Introduction
This special issue of the IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing (JSTARS) presents a curated collection of 32 articles that represent the current state-of-the-art in hyperspectral image analysis. These papers originated from the 9th Workshop on Hyperspectral Image and Signal Processing: Evolution in Remote Sensing (WHISPERS), held in Amsterdam in September 2018. The collection highlights a significant transition in the field: the movement from theoretical signal processing toward mature, AI-driven operational applications.
The Hyperspectral Challenge
Hyperspectral imaging (HSI), or imaging spectroscopy, captures a dense range of the electromagnetic spectrum for every pixel in an image. This provides a unique "spectral fingerprint" for materials, enabling precise identification. However, the high dimensionality and sheer volume of HSI data create substantial processing bottlenecks. Historically, research focused on fundamental tasks such as pixel classification, spectral unmixing (decomposing pixels containing multiple materials), and dimensionality reduction to manage this complexity.
Methodological Shifts: The Rise of AI
The 32 selected articles demonstrate a clear shift in methodology. While traditional pillars like sparse representation and spectral-spatial methods remain relevant, Artificial Intelligence and Deep Learning have emerged as critical drivers. These techniques are pushing hyperspectral analysis beyond traditional statistical limits, moving the field into a new era of automated feature extraction.
Furthermore, the integration of High-Performance Computing (HPC), particularly using GPUs, is becoming standard. This hardware acceleration is essential for meeting the real-time operational requirements of industrial applications, ensuring that complex deep learning models can be deployed effectively in the field by non-experts.
Academic-Industrial Convergence
A defining feature of the 9th WHISPERS workshop was its co-location with SpectroExpo, a professional tradeshow. This partnership signifies a maturing ecosystem where academic tools are finally aligning with industrial needs. The event brought together over 330 participants, including sensor designers, service providers, and researchers.
This convergence is driving "applications-oriented" research. Instead of incremental improvements in algorithmic accuracy, the focus is shifting toward solving real-world problems such as landslide detection, the identification of illicit drug crops, mineral mapping, and environmental resource management.
Conclusion and Future Outlook
The editors conclude that the hyperspectral pipeline—from sensor design to end-user software—has reached industrial viability. Future work will likely see deeper integration of AI into the workflow and a continued expansion of application-specific research. The bridge between academia and industry, strengthened by joint meetings like WHISPERS, remains vital for the continued evolution of remote sensing technologies.
Summary updated