Comparison between linear and quadratic models for Ganoderma classification

Ganoderma is the most destructive disease affecting oil palm plantation in Southeast Asia. Ganoderma infection causes reduction of oil palm productivity and potentially impacting the market value of palm oil in Malaysia. There were no effective treatments to control this disease, meanwhile the conventional methods to detect the disease are time consuming and labor intensive. Terrestrial laser scanner (TLS) was used to obtain 3D features of oil palm trees and the trees’ physical properties were analyzed for Ganoderma detection. Classification models were developed using the physical properties of oil palm trees for Ganoderma classification at four different severity levels. Five physical properties were employed in the study: frond number, frond angle, crown pixel, canopy section at 200 cm (canopy200) and canopy section at 850 cm (canopy850). The aim of this study was to compare between the linear and quadratic classification models. Data of this study were divided into three datasets: training, testing and validation. Training dataset was used as input for the classification model and testing dataset was used to attain the range of value for each severity level for classifying the trees. The classification models were validated using different dataset to determine the ability of the model to classify oil palm trees according to its healthiness level. The results showed that linear model with a combination of frond number, frond angle and canopy200 was a better model. The model could classify the healthiness level of oil palm trees with classification accuracy of 100 % for T0 and T2, 50% for T3, and could separate between the healthy and unhealthy palm trees with 100% accuracy.