Fast cloud detection using low-frequency components of satellite imagery

Being able to rapidly detect clouds in satellite imagery is critical in terms of increasing mission efficiency of ground observation satellites. In this work, a new approach has been proposed for cloud detection involving the utilisation of low-frequency Discrete Wavelet Transform (DWT) components whose sizes are 1/64 of the original image sizes. In the proposed method, several texture features are calculated from the low-frequency DWT components and the cloud pixels are detected by using a K-Nearest Neighbors (KNN) classifier. The proposed method has been tested on the Göktürk-2 images. Since the utilization of low-frequency components leads to a significant loss in detail; the cloud detection performance have decreased to some extent. Nevertheless the obtained results were found to be sufficient for determining the cloudiness rate at a small margin of error, and at approximately 145× increased processing speed.