Ecological monitoring of invasive species through deep learning-based object detection

Water hyacinth (Eichhornia crassipes) is a highly invasive aquatic species that poses significant threats to aquatic ecosystem health and water resource sustainability. Monitoring its spatial distribution offers a valuable ecological indicator for assessing environmental degradation and guiding ecological interventions. However, traditional monitoring methods, such as manual inspection and remote sensing, often struggle with limited scalability, accuracy, and adaptability in complex natural environments. This study explores a novel ecological monitoring approach that uses the spatial distribution of water hyacinth as an indicator of aquatic ecosystem health, leveraging real-time object detection techniques. To implement this, we propose an enhanced YOLO-based model: HydroSpot-YOLO, which integrates an Attentional Scale Sequence Fusion (ASF) mechanism and a P2 detection layer to improve the detection of small and densely clustered targets under challenging conditions such as water reflections, cluttered backgrounds, and variable illumination. A specialized dataset, curated from real-world surveillance footage, was used for model training and validation. To support experimental validation, a specialized dataset was constructed from real-world aquatic surveillance footage, encompassing diverse and visually complex environments. Experimental results demonstrate that the improved model consistently outperforms existing baselines in terms of precision, recall, and mean Average Precision (mAP). These findings confirm the feasibility and effectiveness of applying deep learning-based object detection as an ecological indicator monitoring approach, offering a scalable and automated solution for invasive species management and aquatic ecosystem assessment.