Dataset for Forest Fire Detection

We have curated a dataset to address the problem of forest fire detection. All images in the dataset are 3-channeled with spatial resolution of 250 × 250. We have retrieved these images by searching various search terms in multiple search engines. Afterwards, we thoroughly investigated these images to crop and remove the inappropriate components such as people, fire-extinguishing machinery etc. We ensured that each image only contain the relevant fire region. The dataset is designed for binary problem of fire or no-fire detection in the forests landscape. It is a balanced dataset consisting of 1900 images in total, where 950 images belong to each class. We have divided the dataset into 80:20 for training and testing purposes in our study.

我们精心构建了一款面向森林火灾探测任务的数据集。数据集中所有图像均为3通道（3-channel）格式，空间分辨率为250×250像素。我们通过多搜索引擎检索多种搜索关键词获取这批图像，随后对其进行全面审核，裁剪并移除了人员、灭火设备等无关元素，确保每张图像仅包含与火灾相关的区域。本数据集专为森林场景下的二分类火灾探测任务设计，属于平衡数据集，总计包含1900张图像，两类样本各950张。在本研究中，我们将数据集按照80:20的比例划分为训练集与测试集。