Improving Deep Machine Learning for Early Wildfire Detection from Forest Sensory Images

Wildfires cause irreversible damage, prompting proactive strategies for management and early detection. This research paper aims to utilize a deep-learning model to discern early-stage wildfires in forested regions. Due to the limitations of current public wildfire detection datasets for early fire detection, we prepared a more specialized dataset for the early detection of wildfires in their emerging stages. Using our dataset with a deep machine learning model implemented in TensorFlow and Keras, we are able to effectively detect active fire spots in images captured by satellite and/or surveillance cameras with an accuracy of 86%. The paper meticulously evaluates key classification metrics, including accuracy, precision, recall, and f1 values of our Convolutional Neural Network model. By providing a comprehensive analysis, the research contributes to the advancement of effective early wildfire detection, offering valuable insights for cities and countries grappling with the threats posed by these devastating natural occurrences.