Neural network for remote sensing
Remote sensing technologies, like satellite, airplane and drone images, produce a large amount of data that can help monitor forest canopy health and detect pest damage across vast areas. However, analysing and interpreting such data volumes manually is slow and requires expert knowledge. In FORSAID, we use deep learning-powered computer vision to speed up and improve this process.
These deep learning models are trained with images where pest-affected trees have been identified through field surveys or insect traps. By learning spectral and textural patterns in these images, the models can automatically detect similar occurrences in images of other areas. This enables us to find areas at risk or already affected, even in large and remote forest areas, and monitor changes in forest health over time.
Our goal is to improve early detection of pests in key tree species like oaks, pines and spruces, helping forest managers take timely actions to reduce damage, boost forest resilience and maintain ecosystem services.
Leveraging remote sensing coupled with computer vision, we aim to automate the production of large-area, yet spatially detailed, cover maps to address three main questions:
- Where are the key tree species located?
- Where and when has pest damage occurred?
- Where are future pest outbreaks most likely to occur?
Neural networks for remote sensing in FORSAID will complement other tools, like automated insect traps and environmental DNA, creating a comprehensive and scalable forest monitoring system.
Forest surveillance with artificial intelligence and digital technologies
Ongoing | 2024-2028
- Main funding source
- Horizon Europe (EU Research and Innovation Programme)
- Geographical location
- Italy, Slovenia, France, Portugal, Germany, Sweden, Bulgaria, Denmark, Ukraine, Switzerland