New non-destructive technologies for simultaneous yield, crop condition and quality estimation
Abstract
Accurate yield estimation is a key part of winegrape production, with everything from winery intake planning to sales reliant on yield estimation to some extent. Current yield estimates rely significantly on the experience of the vineyard manager, are labour intensive and typically have an error rate of up to 30%.
More than any other perennial crop, the commercial value of winegrapes depends on fruit composition, with the same variety fetching a ten-fold difference in price depending on its quality. If objective assessment of fruit quality and condition can be done in the field prior to harvest and transport, winery planning can be optimised, a number of potential points of friction between grower and winery can be avoided and additional opportunities would be available to produce fruit to a desired specification.
This project was established to test and develop non-destructive, non-contact technologies for yield prediction/estimation, monitoring of fruit composition and in-field estimation of fruit condition. Colour digital video, stereo imaging, and low-power radar were all used to collect in-field data for yield prediction and estimation at commercial sites.
A lab-based hyperspectral imaging system was used to develop detailed calibrations of fruit composition (quality), using several thousand individually analysed samples. This same system was trialled as a field device, together with a cheaper and more robust field system developed as part of the project and using the same data analytics.
Bespoke data-processing pipelines were developed in all cases to produce grower relevant data from the sensor outputs.
Summary
Accurate yield estimation is a key part of horticultural production, with supply chains and fruit value significantly impacted when production volume doesn’t match expectations. The wine industry is no different in this respect, with everything from winery intake planning to sales reliant on yield estimation to some extent. Current yield estimates rely significantly on the experience of the vineyard manager, are labour intensive and typically have an error rate of up to 30%. Improvements in yield prediction and estimation offer significant value to the industry.
More than any other perennial crop, the commercial potential of winegrapes depends on composition, with the same variety fetching a ten-fold difference in price depending on its quality. Objective assessment of fruit quality is, therefore, paramount. If this can be done in the field prior to harvest and transport, winery planning can be optimised, a number of potential points of friction between grower and winery can be avoided and additional opportunities would be available to the grower to produce fruit to a desired specification. Similarly, accurate and objective pre-harvest assessment of crop condition would minimise winery rejection and provide the best opportunity for a profitable harvest.
With the recent revolution in sensors and data analytics, together with a pervasive online computing infrastructure to support these, there is an opportunity to provide the grape-grower with cost effective tools to predict inter-seasonal variation yield from early in the season, to monitor fruit ripening and quality during maturation and to estimate yield and fruit condition prior to harvest. Furthermore, these tools can be based on whole vineyard assessments using non-destructive measurements, rather than few, small, labour intensive, destructive samples.
This project was established to test and develop technologies to provide non-destructive, non-contact yield prediction, yield estimates, monitoring of fruit composition and in-field estimates of fruit condition. Colour digital video, stereo imaging, and low-power radar were all used to collect in-field data for yield prediction and estimation. All these devices were mounted on commercial vineyard vehicles and used to image commercial vineyards in collaboration with a number of wine companies (primarily Accolade Wine, but also Katnook Estate & Rymill Wine). Bespoke data-processing pipelines were developed in all cases to produce grower relevant data from the sensor outputs.
A lab-based hyperspectral imaging system was used to develop detailed calibrations of fruit composition (quality) against several thousand individual lab analysed samples. This same system was trialled as a field device, but in addition, a cheaper and more robust field system was developed using dual point-spectrometers and the same data analytics.
The project has clearly demonstrated the viability of using colour digital video and machine learning to count inflorescences at scale in the vineyard and thereby provide a yield prediction shortly after budburst. The same method works reliably for bunch counting on green or red grapes, but needs further optimisation for vineyards with heavier canopies. Low-power radar was able to detect fruit irrespective of canopy, but its ability to reliably quantify fruit mass still needs to be demonstrated.
Hyperspectral data analytics were able to accurately quantify key fruit maturity and quality parameters such as sugar content and total organic acids, across a broad range of winegrape varieties and fruit development stages, but estimates of berry nitrogen parameters were only effective on a per genotype basis. Although the calibrations were lab based, their use with field imaging was demonstrated and field devices to collect the relevant data developed and trialled.
It is recommended that each of these technology areas be assessed, with further industry input, for commercialisation potential, particularly considering the dollar value of the data to viticulturist decision making. For example, if the industry values a key parameter as being cost effective at $50 per hectare, then the potential to produce that data at that cost needs to be demonstrated.
For those systems at a lower technology readiness where this assessment it not yet practical, it is recommended to continue research where there is ongoing industry support.
This project was supported by Wine Australia, through funding from the Australian Government Department of Agriculture, Water and the Environment as part of its Rural R&D for Profit program, and CSIRO.