Unmixing-based approach as a tool for classification of oil palm diseases using hyperspectral remote sensing in Colombia
Abstract
Hyperspectral remote sensing has the potential to provide quantitative information on the spatial cover, acquiring relevance for the agronomic management. Traditionally, the diagnosis, management, and control of diseases in oil palm crops is a time-consuming and difficult task given that it needs a visual symptom observation. Currently, oil palm crops deal with diseases and infections. The bud rot disease (PC in Spanish) of the oil palm is one of the most common diseases in Central and South American countries, especially in Colombia. A viable alternative for the identification of diseased palms is the use of hyperspectral images and classification algorithms. Nevertheless, the usual assumption that every pixel of the hyperspectral image can be associated with a unique class label is no longer verified, and mixed pixels cannot be correctly addressed by traditional classifiers. This paper presents an unmixing-based approach as a tool for classification of stress oil palms caused by the bud rot disease, conducted on hyperspectral datasets of oil palm crops from Colombia, through the estimation of abundance maps with three labels: diseased oil palm, healthy oil palm and background (grass-shadow).
- Publication:
-
Remote Sensing for Agriculture, Ecosystems, and Hydrology XX
- Pub Date:
- October 2018
- DOI:
- 10.1117/12.2501805
- Bibcode:
- 2018SPIE10783E..12C