RNFS: a robust nature-inspired feature selection for remote sensing image classification

Remote sensing data is characterized by its complexity and high-dimensional features. Therefore, developing effective feature selection techniques is crucial for minimizing redundancy, boosting computational performance, and achieving high accuracy. Nature-inspired optimization algorithms demonstrate a remarkable ability to efficiently select the most discriminant features. This article presents a novel nature-inspired hybrid metaheuristic algorithm (RNFS) for feature selection that effectively combines the exploitation strength of the Fox Optimization Algorithm (FOA) with the exploitation capacity of the Whale Optimization Algorithm (WOA). The proposed framework aims to improve classification precision and sensitivity while reducing overfitting. The datasets, including WHU-RS19, RSSCN7, and UCM, are used to compare the efficacy of RNFS with FOA and WOA. Experiments are conducted in a MATLAB-based environment, utilizing performance metrics such as accuracy, precision, sensitivity, specificity, and F1-score across 60 trials. The models evaluated included the wide neural network (WNN), cosine k-nearest neighbors (CKNN), and quadratic support vector machine (QSVM). Results demonstrate that the RNFS outperforms both FOA and WOA in all metrics for the WNN, CKNN, and QSVM classifiers, indicating an improvement in the true positive rate by reducing false positives. RNFS achieved the highest testing accuracy and exhibited robust classification performance, confirming its effectiveness for remote sensing applications.