High Resolution Mapping of Vegetation Biodiversity by Hyperspectral Images and Convolutional Autoencoders

A novel methodology leveraging hyperspectral imaging and convolutional autoencoders successfully maps vegetation biodiversity with high resolution, advancing the assessment of forest diversity using Earth Observation (EO) data. The approach, based on spectral variation (SV), utilizes PRISMA mission data to compute NDVI and identify vegetation-covered pixels, which are then clustered to produce detailed biodiversity maps. Applied to five Southern Italian national parks—including Cilento and Vallo di Diano—the study demonstrates the potential of this technique for large-scale ecological monitoring. This method offers a scalable solution for biodiversity assessment, combining high spectral resolution with advanced machine learning to enhance conservation efforts.

Nico, G., Monaco, M., Masci, O., & Cimino, M. G. C. A. (2024). High Resolution Mapping of Vegetation Biodiversity by Hyperspectral Images and Convolutional Autoencoders. IGARSS 2024 - 2024 IEEE International Geoscience and Remote Sensing Symposium, 2247–2250. https://doi.org/10.1109/igarss53475.2024.10640541