Wine Australia – wine grape yield prediction and estimation commercialisation opportunity

Opportunity for

• Agritech software or hardware providers, innovators and commercial partners interested in access to wine grape yield predictions to integrate into existing or new platforms

Opportunity description

Industry challenge:

Poor yield estimation costs the wine industry $100-200 million per year. In-season yield estimation is vital for value chain planning, from the input decisions and crop control in the vineyard through harvest logistics and winery planning to marketing decisions. Current yield estimation protocols can be laborious, involve only a small proportion of the block and be based on visual assessment. Consequently, reliability can be low, with errors of up to 30-40%.

Current opportunity:

Commonwealth Scientific and Industrial Research Organisation (CSIRO) and Wine Australia are seeking engagement from AgriTech solution providers, innovators and commercial partners interested in licensing or partnering to commercialise non-destructive grapevine yield estimation technology. There is a significant opportunity to co-develop products and services using the technology.

Progress has been made validating the technology in the field:

1. Early season yield estimation – four sites and 5 varieties, video capture, machine learning, data pipeline in progress.
2. In season yield estimation – four sites and 5 varieties, video capture, machine learning, data pipeline in progress
3. Pre harvest yield estimation – one site, reflectance radar, machine learning in progress, data pipeline to be determined.

Opportunity background:

Non-destructive grapevine yield estimation technology features and application so far.

Deployed solution for early season and in-season yield potential:

• Video capture using an action-cam (GoPro), or similar, post-budburst (E-L 12-14) when inflorescences clearly visible (early season) and post-fruitset (E-L 29) when individual berries potentially visible in imagery (in season)
• GPS monitors location
• Machine learnt tool identifies and tracks inflorescences (early season) in video
• Machine learnt tool identifies and tracks bunches in video (in season)
• Machine learnt tool identifies non-obscured bunches, segments bunches from images, identifies and counts visible berries (in season)
• Raw or processed data uploaded to cloud service
• Georeferenced inflorescence count (early season), and bunch count and berry count (in season) data provided
• Maps or per block summaries
• Platform uses bunch or inflorescence counts (early season) and bunch counts and berry per bunch estimates from prior seasons to determine likely yield and over or under production relative to target.

For pre-harvest yield estimation:

• Foliage penetration sensor (radar) collects reflectance (or transmission) data after maximum fruit weight reached (E-L 36)
• GPS monitors location
• Computational tool determines fruit mass from calibration on device, at farm base or (preferably) in the cloud to provide a mass estimate
• Raw or processed data uploaded to cloud service
• Georeferenced fruit mass data provided
• Maps or per block summaries.