Land Cover Segmentation using DeepLabV3 and ResNet50

Land cover segmentation has a great importance in various fields, including remote sensing, environmental monitoring, urban planning, agriculture, and natural resource management. It involves a division process of a landscape or region into different classes or categories with respect to the type of land cover in each place. With the recent developments in remote sensing area, high-resolution satellite images can be simply acquired. For an efficient land cover segmentation, in this study, a hybrid approach using deep learning architectures DeepLabV3 and ResNet34 is proposed. The proposed method has been trained and tested using the LandCover AI dataset. As a result, 88.2% F1-score value was obtained with the proposed hybrid approach.