Mango data and images help Australian farmers

Categorised: IMPACT, Newsletter

The development of FruitMaps and auto-harvesting of mangoes at the right time has developed from work undertaken by Central Queensland University’s Institute for Future Farming Systems (IFFS) with the Australian Research Data Commons (ARDC). This involved curation and use of images of mango fruit on trees to ‘train’ a machine vision system, enabling estimation of crop yields and prediction of when to harvest. 

Professor Kerry Walsh, from IFFS, has worked on the use of near infrared spectroscopy (NIRS) for fruit quality assessment for more than 10 years. Supported by the Department of Agriculture and Water Resources and Hort Innovation (ST15005), his team of researchers continue to develop new sensor hardware and applications of existing sensors that can assess fruit and increase productivity without damaging the product. 

FruitMaps provides a visual summary of data from various sensors onto a farm map and into tables to help farmers decide when to harvest. The data displayed includes NIRS data, temperature (‘heat sum’) data and machine vision processed images of trees. The online decision support tool is freely available to Australian growers and has seen adoption in the mango industry.  

“Near Infra-Red (NIR) technology enables growers to pinpoint the optimum harvest window for their crop,” Professor Walsh said.

“The handheld spectroscopic device has been adopted across the Australian mango industry.  Knowing when to pick is useful for farm managers so they can organise labour, purchase packhouse consumables and market their produce.”

Another layer of information in FruitMaps is provided from an image processing pipeline involving input of images of all trees from a farm vehicle mounted camera travelling at 5 km/h. The system can assess the number of flower panicles or number of fruit per image, and is a component of an orchard fruit load estimate. 

Knowing both the expected volume and timing of harvest is very useful to both harvest resourcing and marketing.

The learnings from this project enabled the team to create another data display system for a project supported by Meat and Livestock Australia on soil carbon assessment and mapping.

Another spin-off activity is a system for assessment of colour of coral trout exported to China. Utilising the skills in machine vision developed for the in-field fruit load estimation work, Professor Walsh and his team recently built and tested the world’s first mango auto-harvester with field trials at Yeppoon, Queensland. This first prototype achieved a 75% efficiency in automatically identifying and picking fruit in view. Work continues to move the imaging and auto-harvest technologies to a commercial-ready stage.

machine vision rig

In-field imaging rig for counting fruit on a farm buggy

Image credit: Professor Kerry Walsh