Forest Monitoring System for Early Fire Detection Based on Convolutional Neural Network and UAV imagery

Forest fires are one of the main reasons for environmental degradation. In their early stages, the fires are hard to discover, so a faster and more accurate detection method can help minimize the amount of damage they can inflict. In this paper, we present an approach for autonomous early fire detection, which is based on a system with high degree of reliability and with no need of service or human interaction. To provide the autonomous capabilities to the proposed system, we have developed an object detection method, based on a convolutional neural network, which is presented in the main part of the paper. In order to have a better field of view over the observed area, instead of traditional lookout towers and satellite based monitoring, we use live video feed from an unmanned aerial vehicle (UAV), which patrols over the risky area. To make better predictions on the fire probability, we use not only the optical camera of the UAV, but also an on-board thermal camera. With the help of the software platform Node-RED, we have developed a web-based platform, which can present the acquired data in real-time and can notify the interested parties. The workflow for the development of the web-platform is also described in this paper.