Annotated Datasets of Oil Palm Fruit Bunch Piles for Ripeness Grading Using Deep Learning.

Area of Science:

• Agricultural Engineering
• Computer Vision
• Food Science

Background:

• Palm oil quality is significantly affected by the maturity of fresh fruit bunches (FFB).
• Existing computer vision datasets for FFB maturity classification often lack comprehensive categorization relevant to real-world palm oil mill conditions.
• Accurate FFB maturity assessment is crucial for optimizing palm oil production and quality.

Purpose of the Study:

• To introduce a novel, comprehensive dataset of oil palm FFB images and videos.
• To provide a dataset that accurately reflects the diverse conditions encountered in palm oil mill grading sections.
• To facilitate improved computer vision models for FFB maturity detection and classification.

Main Methods:

• Collected video data using smartphones (1280x720 pixels, .mp4 format).
• Dataset includes 45 single-category FFB videos and 56 multi-category FFB videos.
• Annotated and labeled the dataset with 6 distinct maturity categories: unripe, under-ripe, ripe, overripe, empty bunches, and abnormal fruit.

Main Results:

• A new, large-scale dataset of oil palm FFB was created.
• The dataset captures real-world variations in FFB appearance and maturity.
• The dataset is annotated with 6 categories crucial for industrial grading.

Conclusions:

• The developed dataset provides a valuable resource for advancing computer vision applications in the palm oil industry.
• This dataset will enable the training of more robust and accurate FFB maturity classification models.
• Addressing the limitations of previous datasets will lead to better quality control in palm oil processing.