A novel RGB-driven 3D spectral representation learning framework with imbalance-aware 3D GAN and 3D CNN for oil palm health assessment
Smart Agricultural Technology · 2026-12 · 15 · 102590
DOI: 10.1016/j.atech.2026.102590 ↗Kurz gefasst
Fachartikel in der Zeitschrift Smart Agricultural Technology, 2026-12. Verfasst von Hartono und 5 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 63 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Ministrstvo za visoko šolstvo, znanost in tehnologijo.
Abstract (Originalsprache)
Spectral reconstruction from RGB images has become an essential approach for enhancing plant disease detection, particularly when hyperspectral data are not readily available. This study proposes a novel framework that reconstructs 3D hyperspectral cube from RGB oil palm leaf images by integrating Visible (VIS) and Near-Infrared (NIR) bands. Instead of directly predicting the full spectral cube, the reconstruction process is formulated as a structured and progressive approach. A coarse spectral cube is first generated as an initial spectral prior to approximate the spectral distribution and constrain the solution space. Simultaneously, a feature cube is extracted to capture multi-scale spatial information, including texture, structural characteristics, and contextual patterns, which are critical for reducing spectral ambiguity and improving reconstruction accuracy. To address the significant class imbalance in the dataset, a 3D Generative Adversarial Network (3D GAN) is employed to generate realistic synthetic spectral samples, enhancing the representation of minority classes. This balanced spectral dataset enables more effective learning and reduces bias during model training. The reconstructed and augmented 3D spectral data are then utilized by a 3D Convolutional Neural Network (3D CNN) to perform classification of oil palm leaf conditions, including Brown Spot, White Scale, Healthy, Nitrogen Deficiency, and Potassium Deficiency. Experimental results demonstrate that the proposed method consistently outperforms six benchmark models, achieving the lowest RMSE values (down to 0.112), the highest coefficient of determination (R 2 ) up to 0.930, and superior RPD values reaching 2.946. These findings confirm that the integration of structured spectral reconstruction, GAN-based class imbalance handling, and 3D CNN classification significantly improves prediction accuracy, particularly for minority classes. The proposed framework offers a robust and scalable solution for intelligent oil palm disease detection in precision agriculture.
Förderung
- Ministrstvo za visoko šolstvo, znanost in tehnologijo
Zitieren
APA
Hartono, Zuhanda, M. K., Syah, R., Ongko, E., Kuswardani, R. A., & Suswati (2026). A novel RGB-driven 3D spectral representation learning framework with imbalance-aware 3D GAN and 3D CNN for oil palm health assessment. Smart Agricultural Technology, 15, 102590. https://doi.org/10.1016/j.atech.2026.102590
Vancouver
Hartono, Zuhanda MK, Syah R, Ongko E, Kuswardani RA, Suswati. A novel RGB-driven 3D spectral representation learning framework with imbalance-aware 3D GAN and 3D CNN for oil palm health assessment. Smart Agricultural Technology. 2026;15:102590. doi:10.1016/j.atech.2026.102590
BibTeX
@article{Hartono2026,
author = {Hartono and Zuhanda, Muhammad Khahfi and Syah, Rahmad and Ongko, Erianto and Kuswardani, Retna Astuti and Suswati},
title = {A novel RGB-driven 3D spectral representation learning framework with imbalance-aware 3D GAN and 3D CNN for oil palm health assessment},
journal = {Smart Agricultural Technology},
year = {2026},
volume = {15},
pages = {102590},
doi = {10.1016/j.atech.2026.102590}
}
EUVIMED