Satellite Image Wildfire Detection Using Convolutional Neural Networks (CNN) With Kaggle Wildfire Prediction Dataset

Abstract

Wildfires threaten human lives, property, and the environment, requiring advanced early detection and mitigation methods. This research work employs the use of Convolutional Neural Networks (CNNs) and remote sensors to create an early warning system for wildfires. It stresses the need for immediate action to mitigate destructive effects. In recent years, several tactics have been developed in artificial intelligence (AI), especially in CNNs, enhancing image classification performance. These advancements are utilized in this research to develop a CNN-based model for analyzing images relevant to wildfire incidents to achieve better results than current methods. The study employs the Kaggle Wildfire Prediction Dataset, which includes satellite images distinguishing regions that have experienced wildfires. This dataset is useful for detecting historical trends to train models. The model demonstrates satisfactory learning, as by the final test accuracy of 93.76%. However, the deep model takes more time to be trained than a shallow model due to its size and dimensionality, requiring a training period of 152-170 seconds. Accumulating data from Kaggle can improve the system for detecting wildfires, helping to prevent them more efficiently. To achieve this goal, it is possible to use CNN-based image processing and integrate it with remote sensing data, thus achieving improved and accurate wildfire identification and monitoring. The system will leverage Telegram to notify users once a wildfire is detected in the area, ensuring timely information is received.