Hybrid Deep and Machine Learning Framework for Cloud and Shadow Segmentation in Landsat-8 Imagery

Cloud and shadow interference in satellite imagery reduces the quality and reliability of remote sensing data. The traditional method would face issue to predict data near the shadow and cloud. To address this challenge, this study is focus improve the accuracy the area near shadow and cloud detection in Landsat-8 imagery. The implementation of hybrid module using standard CNN and U-Net CNN and a machine learning model using K-Nearest Neighbors (KNN) on SPARCS and CCA18 Landsat 8 dataset. A hybrid approach was then implemented by integrating CNN outputs and metadata into the second model (KNN/RF), and final evaluation was conducted using accuracy metrics. The research results show that the proposed hybrid deep and machine learning approach improves the accuracy of cloud and shadow segmentation in Landsat-8 imagery. Additionally, the implementation demonstrates that this method can reduce manual effort and computational cost, making it suitable for researchers with limited resources.