A multimodal feature extraction and hybrid selection framework for image-based tea leaf fermentation classification
C. M. Sulaikha1 and A. Somasundaram2
Assistant Professor, Department of Computer Science,Sri Krishna Arts and Science College, Coimbatore,Tamil Nadu,India2
Corresponding Author : C. M. Sulaikha
Recieved : 08-February-2025; Revised : 14-December-2025; Accepted : 18-December-2025
Abstract
Tea fermentation classification aims to categorize tea leaves according to their fermentation states using image analysis techniques, which often involve handling increasingly complex datasets. Conventional classification models frequently underperform due to their limited ability to effectively identify and exploit the most discriminative features from image data. To address this limitation and improve classification accuracy across diverse image datasets, this study proposes a robust framework for feature extraction and feature selection. The framework is specifically designed to enhance the classification of tea fermentation images while maintaining applicability to other image-processing tasks. It consists of four main stages: image acquisition, preprocessing, feature extraction, and feature selection. Feature extraction is performed across three categories—colour, texture, and wavelet transform features. Feature selection begins with correlation analysis (CA) to remove redundant features, followed by linear discriminant analysis (LDA) and principal component analysis (PCA) for dimensionality reduction. The proposed model is evaluated using six machine learning (ML) classifiers and two deep learning (DL) models on both a tea fermentation dataset and the Corel-1K image dataset. Initially, 94 features are extracted and subsequently reduced to 24 using the proposed feature-selection strategy. The results demonstrate superior performance, with the random forest (RF) classifier achieving an accuracy of 98.31% and gradient boosting (GB) achieving 99.71%. The DL models consistently outperform traditional ML techniques, with the convolutional neural network (CNN) showing the best classification performance. The model effectively distinguishes between under-fermented and fermented tea leaves, although over-fermented leaves exhibit slightly higher misclassification rates. Overall, the proposed approach outperforms existing methods in terms of precision, F-measure, and accuracy, albeit with slightly increased computational complexity due to multimodal feature processing and hybrid feature selection. The proposed system offers a novel, adaptable, and effective framework for feature extraction and selection, ensuring improved performance in image-classification tasks. It provides a solid foundation for future research in automated tea fermentation classification and other image-classification applications.
Keywords
Tea fermentation classification, Image processing, Feature extraction, Feature selection, Machine learning, Deep learning.
Cite this article
Sulaikha CM, Somasundaram A. A multimodal feature extraction and hybrid selection framework for image-based tea leaf fermentation classification. International Journal of Advanced Technology and Engineering Exploration. 2025;12(133):1807-1838. DOI : 10.19101/IJATEE.2025.121220208