Summary

This project aims to increase the
seed ginger training dataset for improving the performance of the AI-based
Fusarium Predictor prototype developed in the Ginger Ninja pilot project. The
pilot study was trained using one expert seed ginger grader. It is proposed to
replicate the Ginger Ninja AI-based Fusarium Predictor prototype and distribute
them to a series of farms to allow experienced ginger graders to scan and
classify seed ginger during the seed cutting season. This data will be collected
over four-weeks during seed ginger cutting and then merged with the initial
pilot project data to retrain the AI-based classifier and assess this multiple
“expert” trained performance to the original system.

Program

Ginger

Research Organisation

Queensland University of Technology

Objective Summary

The objectives of this project are:

1. To increase the seed ginger image training dataset for predicting the presence of Fusarium through “expertly” labelling and collecting the datausing the AI-based Fusarium Predictor prototype.
2. To retrain the AI-based Fusarium detection algorithm using the new data and compare performance with existing system.
3. To summarise feedback from operators during the data collection and update the Fusarium Predictor’s with the retrained AI-models