From Traditional to Foundation Models: A Survey for Land Use and Land Cover Hyperspectral Image Classification

ABSTRACT Hyperspectral remote sensing image classification is one of the key research areas of the remote sensing community. The high dimensionality, complex structure of data, and availability of fewer training samples hinder classification performance. Traditional machine learning approaches focus mainly on feature extraction for hyperspectral image classification. The complex relationships among pixels, nonlinearity, and material complexity could not be established with these approaches. This results in a suboptimal solution for fewer training samples in hyperspectral images. Recent advances in deep architectures provide means to improve performance and analyze complex patterns effectively, which were challenging with traditional approaches. The present research systematically describes deep learning models, from basic convolutional neural networks to transfer learning, ensemble learning, attention networks and graph nets. Also, advanced transformer approaches such as Mamba architectures, foundation models and vision‐language models for hyperspectral images with a specific emphasis on land use and land cover mapping. These advanced approaches provide efficient classification and real‐time processing capabilities that allow solutions to other different real‐world applications like agriculture, urban mapping, forestry, and the environment. This research also compares key state‐of‐the‐art methodologies, highlights research challenges, and offers future directions for efficient and accurate classification. This review endorses assimilating multisource data, developing lightweight models for resource‐constrained environments, and progressing explainable deep learning frameworks to improve classification performance. This research also serves as a useful reference for researchers in the hyperspectral remote sensing community, supporting the determination of the most appropriate classification technique specific to a particular remote sensing application. This article is categorized under: Technologies > Classification Technologies > Machine Learning Technologies > Artificial Intelligence