Most of the current smoke detection techniques are developed using CCTV images. To identify the wildfire early, it can be useful to deploy satellite imagery and develop models that can recognize smoke in forest areas. However, very few labelled satellite image datasets are available to build the wildfire smoke detection model. In order to find a solution to this problem, a dataset consisting 23, 644 satellite images was gathered. The dataset is divided into four categories: smoke, smoke with fog, non-smoke, and non-smoke with fog.