Title : Kernel to computation: Identifying optimal feature set for edible nut classification
Abstract:
Efficient edible nut classification relies on transforming raw kernel attributes into a compact and discriminative computational feature set. This study presents a systematic approach to identifying an optimal feature subset that enhances classification accuracy while maintaining computational efficiency. High‑resolution nut images are first preprocessed through segmentation, contour detection, and normalization to isolate individual kernels. From these, a broad set of candidate features—including morphological characteristics (size, aspect ratio, sphericity), color descriptors (RGB/HSV histograms), and texture measures derived from gray‑level co‑occurrence matrices and local binary patterns—is extracted. To reduce redundancy and highlight the most relevant attributes, feature selection methods such as correlation filtering, mutual information analysis, and wrapper‑based recursive elimination are employed. The refined feature subset is then evaluated using machine learning classifiers, including support vector machines, random forests, and lightweight neural networks. Experimental results indicate that an optimized feature set significantly improves model generalization and reduces processing time. This kernel‑to‑computation workflow provides a scalable foundation for automated edible nut classification in agricultural and industrial applications.
