2025/11/01 by Zhiyi Tu, Peng Zhou, Dongsheng Zhang +4 · 1 voice
Environmental Science · #Coral and Marine Ecosystems Studies #Ichthyology and Marine Biology #Marine animal studies overview
paper · doi:10.1093/icesjms/fsaf207
openalex publication_date 2025/11/01 · openalex created_date 2025/11/20 · openalex updated_date 2026/07/29
Abstract Marine megafauna are essential ecological indicators, and underwater video surveys produce large, diverse image collections that are challenging to process automatically. We introduce MBMOT, a curated dataset of 50 deep-sea video sequences (over 23 000 frames) with rich biodiversity annotations, and SMART, an intelligent system for marine megafauna recognition and tracking, which integrates the detection, classification, tracking, and counting functionalities for marine megafauna. With a designed small-target detection module enabling shallow-to-deep feature interaction, SMART achieves a mean average precision (mAP50 - 95) of 63.0%, representing an 1.2% improvement over the standard YOLOv11, and fast processing speed (108.9 ms per frame) when paired with a lightweight DeepSORT tracker. SMART provides an expandable platform for deep-sea resource assessment, ecological monitoring, and behavioral analysis.