Hybrid Deep Learning Model with GLCM and CIELAB Feature Fusion for Tea Leaf Disease Classification

¹ Informatics Engineering Study Program, Faculty of Industrial Technology and Informatics, Universitas Muhammadiyah Prof. Dr. HAMKA, Jakarta Selatan, Indonesia
² Informatics Engineering Study Program, Institut Teknologi Pagar Alam, Sumatera Selatan, Indonesia
³ Department of Electrical Engineering, Faculty of Engineering, Universitas Muhammadiyah Yogyakarta, Yogyakarta, Indonesia
⁴ Smart City Information Systems Study Program, Universitas Tunas Pembangunan Surakarta, Surakarta, Indonesia
⁵ Faculty of Engineering and Computer Science, Universitas Teknokrat Indonesia, Bandar Lampung, Indonesia

^* Corresponding author: This email address is being protected from spambots. You need JavaScript enabled to view it.

Abstract

Tea leaf diseases significantly affect the quality and productivity of tea plants, leading to economic losses in the agricultural sector. This study proposes a hybrid deep learning classification model that integrates the MobileNetV2 architecture with additional features from the CIELAB color space and Gray Level Co-occurrence Matrix (GLCM) texture descriptors. The Tea Sickness Dataset, comprising eight classes of diseased and healthy leaves with approximately one hundred images per class, was used and preprocessed through color conversion, normalization, and moderate augmentation. GLCM texture features were extracted from the luminance channel at four orientations, resulting in twenty statistical features that were fused with CNN features. Experimental results show that the hybrid model achieved an accuracy of eighty-six percent, compared to eighty percent for the baseline MobileNetV2. While these findings confirm the benefits of combining visual and statistical features, the relatively small dataset size and potential class imbalance remain challenges that may limit generalization. Future work will focus on larger and more balanced datasets, alternative fusion strategies, and advanced architecture such as attention mechanisms and explainable AI to further enhance performance and applicability in real- world precision agriculture systems.