Automating coastal bioindicator acquisition with deep learning: crab community monitoring for enhanced wetland management

Major biodiversity changes in the Anthropocene demand enhanced monitoring of ecological communities. Notably, community-level attributes of coastal wetland ecosystem engineers, especially crabs (Brachyura), emerge as crucial ecological indicators. However, traditional expert-based surveys for such data remain labor-intensive, time-consuming, and often invasive. This creates an urgent need for an automated and efficient paradigm shift. We developed an RGB sensor-based automated framework and collected extensive image data from 17 coastal mangrove wetlands across China. This framework integrates expert knowledge with artificial intelligence via a pyramid-style annotation approach, utilizing optimized CNN models (YOLOv5/v8 and EfficientNet) for automated image processing and indicator extraction. Test results show that by integrating attention modules and improved anchors, our model achieved superior performance in crab detection, classification, carapace width measurement, biomass estimation, and burrow detection, matching or exceeding manual methods. It further captured plot-level spatial point patterns, addressing limitations of conventional manual surveys. Our local case study validated that image-extracted community metrics provide independent and essential insights for community analysis, offering a more efficient and comprehensive indicator system than traditional methods. Ecologically, this deep learning-integrated novel method provides an economical solution to expand the dimensionality and breadth of biological data. It enhances management effectiveness by (1) serving as foundational hardware-software for in-situ monitoring and automated data collection (scalable to other benthic fauna), and (2) capturing higher-dimensional community indicators and fine-scale spatial patterns to support biodiversity conservation, blue carbon sequestration, and vegetation protection. • Novel AI-powered framework achieves 0.93 mAP 0.5 in automated coastal crab detection and classification. • The automated system measures crab size and estimates biomass with accuracy matching manual methods. • Successfully tested across 17 mangrove wetlands, proving broad applicability in diverse habitats. • Integrates burrow detection and spatial pattern analysis for comprehensive community assessment. • Provides a cost-effective solution for monitoring benthic fauna and enhancing wetland management.