Machine Learning for Cloud Cover Detection Using Multispectral Satellite Images

Reliable cloud detection in satellite based remote sensing applications is a vital pre-processing step. It can be viewed as a classification technique by partitioning objective pixels into cloud or non-cloud shadow classes. However, certain cloud pixels, particularly narrow pixels in the cloud, can be regarded as a blend of cloud reflection and land-based objects. Hence, it is difficult to classify them using a single classifier. In the present study, a technique based on an ensemble classifier is proposed to identify these pixels. The proposed ensemble classifier uses the Extreme Learning Machine (ELM), Naïve Byes (NB), and K-Nearest Neighbour (KNN) classifier. The model is trained and evaluated using the Spatial Procedures for Automated Removal of Cloud and Shadow (SPARCS) datasets. The findings of the analysis indicate that the proposed classifier provides better classification accuracy. The performance of the model other than RGB band is also analyze and results shows that adding images from different bands significantly improves the results.