Oil palm detection and delineation using local maxima, template matching and seeded region growing

Abstract

Oil palm (Elaeis guineensis Jacq.) is recognized as a golden crop and it contributes significantly to the economic development of Malaysia. Oil palm detection and delineation are important stepping stones for the practice of precision agriculture in the oil palm industry and it could be done so with remote sensing applications. This research aims to develop a semi-automatic, streamlined approach of oil palm detection and delineation using a combination of template matching, local maxima and seeded region growing with Worldview-2 data. The performance of the proposed methods was assessed in various aspects while taking into consideration the different planting conditions, age, and height. The proposed methods of oil palm detection managed to achieve high accuracy with overall precision and recall rate of 83% and 90% respectively and planimetric accuracy of 0.84 m root mean square error. The overall accuracy index is recorded at 71.2%. It was found that different planting conditions affect the detection accuracy to a certain degree where oil palms in optimal planting conditions are the most accurately detected with an accuracy index of 89.5%. Meanwhile, the parameters of age and height were found to have no significant effect on the planimetric accuracy or its positional accuracy. Oil palm delineation scored a high segmentation accuracy with only a 25% error rate. The proposed methods are feasible for oil palm detection with their simple, streamlined and user-friendly features and the application of this approach can be extended to other regions of oil palms with similar conditions.