A Two-Stage Approach Combining Feature Selection and Construction for Hyperspectral Crop Classification

Hyperspectral image classification is effective for obtaining high-precision crop distribution maps. However, learning discriminative features from hyperspectral images to improve classification accuracy remains a highly focused research topic. Existing methods often fail to fully integrate feature selection and feature construction, which may result in learned feature sets with lower information levels and increase the risk of overfitting. To address these issues, this paper proposes a two-stage approach that combines feature selection and feature construction for hyperspectral image classification. In the first stage, feature selection is used to remove redundant features and obtain a high-quality feature subset. In the second stage, feature construction is employed to generate more discriminative high-level features, aiming to find the optimal combination of high-level and original features. Comparative experiments on four hyperspectral image datasets demonstrate that the proposed approach significantly outperforms nine baseline methods in crop classification tasks.