LAUSR

ABSTRACT Remote sensing scene categorization (RSSC) is a long-standing, vital, and complex issue in computer vision. It seeks to classify a scene into one of the predetermined scene groups by analysing the entire image. The rise of large-scale datasets and the resurgence of deep learning-based methods, which directly learn potent feature representations from large amounts of raw data, have led to a lot of progress in representing and classifying RS scenes. Convolutional neural networks (CNN) are among the varieties of deep neural networks that have been the subject of the most research. Taking advantage of the swift increase in the amount of labelled samples and the major enhancements in the strength of processing units, CNNs research has advanced swiftly, producing state-of-the-art results on a number of applications. In this overview, we present a comprehensive evaluation of earlier published surveys and recent CNN-based approaches for RSSC. This study covers more than 100 significant works on scene categorization, including problems, benchmark datasets, and qualitative performance evaluation. In view of the results so far, this study concludes with a list of intriguing research opportunities.