. While neural network- based enhancement methods have advanced, challenges like reliance on approximate reference maps and limited global dependencies hinder performance. To address this, the authors propose Mamba-SAU, a two-phase framework, augmented distribution estimation and consensus refinement, that reduces reference-induced errors. Incorporating the Mamba module (inspired by structured state space models) enables efficient global context awareness. Experiments demonstrate Mamba- SAU's superiority over state-of-the-art methods, validated by subjective and objective metrics.

05 - A parameterized model for the real-time reconstruction of underwater sound speed fields in oceanic eddy environment

Authors: Luochuan Xu, Anmin Zhang, Xuehai Sun, Jian Xu, Yi Liu, Dan Chen, and Linglong Chen This paper introduces a parameterized model to assess the impact of oceanic eddies on underwater sound speed fields using satellite remote sensing and Argo float data. The model distinguishes eddy structures from environmental regions to preserve acoustic characteristics while accounting for external influences. Unlike numerical simulations or in situ observations, it reconstructs sound speed fields in real-time within seconds using only surface eddy characteristics, significantly improving efficiency. Additionally, it outperforms conventional methods, achieving 6.7% higher accuracy in sound speed reconstruction and 13.5% better accuracy in acoustic propagation loss calculations.

23 - Further enhancing the sharpness of an acoustic focus employing split processing

Authors: Ting Li, Jiajing Wang, Yang Li, Sheng Li A low-frequency source traditionally suffers from a large focus, even when processed with artificial iterative phase- conjugated processing (AIPCP), which can surpass the half-wavelength limit in homogeneous media. To improve this, split artificial iterative phase-conjugated processing (SAIPCP) is introduced by combining AIPCP with split beamforming. Through theoretical analysis, simulations, and audible sound field experiments using a uniform linear array, SAIPCP demonstrates a smaller and sharper focal spot than AIPCP in both near and far fields, with reduced side-lobe interference. Notably, SAIPCP achieves a focus sharper than half a wavelength, outperforming AIPCP in precision.

25 - Leveraging learned monocular depth prediction for pose estimation and mapping on unmanned underwater vehicles

Authors: Marco Job, David Botta, Victor Reijgwart, Luca Ebner, Andrej Studer, Roland Siegwart, and Eleni Kelasidi This paper introduces a visual-acoustic fusion framework for real-time localization and 3D mapping in dynamic underwater fish farming environments. It estimates UUV pose relative to fish net pens using monocular vision, combining deep learning-based depth prediction with sparse FFT-derived depth priors, and fuses these with acoustic data for global pose estimation. Predicted depth images are integrated into the wavemap framework to generate real-time 3D maps. Evaluations on industrial fish farm datasets confirm accurate net-relative and global pose estimation, and robust mapping for autonomous inspection and navigation.

24 - Advanced hyperspectral image processing and machine learning approaches for early detection of wheat stem rust

Authors: Alexander Fedotov, Danila Eremenko, Daria Kuznetsova, Olga Baranova, and Anton Terentev This study introduces a preprocessing pipeline for hyperspectral data in plant disease detection, addressing challenges like spectral variability and noise. Evaluated on early stem rust detection in wheat, the pipeline significantly improved classification models (F1-scores rising from 0.67–0.75 to 0.86–0.94) and enabled detection of asymptomatic infections as early as 4 days post-inoculation. The framework’s robustness suggests broader applications in vegetative health monitoring, environmental assessment, and material classification, while offering a scalable solution for integration into unmanned and satellite systems.