Dataset Overview

Dataset Name

• Kaggle‑FIRE‑Dataset

Dataset Source

• From the Kaggle platform, link: Kaggle Fire Data Set

Background

• The dataset was created by a team during the 2018 NASA Space Apps Challenge, with the purpose of developing a model that can recognize images containing fire.

Content

• Classification Task: Binary classification between fire images and non‑fire images.
• Fire Images: 755 outdoor fire images, some containing dense smoke.
• Non‑Fire Images: 244 natural images (forests, trees, grasslands, rivers, people, foggy forests, lakes, animals, roads, waterfalls).
• Class Imbalance: The two classes are imbalanced; it is advised to keep an equal number of images per class in the validation set (e.g., 40 images per class).

Application

• A convolutional neural network was trained to detect the presence of fire in images, achieving an accuracy of 97%.