Keeping tabs on the fruit of the vine

By RICHARD WOOD

A kiwifruit industry research company is evaluating technology that aims to electronically count flower buds or fruit on the vine.

Kiwifruit Supply Research (KSRL) is looking at a proposal from Auckland IT consulting firm Simpl to use "Neural Net" technology to evaluate digital photographs from orchards.

Neural Net works like the human brain, by recognising patterns.

The aim is to replace smaller-scale manual counting methods, improve estimates of kiwifruit crop size, and prevent excessive handling and marketing costs.

Simpl is attempting to find commercial uses for Neural Net processor technology and a source of further funding to establish a lead globally in the area. It has applied for funding through Technology NZ.

Rob Craig, KSRL director and managing director of Franklin-based Punchbowl coolstores, said the kiwifruit industry often struggled with crop estimates.

"From bud break through to harvest it's important to know what's sitting on the vine."

Industry-wide, logistics and marketing spend can be tailored to the total expected, so the more accurate the better, Craig said.

"They need to put the promotional plans in six months ahead, particularly when you've got a new product like the Gold kiwi where they are spending very heavily on promotion."

At an individual packhouse or orchard the technology may improve overall predictive capability and management and help determine optimum culling and pruning levels.

Pruning strategies can be geared to a specific fruit size mix that optimises market conditions.

Simpl technologist Paul Burnett said the long-term plan was to make a version for orchardists, who could do their own counts just after thinning and before harvest.

"That information gives them some new decisions about labour, about what they are going tell the packhouse to expect, and the sizes."

Burnett said the firm had been looking for technology for a few years to deal with some image-recognition needs, including kiwifruit counting, before it stumbled upon the Archimedes Neural Net project at Stanford University run by New Zealander Neil Scott. Scott gave a talk on the topic in New Zealand in 2001.

Burnett said Simpl was also talking to the Auckland University of Technology to tap into its work in the area of image recognition.

Simpl has spent the last 12 months getting to grips with the technology and how to get it to recognise an object in a picture. In the kiwifruit example it would look at a kiwifruit's defining characteristics, such as the remains of the flower at the bottom, and texture.

Simpl proposes taking much larger samples by taking photographs along rows, and perhaps even entire orchards.

It has a laboratory prototype and is ready to move it into the field, where the push will be to make it more robust.

Simpl is also floating the possibilities for the forestry industry in terms of recognising where the knots are in a tree prior to milling, and other forms of horticulture such as apples and grapes.

Craig said there was export potential as the grape industry, in particular, was concerned about crop loads.