LiteNet: A Resource-Efficient Method for Cloud Detection in Remote Sensing Imagery

The accuracy of Earth observation applications is hindered by cloud cover in remote sensing imagery, which presents a formidable challenge. Reliable information extraction from satellite data depends on the prompt and accurate identification of clouds. Artificial intelligence, particularly deep learning techniques, has shown promise in addressing this challenge. However, existing models often prioritize accuracy over simplicity. In response, this paper introduces LiteNet, a neural network based on the UNet architecture but with lower complexity. Developed through meticulous adjustments and informed by Landsat-8 dataset analysis, LiteNet strikes a balance between accuracy and simplicity for cloud detection. By fine-tuning layers and hyperparameters, the model offers an improved solution for cloud identification in remote sensing imagery, crucial for tasks like land cover classification and environmental monitoring. LiteNet’s streamlined design enhances efficiency without sacrificing performance, making it a valuable asset for Earth observation applications.