TITLE:
Early Detection of Cocoa Swollen Shoot Using Hyperspectral Reflectance Spectroscopy
AUTHORS:
Nimongon Seydou Silué, Penétjiligué Adama Soro, Amara Kamate, Emma Georgina Hueva Zoro, Kouabenan Anicet Kouakou, Adjo Viviane Adohi-Krou
KEYWORDS:
Swollen Shoot Virus, Early Detection, Spectroscopy, Cocoa Tree, Hyperspectral Reflectance, Machine Learning, Vegetation Indices
JOURNAL NAME:
Spectral Analysis Review,
Vol.9 No.1,
June
16,
2026
ABSTRACT: Cocoa swollen shoot is a viral disease prevalent mainly in West Africa. It is responsible for significant yield losses. Effective monitoring and accurate detection of this disease are essential to ensure stable and reliable cocoa production, as well as the income of thousands of farmers. Current standard methods often rely on visual inspection for disease symptoms. This method is costly, time-consuming, and prone to errors due to the subjectivity of inspectors. Recent advances in precision agriculture, using spectral data at different scales (leaf, canopy, and space), offer the potential to solve these problems at low cost and with high efficiency. Spectral vegetation indices are widely used for the indirect detection of plant diseases. Therefore, this work aims to evaluate the potential of hyperspectral reflectance spectroscopy, using vegetation indices, for the early detection of cocoa swollen shoot. Thus, reflectance spectra of healthy, asymptomatic and symptomatic cocoa tree leaves were collected using an Ocean Optics USB 4000 spectrometer, with a wavelength range from 350 to 1100 nm. To reduce scattering and noise effects in the collected data, we applied three essential spectroscopic preprocessing methods: the Multiplicative Scattering Correction (MSC), the Standard Normal Variable (SNV), and the first order Savitzky-Golay derivative (SG-D1). Principal Component Analysis (PCA) identified MSC as the most effective preprocessing method. Subsequently, eight vegetation indices (NDVI, GNDVI, Datt5, Ctr2, PSRI, SIPI, PRI, Lic2) selected from the literature were calculated. Using classification algorithms such as Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Decision Tree (DT) and Random Forest (RF) applied to all eight vegetation indices, the performance of each index was evaluated for the infection detection and the classification of the three types of cocoa trees. The results obtained indicate that the most efficient model is RF combined with MSC pretreatment with an overall accuracy of 97.59%, thus showing the strong potential of reflectance spectroscopy for the early detection of cocoa tree infection by the swollen shoot virus.