Intelligent monitoring of coastal outfalls via multi-source remote sensing image fusion.

Coastal outfalls are major point sources of pollutants entering the marine environment, severely threatening coastal water quality and marine ecosystems. Effective monitoring of these outfalls is therefore essential for pollution control and ecosystem protection. Existing monitoring approaches face significant limitations: manual inspections are inefficient and labor-intensive, visible light UAV imagery provides sufficient spatial resolution but lacks the spectral depth required for precise material identification, and multispectral satellite data has insufficient spatial resolution. Remote sensing images provide critical information for outfall detection. We propose an intelligent recognition method that integrates multi-source image fusion with a deep learning object detection model. Using unmanned aerial vehicles, we acquired 1657 high-resolution multispectral images, classified into four morphological types and annotated. Six fusion algorithms generated enhanced datasets, evaluated with multiple quality metrics. The model was trained on each dataset, with performances compared to identify the optimal fusion approach for improved accuracy, robustness, and efficiency. This method effectively advances deep learning-based coastal outfall recognition and supports intelligent marine environmental monitoring.