A computer vision algorithm that pinpoints unripe citrus fruit within images of the fruit trees could improve crop yields and revenues for farmers, a study by the University of Florida (UF) has shown.
The algorithm developed by Wonsuk Lee, a University of Florida agricultural and biological engineering professor, was able to find 80 per cent of the immature fruit. Lee’s study, published in the January issue of the journal Biosystems Engineering, could eventually help Florida’s $9 billion per annum citrus industry.
Using a digital camera, two of Lee’s former students took 240 photos of fruit from a research grove at UF’s Institute of Food and Agricultural Sciences on the Gainesville campus. Because of the scope of the study, these are preliminary findings, Lee said, but they hold promise for growers seeking to boost the accuracy of their crop-yield estimates.
The findings are part of Lee’s research goal of developing an electronic system that can see and count fruit using machine vision.
The system includes a digital camera, a portable computer, GPS receiver and software designed by Lee and his graduate students. Ultimately, growers would like a machine that drives itself through groves, but researchers aren’t there yet, Lee said.
The accuracy rate means growers can use the model to know well before harvest how much ripe fruit is on their trees, Lee said. Therefore, they can plan harvesting better, predict crop yields and possibly make more money, he said.
Harvesting accounts for about 30 per cent of the cost of citrus production, Lee said. With Lee’s system, growers can determine the optimal time to harvest.
The study, co-authored by UF computer and information sciences doctoral student Subhajit Sengupta, details the yield-estimation method, which may also someday help growers identify the least productive parts of their groves so they can find out why.
'You have to find the cause of those and correct those so you can increase yield and profit, eventually,' he said.
In smaller groves, it’s possible to photograph every tree, Lee said. But for those that span thousands of acres, operators would photograph trees in representative parts of the grove and use the results to make projections.
For now, Lee said, he and one of his graduate students are working on developing the self-running machine vision system that would automatically gauge a grove.
