• Download report PDF

  Download

Growing ginger is resource intensive; one of the most resource-intensive periods for ginger production is from planting through to canopy closure (August-October to January-February). During this period, the slow-growing ginger is out-competed by fast-growing weeds. If this competition is not managed, growers can experience crop losses of up to 80%. The industry currently relies on manual spot-spraying to control weeds during this period. The labour required to carry out this task accounts for the largest cost in ginger production systems.

This project has begun to address this issue by developing an algorithm for the detection of weeds in ginger crops. This algorithm is designed to be applied to an autonomous platform for the continual, autonomous control of weeds in ginger crops.

At present, the algorithm has been refined to provide weed recognition precision and recall of 68% and 93%, respectively, on a representative set of test images. Both metrics are well above the general industry averages for weed recognition. These high recall and precision values represent high-level weed detection accuracy for the safe and targeted delivery of weed control treatments to grass and broadleaf weeds.

The outcomes of this research have the potential to generate large labour cost savings for ginger growers. Additionally, greater control of weeds has the potential to increase ginger crop yields, and the targeted nature of the site-specific weed control will help reduce chemical use and improved the sustainability of the industry.

Expansion of the image library used to train the algorithm is required to account for variability in ginger growing conditions and ensure the accuracy of weed detection in a variety of ginger cropping situations.

Strategies for the commercialisation of this technology are being investigated. These will include the refinement of the algorithm and development of a suitable platform to enable a suitable weed control method to be delivered.