Palm Oil Loose Fruit Detection Using YOLOv4 Algorithm

Tiong, Yeong Ming (2023) Palm Oil Loose Fruit Detection Using YOLOv4 Algorithm. Final Year Project (Bachelor), Tunku Abdul Rahman University of Management and Technology.

Text Full Text -Tiong Yeong Ming.pdf Restricted to Registered users only Download (14MB)

Abstract

This paper proposes a deep learning-based automatic loose fruit detection method for oil palm using machine vision. The objective is to reduce the time taken to pick up loose fruits manually and to increase productivity. The problem statement highlights the limitations of traditional methods and the need for an automatic detection method. The study compares two popular algorithms, Faster R-CNN and YOLO, and notes the limitations of Faster R-CNN, which achieves higher accuracy but has low frame rates. The study discusses the use of CNN in both algorithms and the importance of real-time detection. The study reviews previous research on YOLOv3 and YOLOv4 algorithms and their modifications, including the use of DenseNet backbone, for detecting the maturity level of oil palm. The methodology involves training and evaluating the models using different image preprocessing datasets, with the implication of different input size configurations in the architectures, learning rate and modifications on the network architectures. The results show that the modified YOLOv4 algorithm achieves higher accuracy (83.25%) for detecting palm oil loose fruit.