UPM Institutional Repository

Identification of basal stem rot disease in oil palm tree using thermal imaging technique


Abdol Lajis, Ghaibulna (2018) Identification of basal stem rot disease in oil palm tree using thermal imaging technique. Masters thesis, Universiti Putra Malaysia.


Basal stem rot (BSR), caused by Ganoderma boninense is known as one of the deadliest diseases in the oil palm plantations in Southeast Asia. Ganoderma could reduce the productivity of oil palm plantations and possibly reduce the market value of palm oil in Malaysia. The available technique of BSR detection is time-consuming and human dependence. This study focuses on detecting the oil palm tree infected by BSR using thermal imaging technique. In order to find a suitable time to capture the thermal images, thermal images of canopy and trunk sections of the oil palm trees from healthy and BSR-infected trees were captured in the morning (9 to 12 pm) and afternoon (12 to 3 pm) session. The images were pre-processed using FLIR QuickReport 1.2 (FLIR Systems, Inc., Oregon, United States). The images were then processed using MATLAB software (Version R2016b, The MathWorks Inc., Massachusetts, United States) to extract pixel value representing thermal properties of the trees. After that, statistical analysis was done using these pixel values. The result from T-test has shown that thermal images taken at canopy section during the afternoon session have a significant difference (α<0.05) between healthy and BSR-infected trees. There were four features extracted from the images of canopy section namely minimum, maximum, mean and standard deviation value. Based on the statistical analysis, only mean of the pixel value gave a significant difference with a P value of 0.0052. For the maximum feature, all the data has the same value regardless of the healthiness condition, hence this feature will not be used for further analysis. Four different types of classifier namely, Linear Discriminant Analysis (LDA), Quadratic Discriminant Analysis (QDA), Support Vector Machine (SVM) and k-nearest neighbour (kNN) were used and compared. Input parameters were taken from different combinations of the features to classify the healthy and BSR-infected trees. The result showed that quadratic SVM with the input parameter using a combination of minimum and mean gave the highest percentage of accuracy with 67.0%. In order to improve the accuracy, new indices called; healthy variance (YH) BSR-infected variance (YUH) and all variance (YALL) were developed based on the squared value of the difference between the mean intensity value of an oil palm tree and the averaged mean intensity value of healthy, BSR-infected and all samples accordingly. However, it only gave the best accuracy at 62.3% from the combination of minimum, mean, standard deviation and YUH using linear SVM classifier. As a result, the Principal Component Analysis (PCA) was introduced to extract the most suitable features among six features available. The score plot of PC1 versus PC3 has shown that there were two distinguishable trendlines where the BSR-infected tree is located outside the trendline of the healthy trees. Values of PC1 and PC3 were later used for classification using all fourteen different types of classification model. Based on the results, the quadratic SVM model gave the best classification with the highest accuracy of 89.2% for the training set and 84.4% for the test set. Based on this study, it can be concluded that thermal imaging has the potential for BSR detection in oil palm trees.

Download File

FK 2018 149 - IR.pdf

Download (1MB) | Preview

Additional Metadata

Item Type: Thesis (Masters)
Subject: Agricultural innovations
Subject: Oil palm - Technological innovations
Subject: Infrared imaging
Call Number: FK 2018 149
Chairman Supervisor: Siti Khairunniza Bejo, PhD
Divisions: Faculty of Engineering
Depositing User: Ms. Nur Faseha Mohd Kadim
Date Deposited: 27 Nov 2019 08:24
Last Modified: 29 Nov 2019 08:27
URI: http://psasir.upm.edu.my/id/eprint/76066
Statistic Details: View Download Statistic

Actions (login required)

View Item View Item