Wildfire Detection from Sentinel Imagery Using Convolutional Neural Network (CNN)

Abstract

Wildfires are a significant threat to the environment and human life, and early detection is crucial for effective wildfire management. The aim of this research work is to develop a deep learning model using CNN method for detecting wildfire using satellite imagery. The CNN model development process involves splitting the dataset into training and testing sets, pre-processing the data using ImageDataGenerator function, building a deep learning model using Sequential function, compiling the model using Adam optimizer, categorical cross-entropy loss function, and accuracy metric, and training the model using the fit generator function. The CNN model architecture includes Conv2D, MaxPooling2D, and Dense layers.

The dataset was collected from Sentinel-2 L1C, including 159 fire images and 149 non-fire images from different places in the Mediterranean region of Turkey. The CNN model was developed using the Keras library and trained for 200 epochs using the Adam optimizer. The model achieved an accuracy of 92.5% and a loss of 0.22 on the test set, outperforming existing methods for wildfire detection. The research work contributes to the field of wildfire science and management by providing a deep learning CNN detection model that can accurately predict wildfire behavior using satellite imagery. The outcomes of the research work can enable more effective and efficient wildfire mitigation efforts, improving the safety and well-being of communities affected by wildfires.