Airborne Data Monitoring
This page provides resources and information about using machine learning for airborne data monitoring. Airborne data includes imagery and other sensor data collected from drones, aircraft, and satellites, providing valuable information for biodiversity monitoring at various spatial scales.
Esta página proporciona recursos e información sobre el uso del aprendizaje automático para el monitoreo de datos aéreos. Los datos aéreos incluyen imágenes y otros datos de sensores recopilados de drones, aeronaves y satélites, proporcionando información valiosa para el monitoreo de la biodiversidad en varias escalas espaciales.
Summary
Machine learning for airborne data analysis largely focuses on two main tasks:
- Scene classification and segmentation
- Object detection
Example of airborne imagery with annotations for biodiversity monitoring
Source: Xu et al. (2024) Machine learning for biodiversity monitoring: A review
Data
Several biodiversity datasets are available for training and testing machine learning models:
- LILA.Science Datasets - A comprehensive collection of biodiversity datasets
- Biodiversity Package Dataset Compilation - A curated list of known biodiversity datasets
Common Challenges and Solutions
Key challenges in airborne data analysis include:
- Changes in background and image resolution
- Small object sizes
- Fine-grained differences in object classes
- Large data sizes
- Rare objects
Existing Tools
Several specialized tools are available for airborne and satellite data analysis:
- DeepForest - A Python package for deep learning on forest imagery
- TorchGeo - A PyTorch library for geospatial data (primarily satellite)
- Terratorch - A PyTorch library for satellite imagery analysis
Worked Example Notebook
Below is a video tutorial demonstrating how to use DeepForest for airborne biodiversity prediction:
This video provides a comprehensive walkthrough of using DeepForest, a Python package for deep learning on forest imagery, to predict biodiversity from airborne imagery. The tutorial covers data preparation, model training, and evaluation of results.
You can also explore the worked example notebook directly: DeepForest Example Notebook.