|
Enhancing Hyperspectral Image Classification Accuracy Using the Artificial Bee Colony Algorithm
|
Ali Asghar Torahi *   |
|
|
|
Abstract: (13 Views) |
Hyperspectral imaging, due to its capability in accurate material discrimination, is considered a powerful tool in Earth sciences and remote sensing. However, the inherent challenge of mixed pixels—arising from the limited spatial resolution of sensors—significantly reduces the accuracy of conventional classification methods. The main objective of this study is to propose and evaluate a novel swarm intelligence–based approach to improve spectral unmixing and consequently enhance classification accuracy. In this context, the performance of the Artificial Bee Colony (ABC) algorithm, as a global optimization method that does not rely on the assumption of pure pixels in the image, was compared with three conventional geometric algorithms: Pixel Purity Index (PPI), N-Finder, and Sequential Maximum Angle Convex Cone (SMACC). The research methodology was implemented on the standard AVIRIS dataset of the Cuprite region, Nevada. After applying essential preprocessing steps, including atmospheric correction (FLAASH) and dimensionality reduction (MNF), the endmembers of key minerals (Alunite, Kaolinite, and Chalcedony) were extracted using all four algorithms. The quality of the extracted endmembers was evaluated through quantitative metrics such as Spectral Angle Mapper (SAM), Euclidean Distance (ED), and Root Mean Square Error (RMSE), in comparison with the USGS spectral library. Finally, image classification was carried out using the SAM algorithm, and classification accuracy was assessed with Overall Accuracy (OA) and the Kappa coefficient. The key findings highlight the superior performance of the ABC algorithm, which achieved an overall classification accuracy of 91.14% and a Kappa coefficient of 78.42%, while the best result among geometric methods (PPI) recorded 86.19% OA and 62.69% Kappa. These results confirm the research hypothesis and demonstrate the high reliability of the ABC algorithm for precise hyperspectral image analysis in real-world environments lacking pure pixels. |
|
| Keywords: Hyperspectral Imaging, Endmember Extraction, Artificial Bee Colony, Spectral Unmixing, Classification Accuracy, AVIRIS Cuprite |
|
|
|
Type of Study: Research |
Subject:
Photo&RS
Received: 2025/09/23 | Accepted: 2026/06/4
|
|
|
|
|
|
|
| Send email to the article author |
|
|
|
| Add your comments about this article |
|
|
|
