UAV-based super-resolution instance segmentation enables ecological monitoring of sea cucumbers in shallow coral reef habitats.

Accurate monitoring of sea cucumbers from UAV imagery remains challenging because submerged individuals often exhibit low contrast, camouflage against coral-rubble backgrounds, and weak boundary information under water-surface disturbance. In particular, key visual cues such as body contours, surface texture, and papillae-like structures can be severely blurred in aerial images, limiting the reliability of automated recognition. To address these challenges, this study proposes SR-HoloMambaNet, an integrated framework with a direct high-resolution path and a super-resolution-assisted path. Native high-resolution patches can be processed directly by HoloMambaNet, while resolution-limited patches are reconstructed by MambaIR at a 4× magnification factor before entering the same instance-segmentation network. HoloMambaNet incorporates VSSBlock, CSwinTR, and Large Separable Kernel Attention (LSKA) to improve long-range dependency modeling, multi-scale feature representation, and segmentation under camouflage and complex seabed backgrounds. Experimental results showed that SRR substantially improved the downstream segmentation performance of degraded images, increasing the mean average precision (mAP@50) from 63.7% with bicubic interpolation to 82.7% with MambaIR reconstruction. On high-quality UAV images, the proposed HoloMambaNet achieved 93.7% precision, 82.0% recall, and 95.7% mAP@50 with only 4.10 M parameters and an 8.7 MB model size, outperforming mainstream instance-segmentation models such as Mask R-CNN, RT-DETR, and YOLO-series baselines. Under a frozen-source zero-shot protocol on five public external datasets, HoloMambaNet achieved the highest box mAP@50 on each dataset and a macro-average of 41.1%. The expert-verified annotation masks were also used to derive annotation-based reference summaries of patch-scale abundance, image-space apparent cover, morphology-related proxies, and local spatial arrangement. These results indicate that SR-HoloMambaNet provides a practical and lightweight solution for UAV-based shallow-water sea cucumber monitoring, with potential applications in large-scale benthic resource assessment and coral reef ecological conservation.