Development of basal stem rot disease detection model using terrestrial laser scanning data of oil palm crown structure

Terrestrial laser scanning (TLS) technology is an active remote sensing imaging method stated to be one of the latest advances and innovations for plant phenotyping and plant structure characterisation. It can provide accurate information via high-resolution scans on tree’s dimensions and morphology...

Full description

Saved in:
Bibliographic Details
Main Author: Husin, Nur Azuan
Format: Thesis
Language:English
Published: 2020
Subjects:
Online Access:http://psasir.upm.edu.my/id/eprint/99118/1/FK%202020%2075%20IR.pdf
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Terrestrial laser scanning (TLS) technology is an active remote sensing imaging method stated to be one of the latest advances and innovations for plant phenotyping and plant structure characterisation. It can provide accurate information via high-resolution scans on tree’s dimensions and morphology, which are important indicators of the plant’s health and development. Basal Stem Rot (BSR) is the most destructive disease of oil palm in Malaysia caused by white-rot fungus Ganoderma boninense. The infected trees show foliar symptoms such as flattening and hanging-down of the canopy, the appearance of many unopened spears, shorter leaves and smaller size of the crown. Various remote sensing approaches have been used to detect BSR. However, none of them using TLS. Furthermore, even the tree dies less than 12 months after infection, current study only monitors the tree at 6 and 12 months after infection. Therefore, this study proposes the use of TLS data of crown properties to detect BSR. This includes the study of crown and frond parts of the oil palm trees to develop a model suitable for BSR detection and also analysis of the changes using multi-temporal data of 2 and 4 months gap. A total of 40 samples of oil palm trees at the age of nine-years-old have been selected with 10 trees for each healthiness level were predetermined by the experts in the same plot. The trees were categorized into four healthiness levels - T0, T1, T2 and T3 represents the healthy, mildly infected, moderately infected and severely infected, respectively. Another 40 samples of oil palm tree taken from different plot were used for prediction. TLS was mounted at a height of 1 m and each palm was scanned at four scan positions at a distance of 1.5 m around the tree. The recorded laser scans were synched and merged to create a cluster of point clouds. Crown stratification was done to get a density of point cloud at specific strata (Cn). Meanwhile, the crown area, frond number and frond angle were gathered by processing the top-view of point cloud data. Analysis of Variance (ANOVA) at 5% significant level and four post-hoc tests - Student’s (Student-Newman-Keuls, SNK), Tukey-Kramer HSD (Honest Significance Difference), Hsu’s MCB (Multiple Comparison Best) and Dunnett’s were used to find significant features to be used as input parameter(s) of three different approaches of classification models, i.e., single parameter, combined parameters and machine learning. Results of the crown profile have shown that the upper parts of healthy tree are more densed compared to unhealthy. Five features were identified to be significant to classify BSR at four severity levels, namely C200 (strata at 200 cm from the top), C850 (strata at 850 cm from the top), crown area, frond angle and frond number. For a single parameter approach, models developed using frond number and frond angle gave the best results with both gave 100% healthy level classification, 81.67% healthy-unhealthy classification and 72.5% four severity levels of infection classification among all five parameters. Linear model using frond number, frond angle and C200 produced the best result among 118 classification polynomial models with 100% healthy level classification, 86.67% healthy-unhealthy classification and 80% four severity levels of infection classification. For the machine learning approach, the Kernel Naïve Bayes that used PC1 and PC2 as inputs gave the best results with 100% healthy and T1 (mild infection) levels of classifications, 90% healthy-unhealthy classification and 85% four severity level of infection classification compared to other 72 classification models. This model has also been identified as the best model to detect at an early stage and classify the severity level of BSR. Meanwhile, based on the results of multi-temporal analysis, compared to the unhealthy trees, the crown area and frond angle of healthy trees did not give significant changes during 2 and 4 months gap. It shows that even though there were changes in oil palm’s architecture due to a normal growth of the healthy trees, the changes were trivial and more stable. It can be concluded that the major contribution of this study is on the development of a model suitable for BSR disease detection in an oil palm tree due to Ganoderma boninense and also the capability of the model to classify its severity level of infection at very early stage (T1 – mild infection) using machine learning technique and TLS data of the crown properties. The proposed method hopefully can help better disease management at the oil palm plantation which thus can increase the oil palm yield.