Context and Justification
Forest fires represent a critical threat to ecosystems in Ecuador's Andean and Amazonian regions, with direct impacts on biodiversity, water resources, and rural communities. Late detection is one of the factors that most amplifies damage: reducing response time from hours to minutes can save thousands of hectares of native vegetation.
UAVs equipped with thermal sensors and infrared cameras offer an ideal platform for autonomous patrolling and early detection of heat sources in hard-to-reach areas. This project integrates computer vision, AI-based classification, and real-time communication to build an alert system that operates autonomously over high-risk forested areas.
Main Research Goal
Specific Scope
- Select and integrate a lightweight thermal and infrared sensor in a multirotor UAV platform with a minimum flight autonomy of 25 minutes.
- Design and implement a thermal image classification algorithm based on convolutional neural networks (CNN) to discriminate heat sources of anthropogenic or natural origin.
- Develop an automatic georeferencing module that associates each detection with precise GPS coordinates for risk map generation.
- Validate the system through controlled detection tests over calibrated heat sources and patrol missions in rural areas of Pichincha province.
Research Products
UAV prototype with integrated thermal sensor, embedded processing system, and real-time data transmission capability to base station.
Trained and validated convolutional neural network model for thermal image classification with true positive rate above 92%.
Georeferenced alert system with cartographic visualization interface for identification and tracking of detected fire hotspots.
Experimental validation report with classifier performance metrics and territorial coverage analysis per patrol mission.