Deep Learning Techniques for Wildfire Investigation with Drones

The forest fire is a wildfire that spreads uncontrollably, causing significant damage to the environment and living beings. It leads to the formation of smoke and fine particulate matter (PM2.5) that contaminates the air. This research has three objectives: To develop a wildfire detection model using deep learning techniques. The YOLO (You Only Look Once) algorithm was used to train the images, aiming for a high level of accuracy with a mean average precision (mAP) of 0.8, which is highly satisfactory. To develop an automated wildfire image detection system by installing Raspberry Pi devices and cameras on drones. This system aims to provide fast and reliable identification of wildfire events. To develop a wildfire alert system by sending detectable data to LINE Open Chat for quick verification of wildfire events, thus reducing damage and building community confidence in safety. The research began by preparing many images of forest fires using the Roboflow platform and training the model using the Google Colab platform. Subsequently, the trained model was installed on Raspberry Pi devices, and a program was written to send notifications to LINE Open Chat. In the testing phase, the model detected approximately 99.448% of the total number of images, with an accuracy of 0.928 for fire detection and 0.803 for smoke detection, resulting in an average mAP of 0.865. Through the utilization of advanced technology and collaboration with relevant agencies, the research aims to develop an effective and efficient system to enhance wildfire management and mitigate its severe consequences.