View Proposal


Proposer
Ian Tan
Title
Are We There Yet?: Evaluating Zero-Shot AI as FFB Ripeness Grader
Goal
To evaluated the capability of current AI for classifying Oil Palm Fresh Fruit Bunch into categories it has never explicitly trained on.
Description
Accurate ripeness grading of oil palm Fresh Fruit Bunches (FFB) directly determines oil extraction yield and mill profitability. While supervised computer vision models achieve high accuracy, they rely heavily on massive, domain-specific labeled datasets that are labour-intensive to gather and build. This project explores the frontier of zero-shot learning in precision agriculture by evaluating whether state-of-the-art vision-language models (VLMs) can reliably grade FFB maturity stages (e.g., unripe, ripe, overripe) without explicit supervised training on palm fruit data. By benchmarking zero-shot AI capabilities against unseen agricultural classes, this study investigates the viability of deploying out-of-the-box foundation models to streamline quality assessment in the palm oil supply chain. The expected outcome from this project is a technical documentation which compares with current state of the art FFB ripeness classification.
Resources
A small dataset of about 300 oil palm ffb images will be provided by the industry partner. VLM such as Qwen Image 2.0 Pro, Nano Banana Pro, Seedream 5.0 tokens will be sourced from a partner.
Background
This is work from past oil palm ripeness classification: * Teh, W.Y. and Tan, I., 2021. Coloured Edge Maps for Oil Palm Ripeness Classification. In British Machine Vision Conference (BMVC). * Tan, I.K.T., Lim, Y.H. and Hon, N.H., 2021. Towards palm bunch ripeness classification using colour and canny edge detection. In Computational Science and Technology: 7th ICCST 2020, Pattaya, Thailand, 29–30 August, 2020 (pp. 41-50). Springer Singapore.
Url
Difficulty Level
Easy
Ethical Approval
None
Number Of Students
1
Supervisor
Ian Tan
Keywords
oil palm, fresh fruit bunch ripeness, computer vision, classification
Degrees
Bachelor of Science in Computing Science