Deep learning versus traditional artificial intelligence algorithms for remote sensing applications

Remote sensing is an important mechanism for monitoring and oversight the environment, natural resources, and human activities. In this research, the focus was on utilizing deep learning (DL) algorithms, specifically a Convolutional Neural Network (CNNs) model with a U-Net architecture, for the classification of satellite image and because of there are multiple and extensive research contributions to the study of traditional artificial intelligence approaches in the field of remote sensing, this study was limited to a comprehensive review of some previous studies in this field. As for DL, a U-Net model was developed, utilized and it was trained on Sentinel-2 satellite image have a spatial resolution of 10 meters. The training stage was centered on data covering the Egypt especially the Delta region and the Nile Valley due to their agricultural and environmental importance. The model achieved a classification and segmentation accuracy of 94.14% and an Jaccard coefficient (IOU) value of 87.64%. These results confirm the strong potential of DL models, such as U-Net that resulted high accuracy in satellite image classification and segmentation tasks, especially when used specific geographical regions with distinctive characteristics.