Crown counting and mapping of missing oil palm tree using airborne imaging system

Unmanned aerial vehicles (UAV) have been recently deployed in the agriculture sector for various applications such as crop monitoring and yield prediction in order to achieve the precise agriculture objectives. In Malaysia, there are plenty of large-scale oil palm plantations, but the counting of oil palm trees and missing trees processes are still manually done which is time consuming and costly. The demand of automated missing oil palm tree counting is increasing yet none of relevant research has been published. Therefore, this study implemented different image processing techniques to true color and multispectral UAV imageries for automated existing and missing oil palm tree counting. The study found that multispectral data provides higher accuracy compared to visible bands data and Normalized Difference Vegetation Index (NDVI) index is the best index for oil palm discrimination among all the vegetation indices tested in this research. Hue, Saturation, Value (HSV) color space was found to be effective for distinguishing between the trees and weeds while the index of subtraction red from green (RG) is useful for separation of soil and greenness objects in tree identification. Integration of vegetation indices, color space model conversion and blob analysis was found capable to generate a better oil palm discrimination map. Watershed segmentation function was employed to separate the connected trees and missing trees regions. Different structuring elements of morphological filtering were applied to oil palm trees and missing trees counting for the purpose of noise removal. The undetected group of missing oil palms trees are estimated based on the planting pattern design. Over-counting error can be eliminated by merging the detected trees which depart from each other within a threshold value. For comparison purpose, the other approach of counting trees was carried out; the commercial eCognition software was tested for oil palm counting in template matching approach. The overall accuracy of counting existing oil palm trees using the approach developed in this study is 93.3% while missing trees detection gives the detection accuracy of 89.2%. The oil palm tree counting from eCognition software provides the average counting accuracy of 90.3%. The study found that the proposed method is capable to give the root-mean-square-error (RMSE) of 1.02m and 0.10m for existing tree counting and missing oil palm tree counting respectively. A graphical user interface by using MATLAB language was created based on the best executed method to aid users in counting the trees automatically. This helps to facilitate semi-skillful users to implement the tree counting algorithms. Users are required to input three parameters which are, image resolution, planting distance of oil palm trees as well as the diameter of oil palm crown, in order to operate the interface. The total number of existing oil palm trees and the number of missing trees together with its individual coordinates can be obtained using this graphical interface.

Text FK 2019 106 - ir.pdf Download (1MB)

Actions (login required)

View Item