Multi-Sensor Satellite Image Cloud Classification with Deep Learning

Advancements in remote sensing (RS) have greatly increased the availability of satellite imagery, enabling diverse applications. However, cloud-cover remains a significant challenge, obstructing the acquisition and analysis of optical satellite images. This necessitates advanced cloud classification techniques as a crucial preprocessing step. The proposed MobileNetV2-based convolutional neural network architecture demonstrates unprecedented cloud classification capabilities, effectively processing imagery across diverse spatial resolutions ranging from 10 to 56 meters. By leveraging sophisticated machine learning algorithms, the model transcends traditional pixel-based thresholding techniques, capturing complex inter-pixel spatial correlations with remarkable precision. Experimental validation across heterogeneous datasets including IRS (AWiFS, LISS-III), Landsat, and Sentinel platforms reveals exceptional performance metrics, consistently achieving classification accuracies exceeding 99.94%.