Citation

Kwang, Chee Seng (2026) Hyperspectral imaging signatures for early detection of Ganoderma disease using deep neural networks. Masters thesis, Multimedia University.

Full text not available from this repository.

Abstract

The oil palm industry is one of the main contributors to the Malaysian national economy. However, the industry currently faces a challenge in dealing with Ganoderma infection, a destructive disease affecting oil palm trees and causing base stem rot (BSR). As hyperspectral imaging provides richer spectral information and large-scale monitoring compared to other sensors like multispectral, offering a greater potential for subtle disease detection, and traditional statistical analysis can be timeconsuming and less effective for complex spectral patterns, this research proposes a method combining deep learning models. The study utilises a dataset of hyperspectral images captured from oil palm plantations in Johor, Malaysia, covering the spectral range from 500 nm to 900 nm. Unlike existing studies that utilise both spatial and spectral-extracted features for detection. This research seeks to leverage both spatial and rich spectral information from hyperspectral images to distinguish between healthy and infected trees, providing more accurate and insightful results for early intervention and disease management in oil palm plantations. Several preprocessing techniques have been developed before implementing the prediction, which include feature-based band alignment, instance segmentation, and dataset conversion. A deep learning model was trained to classify the disease stages of the oil palm trees. To ensure transparency and build trust in the model's predictions, explainable artificial intelligence (XAI) techniques, specifically integrated gradients, were employed to interpret the model's decision-making process. The proposed model achieved a high classification accuracy of 92%. The XAI analysis successfully identified the key spectral bands most influential in distinguishing between healthy and infected trees; they are the Green Region (500nm – 570nm), the Red-Edge Region (680nm – 750nm), and the Near-infrared (NIR) region (750nm – 900nm). This reveals that wavelengths associated with chlorophyll content and water stress were critical for early disease detection.

Item Type:	Thesis (Masters)
Additional Information:	Call No.: TR267.73 .K83 2026
Uncontrolled Keywords:	Hyperspectral imaging
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering
Divisions:	Faculty of Information Science and Technology (FIST)
Depositing User:	Ms Nurul Iqtiani Ahmad
Date Deposited:	22 Jul 2026 08:29
Last Modified:	22 Jul 2026 08:29
URII:	http://shdl.mmu.edu.my/id/eprint/16377

Downloads

Edit (login required)