Analysis of multispectral imagery from unmanned aerial vehicle (UAV) using object-based image analysis for detection of Ganoderma disease in oil palm
Ganoderma disease that affects oil palms has caused huge losses to the palm oil industry in Malaysia. To curb widespread infection and mitigate further losses, attempts have been made to detect infected oil palms automatically so that they can be treated or destroyed. The multispectral remote sensin...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Published: |
Malaysian Palm Oil Board
2020
|
| Subjects: | |
| Online Access: | http://eprints.um.edu.my/37440/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Ganoderma disease that affects oil palms has caused huge losses to the palm oil industry in Malaysia. To curb widespread infection and mitigate further losses, attempts have been made to detect infected oil palms automatically so that they can be treated or destroyed. The multispectral remote sensing technology can be employed to this effect efficiently. From the aerial images, infected oil palms can be detected and classified according to the Ganoderma Disease Severity Index (GDSI). In this study, object-based image analysis (OBIA) was performed to classify oil palms in a selected area into three classes namely; healthy (T0), moderately infected (T2) and severely infected (T3). It would be desirable if lightly infected oil palms could also be categorised as a class. However , it was extremely difficult to discriminate lightly infected oil palms from the healthy ones just by analysing the aerial images since symptoms of early infection were not evident in the fronds yet. Images of each individual band as well as those obtained by combining two, three or four bands of the available spectra were analysed. The OBIA was conducted using example-based feature extraction procedure and various OBIA settings were tested to achieve a number of classification results. The accuracies of the results are quantified by comparing the results with the ground truth data. The results suggest that the combination of Edge-based segmentation and merge algorithm using Full-Lambda Schedule (ITS), Support Vector Machine (SVM) classifier and three-band data of (G_R_NIR) scores the highest accuracy of (91.8%). When data of individual bands were tested using the same algorithm and classifier, they obtained moderate accuracies ranging from 65.5%-76.2%. However, when data of two, three and four bands were combined, better results with classification accuracies from 70%-90% were recorded. These results show that the OBIA can be used to analyse multispectral images of oil palms to detect moderate and severe infection of Ganoderma disease. Detection of early infection of Ganoderma is feasible if more advanced algorithms and classifiers can be used with multispectral and hyperspectral aerial images. |
|---|