Abstract
Due to the increasing demand for hyperspectral image (HSI) classification, there is a need for improvements and enhancements to achieve more accurate and cost-effective results. Image processing plays a significant role in HSI classification, primarily used for image smoothing and denoising. Filtering, a popular method in image processing, is typically based on mathematical equations. However, in this study, filtering is treated as an optimization problem to provide a novel filter for HSI processing and classification. An optimized filter (OF) was generated and optimized using genetic algorithm (GA) based on the Pavia University (PU) dataset, which preprocessed using Minimum Noise Fraction (MNF). Subsequently, the OF was applied to HSI classification for three datasets using Extreme Gradient Boosting (XGB). The results were compared with median filter (MF) and Gaussian filter (GF). The findings demonstrated that, in comparison to MF and GF, OF exhibited the strongest enhancement and achieved the highest accuracy in most situations, including different sampling scenarios for various datasets. Moreover, OF demonstrated excellent performance in aiding HSI classification, especially in classes with a higher number of samples. The study's outcomes highlight the feasibility of generating a filter specifically for HSI processing and classification using GA, which is deemed acceptable and effective. Based on the results, filtering has evolved into an optimization problem, expanding beyond being solely a mathematical problem. Filters can now be generated and optimized based on the goals and requirements of image-related tasks, extending beyond HSI applications.