Advancing Hyperspectral Image Classification with Deep Features Learning and Evolutionary Algorithms

Abstract. Hyperspectral Images (HSI) reveal the secrets of land cover at a granular level, capturing hundreds of narrow spectral bands rich in detailed information. However, the sheer dimensionality of these data poses significant challenges to traditional Machine Learning (ML) methods. This research tackles the high-dimensional challenge of HSI classification with an advanced hybrid framework, leveraging the power of Deep Learning (DL), ML and Evolutionary Algorithms (EA) to conquer this challenge and achieve accurate HSI classification. We unleash the data's inherent wisdom via deep Features Extraction (FE) and optimize the representation through EA. Experiments on the Hyperion Earth Observation-1 (EO-1) show that our approach outperforms state-of-the-art ML based methods in analyzing Earth's diverse landscapes. In addition, the experiments conducted on simulated benchmarks validate the superior performance of the proposed approach compared to the baseline ML model in terms of prediction accuracy and F1-score.