A University of Florida Institute of Food and Agricultural Sciences researcher has developed a machine vision system to measure the quantity of a citrus grove’s dropped and decayed oranges. The fallen fruit provides an indication of a so-far incurable disease that has been spreading through Florida’s trees since its first appearance in the state in 2005.
The condition, known as Huanglongbing (HLB) or citrus greening disease, leads to shrunken, lopsided, and bitter fruit that growers cannot sell.
To detect citrus greening, Wonsuk “Daniel” Lee, a UF/IFAS professor of agricultural and biological engineering, created an imaging box containing a video camera, illumination device, and computer. The ground-based system takes videos underneath the tree canopies as operators drive around the grove.
By extracting individual frames from the video and correcting image brightness, potential fruit objects are located based on shape, color, and texture. A measurement of the color’s saturation provides differentiation between recently dropped and decayed fruit.
Rotten fruit, for example, has a dark color: black or dark brown. A recently dropped orange, conversely, maintains its brighter hue.
Bringing Machine Vision Outdoors
To spot the dropped oranges, Lee employed the same machine vision technology used to inspect parts on a factory floor. Incorporating machine vision indoors, however, is straightforward compared to a sunlit citrus grove that has varied and unpredictable illumination levels.
“You might have a very bright area and a shady area together in one image,” said Lee.” The first step in our algorithm was to correct the brightness so that we can have uniform brightness in a single image and detect the object better.”
A computer-processing correction method called Contrast Limited Adaptive Histogram Equalization, originally used in medical imaging, redistributes an image’s lighting values to provide consistent brightness across various pictures.
According to a September 2016 paper co-authored by Lee called “Detection of Dropped Citrus Fruit on the Ground and Evaluation of Decay Stages in Varying Illumination Conditions,” the results showed all processed images had desired brightness levels (152 out of a standard 255 RGB output), with a standard deviation of 1.0.
Beyond the randomness of outdoor illumination, the unstructured nature of a citrus grove environment provides further challenges for machine vision detection. At times, for example, multiple fruits are touching each other, or one orange may be hidden by branches or leaves.
By determining the roundness of a given object, its color, and its texture, the machine vision algorithm separates the objects and identifies remaining fruit portions.
Lee’s study also indicated that fruit was identified with 89.6 percent accuracy; false positives were measured at 5.0 percent. False classifications of decay stages of fruit were 4.2 percent and 18.5 percent for recently dropped fruit and rotten fruit, respectively.
The Huanglongbing (HLB) disease was first spotted in July 2004 in Brazil. In August 2005, the disease was found in the south Florida region of Homestead and Florida City. The infection has compromised orange production and led to the loss of millions of trees.
At a keynote address to the American Peanut Research and Education Society annual meeting in Clearwater, FL, in July of 2016, Bob Shatters, a research molecular biologist with the USDA’s Agricultural Research Service, said 80 to 90 percent of Florida is either infected or will be infected by citrus greening. Greening has also been found in California, Texas, and Arizona.
The Florida university’s machine vision system efficiently provides a valuable map for farmers: an instant look at a citrus grove’s health.
“This is an all-automatic system,” said Lee. “Think about the size of a citrus grove in Florida. You cannot manually inspect and count the number of dropped fruit.”
Lee plans to commercialize the machine vision system to create a realtime “dropped-fruit” mapping system for growers.