LULC Image Classification with Convolutional Neural Network

The topic of land use and land cover classification (LULC) has attracted the interest of many researchers in recent times. A variety of techniques have been proposed for LULC and while some of them are semantic segmentation-based, others are classifying an entire image to determine its class. The semantic segmentation approaches label objects as members of a class by assigning a different colour to each class. In this work, we investigate class heterogeneity, which so far, to the best of our knowledge, has not been explored in LULC or scene classification. We carefully cluster the 21 classes of the UC Merced dataset into four superclasses based on their textural, spectral, or structural similarities and use the dataset to test the performance of our model. We also demonstrate the efficiency and accuracy of our deep learning approach, reporting $a$ superior performance of our model in terms of Accuracy, Precision, Recall, and F1 score.