A novel image fusion method based on UAV and Sentinel-2 for environmental monitoring.

In recent years, with the rapid development of remote sensing technology, environmental monitoring in mining areas using remote sensing imagery has gained increasing attention. Due to the small scale of mining areas, the resolution of satellite remote sensing imagery is insufficient for detailed monitoring needs. UAV remote sensing imagery provides high resolution, but its monitoring range is limited and lacks access to historical data. Furthermore, effectively fusing multi-source data with disparate spatial-temporal characteristics to accurately capture the complex dynamic changes in mining areas remains a key methodological challenge.To address this, this study, utilizing UAV remote sensing imagery and Sentinel-2 satellite imagery acquired on September 5, 2023, from the Erlintu mining area, proposes a novel fusion method aimed at achieving small-scale, long-term environmental monitoring in mining areas.First, the spatial resolution of both UAV and Sentinel-2 imagery is resampled to 0.1 m. Second, a two-layer preprocessing approach is applied to enhance data quality. Third, a stacked inversion model based on an ensemble learning framework is developed. Finally, using high-resolution UAV imagery as the reference, and original, resampled, and model-inverted Sentinel-2 imagery as experimental values, accuracy is assessed and analyzed with Mean Absolute Percentage Error (MAPE) as the metric. Results demonstrate that the stacked learning model, combined with cubic convolution resampling, reduces the MAPE of NDVI values between Sentinel-2 and UAV imagery from 54.31 to 10.01%, markedly improving accuracy. This study further uncovers the synergistic effect of resampling techniques and model architecture, offering reliable data support for small-scale, long-term environmental monitoring in mining areas.