article · IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
In the world of remote sensing, hyperspectral imaging has emerged as a powerful tool that captures incredibly detailed information about our environment. These images contain hundreds of spectral bands that reveal what the human eye cannot see, making them invaluable for applications ranging from precision agriculture to environmental monitoring. However, to extract insights from complex data requires sophisticated analytical approaches. Our research dives into the performance comparison of two popular machine learning approaches: the Support Vector Machine (SVM) and the more recent deep learning-based Stacked Autoencoder (SAE). We wanted to understand which approach works better under different real-world conditions that researchers and practitioners face. Through extensive experiments across five diverse public hyperspectral datasets, we discovered that the choice between these models isn't straightforward, it depends significantly on your specific circumstances. When labeled data is scarce and that is a common challenge in remote sensing, SVM proves more reliable and efficient. Conversely, when abundant training data is available, SAE demonstrates impressive capabilities in learning complex patterns. One interesting finding was how active learning as a technique that intelligently selects the most informative samples for labeling, improved SAE's performance on medium-sized datasets, potentially offering a practical solution to the data scarcity problem. The proposed approaches showed vulnerability to noise, highlighting the importance of preprocessing steps in real-world applications. Although SVM generally requires less computational resources, SAE's potential to handle large and complex datasets makes it an attractive option when the appropriate computing infrastructure is available. The model training also achieved high accuracy, compared to other models published in the literature. The results achieved provide practical path to researchers and practitioners navigating the complex landscape of hyperspectral image analysis for helping them choose the most suitable approach for their specific constraints and requirements.
This page summarises published work. The authoritative version sits with the publisher.
DOI: 10.1109/jstars.2025.3580654
Is something wrong with this record? Report it or request removal.
Discussion
Have you built on this work, tried to replicate it, or seen it applied in practice? Share what you know. Verified researchers and MARATTO™ domain experts can open a discussion, and any member can reply. Contributions are reviewed before they appear.
No discussion yet. Open the first thread.
New to MARATTO™? Create a free account.