Airborne optical image classification model for detection of Ganoderma disease in oil palm / Mohamad Izzuddin Anuar
One of the most common diseases in Malaysia in oil palm is Ganoderma basal stem rot (BSR) and causes significant yield loss during oil palm lifetime. Early disease detection of the Ganoderma BSR disease has been developed using lab-based technologies. However, those technologies require the sampling...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Published: |
2024
|
| Subjects: | |
| Online Access: | http://studentsrepo.um.edu.my/15370/2/Mohamad_Izzuddin_Anuar.pdf http://studentsrepo.um.edu.my/15370/1/Mohamad_Izzuddin_Anuar.pdf http://studentsrepo.um.edu.my/15370/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Universiti Malaya |
| Summary: | One of the most common diseases in Malaysia in oil palm is Ganoderma basal stem rot (BSR) and causes significant yield loss during oil palm lifetime. Early disease detection of the Ganoderma BSR disease has been developed using lab-based technologies. However, those technologies require the sampling of each oil palm in the field which consumes loads of labour and time. This study uses airborne optical remote sensing (AORS) images from multispectral and hyperspectral drones for detection of Ganoderma BSR disease in oil palm. Object-based Image Analysis (OBIA) was used to classify between Ganoderma BSR Disease Severity Index (GDSI). In OBIA, the segmentation parameters, the Edge is set to 30 while Merge is set to 70 to segment each of individual oil palm canopy fronds and avoid overlapping between canopies. Then, two classifiers which are 1) Support Vector Machines (SVM); and 2) K-Nearest Neighbour (KNN) were used to classify the segmented oil palm canopy into GDSI. Multispectral image classification using OBIA resulted a good accuracy (>90%) but unable to classify the mild/early disease severity. This study introduces a new procedure to process hyperspectral image for GDSI classification that starts with image colour-balancing, image transform using Continuum-Removed (CR) and First Derivative of Spectral Reflectance (FDSR), denoising using Savitzky-Golay filter and generating new images using significant wavelengths. Then, the new images were analysed using OBIA and results show high overall accuracy (92.5%) for GDSI classification with 85 % accuracy for early/mild Ganoderma BSR infection classification. The result showed that hyperspectral image performed better for detection of the early infection of Ganoderma BSR disease in oil palm compared to multispectral image
|
|---|