Oil Palm Spear Detection Using Machine Learning

Tan, Shing Jian (2023) Oil Palm Spear Detection Using Machine Learning. Final Year Project (Bachelor), Tunku Abdul Rahman University of Management and Technology.

Text Full Text -Tan Shing Jian.pdf Restricted to Registered users only Download (3MB)

Abstract

The MANTIS pesticide spraying system developed by Genting Plantation has the potential to overcome the technical limitation of targeted application to prevent the infestation of rhinoceros beetles but is restricted by the pesticide spray control design. This research proposes to investigate the viability using of machine learning in oil palm spear detection by determining the position of the oil palm spear from the top-down position at a close range, evaluate the model detection performance and deploy the model to an edge computing device. The methodology used involved collecting dataset from the field at Genting Tebong Estate and utilising the TensorFlow framework to train the 4 object detection models with 2 different CNN architectures. The results show that EfficientDet D1 is able to perform well regardless of the size of dataset bounding box given while the d0 variant requires smaller bounding box which just fit the size of the target object. Both EfficientDet class models are able to generalise the features of the oil palm spear and its detection results on the video dataset corresponds to the precision and recall trend while the same is not true for the SSD MobileNet class models.