Performance Evaluation of Support Vector Machine and Stacked Autoencoder for Hyperspectral Image Analysis

In remote sensing, hyperspectral imaging has emerged as a powerful tool, capturing hundreds of spectral bands that reveal information beyond human vision. These images are invaluable for applications such as precision agriculture and environmental monitoring. However, extracting meaningful insights from hyperspectral data requires advanced analytical techniques. This research compares two popular machine learning approaches for hyperspectral image classification: support vector machines (SVM) and deep learning-based stacked autoencoders (SAE). Our goal was to evaluate their performance under diverse real-world conditions. Extensive experiments were conducted on five public hyperspectral datasets, revealing that model effectiveness depends on specific circumstances. When labeled data are limited, SVM proves more reliable and efficient. Conversely, SAE excels with abundant training data due to its ability to learn complex patterns. Incorporating active learning to select the most informative samples further enhances SAE performance on medium-sized datasets, offering a solution to data scarcity. Both methods showed sensitivity to noise, emphasizing the importance of preprocessing. While SVM requires fewer computational resources, SAE is advantageous for handling large, complex datasets with adequate infrastructure. Overall, the proposed approaches achieved high accuracy compared to existing models, providing practical guidance for researchers and practitioners in selecting suitable methods for hyperspectral image analysis based on dataset size, computational capacity, and labeling constraints.

Keywords: Hyperspectral imaging, Remote sensing, Support vector machine (SVM), Stacked autoencoder (SAE), Deep learning, Active learning, Image classification.