Recent Advancements on Hyperspectral Image Reconstruction from a Compressive Measurement

Hyperspectral (HS) image reconstruction has emerged as a crucial research direction in computational imaging, enabling the retrieval of high-resolution spectral data from compressive snapshot measurements. With the rapid evolution of deep neural networks, HS image reconstruction technique has significantly advanced in both accuracy and efficiency, enabling more precise spectral recovery across many applications. This survey provides a comprehensive review of recent advances in HS image reconstruction strategies, being categorized into traditional model-based approaches, deep learning-based techniques, and hybrid pipelines that integrate data-driven learned prior knowledge with the mathematical formulation of the compressive degradation process. We discuss the theoretical foundations, advantages, and limitations of all three different strategies, highlighting key developments such as sparsity-based reconstruction and low-rank modeling in the model-based methods, convolutional neural network to transformer architecture evolution in the learning-based approaches, and deep unfolding framework in the hybrid pipelines. In addition, we investigate benchmark datasets, evaluation metrics, and key challenges—including spectral distortion, computational efficiency, and generalizability across diverse scenarios—along with potential future trends aimed at addressing these limitations. This survey aims to serve as a valuable resource for researchers and practitioners seeking to advance the state of the art in HS image reconstruction.