Use of machine learning to determine extraneous matter and billet length in cane consignments

Cost pressures in harvesting and transport in the sugar industry have led to higher extraneous matter (EM) levels and shorter billets in harvested green cane, and there is no cost-effective procedure to measure either of these important parameters. Manually sorting and weighing grab samples of cane (as used to be standard practice in the 1970s and early 1980s) is no longer a viable option for Australian sugar mills due to prohibitive costs and shortage of labour. Tully Mill is the only factory that still regularly undertakes manual sorting (*using an automated EM lab) but, even then, fewer than 15% of consignments are analysed. Measurement of EM% mass and billet length on each cane consignment will allow the industry to collaboratively determine appropriate target values for each cane district, depending on the district’s circumstances. Once adopted by the industry the measurements of EM and billet length provide large datasets to support research to maximise the value of the harvested crop in a mill district.

QUT and Tully Mill in July 2022 completed an SRA and SRI funded small milling research project to evaluate the potential for machine learning technology to undertake this task. That work is the forerunner of this project.

• Develop a large pool of cane images and reference extraneous matter (EM) and billet length data from cane supplied to three factories.
• Develop and demonstrate a machine learning (ML) system to analyse images of cane consignments on the conveyor as it enters the factory to provide estimates of EM content and billet length.
• Develop an adoption-ready EM and billet length measurement system and make it available for implementation in all Australian sugarcane milling districts.