Mar Environ Res 2025 Sep 10;212:107510. doi: 10.1016/j.marenvres.2025.107510.

A novel underwater Holothurians monitoring system using consumer-grade amphibious UAV with Mamba-based Super-Resolution Reconstruction and enhanced YOLOv10.

???displayArticle.abstract???
Holothurians (commonly known as sea cucumbers) have economic and ecological value, making their monitoring essential for understanding ecosystem status and decision-making in ecological conservation and fishery management. Traditional monitoring of Holothurians has primarily relied on in situ visual census conducted by divers, which is labor-intensive, time-consuming, and often limited in spatial coverage. In recent years, underwater video and photographic surveys have emerged as supplementary tools, offering the potential for broader-scale and digital documentation. However, these imaging-based techniques remain constrained by limitations in image quality and resolution, often rely on outdated algorithms, lack ecological specificity, and cannot generate georeferenced orthophoto maps enriched with visualized demographic parameters. High-quality underwater images are essential for identifying species-specific morphological characteristics of Holothurians in video- or photo-based monitoring. However, the complexity of underwater environments poses significant challenges in acquiring such images. These limitations hinder accurate species identification and quantitative analysis, especially when relying on automated image analysis. Therefore, super-resolution reconstruction is necessary to enhance image clarity and detail, enabling more reliable mapping and demographic monitoring. Based on a field survey conducted in Koh Tao, Thailand, we developed a novel Holothurians monitoring system that integrates a custom-designed consumer-grade amphibious unoccupied aerial vehicle (AUAV), a Mamba-based super-resolution reconstruction technique, and an improved instance segmentation model (YOLOv10-SEA). Super-resolution images reconstructed by the MambaIR model were used as input for the YOLOv10-SEA model to perform instance segmentation of Holothurians. Among seven super-resolution methods evaluated (Bicubic, SRCNN, EDSR, RCAN, SwinIR, Real-ESRGAN, and MambaIR), MambaIR achieved the highest quantitative (PSNR: 45.58 dB, SSIM: 97.95%) and qualitative image quality. Instance segmentation on MambaIR-generated super-resolution images achieved a mean Average Precision at IoU 0.5 (mAP50) of 99.9%, using an optimal magnification factor of 4. This was achieved using YOLOv10-SEA, a lightweight model modified from YOLOv10 with three architectural changes, which improved mAP50 by 6.1% while keeping the model compact (16.2 MB). Segmentation was performed on individual super-resolution images, and the resulting outputs were subsequently used to generate orthomosaic habitat maps and extract demographic parameters, revealing localized distribution patterns and a stable size structure of holothurians. These findings demonstrate that the proposed system enables efficient and accurate monitoring of Holothurians, while also providing a pipeline to monitor other underwater objects, thereby benefiting conservation communities and fishery resource managers.

???displayArticle.pubmedLink??? 40939273
???displayArticle.link??? Mar Environ Res