Classification of basal stem rot disease in oil palm using dielectric spectroscopy

Dielectric spectroscopy method has been identified to be able to classify plant diseases. However, studies on its application on basal stem rot (BSR) disease in oil palm are yet to be explored. This study investigated the feasibility of utilizing dielectric spectral properties such as impedance, capacitance, dielectric constant, and dissipation factor in classifying BSR disease in oil palm trees. Leaflet samples from different oil palm trees (healthy, mild, moderate, and severely-infected) were collected and dielectric properties were measured using a solid test fixture connected to an impedance analyzer at a frequency range of 100 kHz–30 MHz with 500 spectral intervals. Two data reduction methods were used 1) feature selections methods and 2) principal component analysis (PCA). First, features selection algorithms (genetic algorithm (GA), random forest (RF), and support vector machine-feature selection (SVM-FS)) were used to select the most significant frequencies. Then, data at the most significant frequencies were served as the input of six classifiers, namely: support vector machine (SVM), artificial neural networks (ANN), linear discriminant analysis (LDA), quadratic discriminant analysis (QDA), k-nearest neighbours (kNN) and Naïve Bayes (NB) to classify different levels of BSR disease. In a separate analysis, PCA was used as the data reduction method and the highest principal components (PCs) were served as the inputs to the classifiers. An analysis was done to see the effect of implementing different data reduction algorithms in classifying BSR disease. The results showed that the impedance parameter was the best in classifying BSR severity levels compared to the other dielectric properties. The severelyinfected oil palm leaflets had the highest mean impedance (11.95 Ω) and the lowest was found at the healthy oil palm leaflets (4.56 Ω). For feature selection algorithms, SVM-FS model gave the best classification accuracies compared to GA and RF; ranged from 81.82% to 88.64% with SVM and kNN as the best classifiers. Better classification accuracy was achieved using QDA classifier when implementing PCA dimensionality reduction with the accuracy of 96.36%. Without implementing any data reduction algorithm, the highest classification accuracy was found in SVM classifier with 79.55%. As such, this study demonstrates the potentials of utilizing dielectric spectral properties of oil palm leaflets in classifying the BSR diseases of the trees.

Text FK 2019 27 ir.pdf Download (1MB)

Actions (login required)

View Item