Oil Palm Fruits Dataset in Plantations for Harvest Estimation Using Digital Census and Smartphone.

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Accurate monitoring of oil palm Fresh Fruit Bunches (FFBs) is crucial for optimizing plantation operations.
• Existing datasets may not fully capture the complexities of real-world harvesting conditions.

Purpose of the Study:

• To present a comprehensive, expertly annotated dataset of oil palm FFBs across five maturity stages.
• To facilitate the development of deep learning models for FFB detection and classification.

Main Methods:

• Collected smartphone video data of oil palm trees in Central Kalimantan, Indonesia.
• Extracted and annotated video frames using Computer Vision Annotation Tool (CVAT) in COCO format.
• Utilized data augmentation to enhance dataset variation and address class imbalance.

Main Results:

• A dataset comprising 10,207 training, 2,896 validation, and 1,400 test images was created.
• The dataset includes images with real-world challenges like occlusion, blurriness, and low contrast.
• The dataset supports object detection tasks for FFB analysis.

Conclusions:

• The developed FFB image dataset is foundational for advancing AI in oil palm cultivation.
• This resource will support improved harvest timing, yield prediction, and resource management.
• The dataset's diversity and annotations enable robust deep learning model training.