Efficient Wildfire Detection Framework Based on Artificial Intelligence Using Convolutional Neural Network and Multi-Color Filtering

Wildfire prevails to be a high-risk natural disaster, posing serious damage to both human populations and the environment. As a nation with one of the largest forested regions, Indonesia faces the recurring challenge of wildfires, which annually rank among its top concerns. The significant impact of wildfires on the environment, development, and economic growth motivates this research to create a detection model based on deep learning methods. Through utilizing surveillance tools with the likes of CCTV, UAVs, and satellites, the proposed model aims to pinpoint the precise locations of fire incidents in images, achieved through an Internet of Things device-driven process. This will enable a more efficient and effective response in controlling forest fires, as well as supporting sustainable development. By employing a Convolutional Neural Network-based model using MobileNetV2 with additional fully connected layers for wildfire event classification in images, as well as multi-color filtering for segmenting fire images, the proposed model yielded impressive results. It achieved a remarkable 98.95% accuracy in classifying wildfire images and an Intersection over Union (IoU) score of 0.37 for accurately segmenting specific fire images. These outcomes surpass previous research, underscoring the proposed model's capacity to effectively detect wildfire presence in images and accurately delineate fire boundaries. The proposed model is envisioned to be seamlessly integrated into a system capable of providing real-time information on wildfire locations, serving as an effective mitigation and response solution.