Intelligent Multimodal Fire Detection System for Drone-based Forest Surveillance

The forest fire causes many threats to wildlife, natural habitats, and humans, as it can also damage the economy . Timely mitigations are needed to solve this issue rather than to avoid major disasters. This paper presents a multi modal approach that uses the fusion of RGB and thermal images to build an efficient fire detection system . The object detection algorithms such as EfficientDet, CenterNet, RetinaNet, FireDetXplainer, ShuffleNet, and Faster R-CNN are compared with YOLO in terms of accuracy, speed, and effectiveness in environmental conditions. Using this, the proposed system was developed with optimized YOLOv12 integrating a modified FireDetXplainer for drone usage. This model provides major advantages in deployment to the real-time environment, as it shows greater efficiency in detecting fires in forest regions. This model can be used to monitor fire incidents, and a timely alert is sent to mobile devices. The alert system is capable of providing specific geo-location, fire intensity, and predictive analytics. This can be helpful for emergency teams to take accurate, timely actions. The experiment results for the proposed model are evaluated, as it achieves greater accuracy and precision in detecting and localizing specific regions, making it a reliable solution for wildlife management.