Drones have been used for years to help identify the overall plant health of crops, but the cost of that equipment has been a limitation for many farmers. A study funded by the Louisiana Soybean and Grain Research and Promotion Board has been showing comparable effectiveness from less-expensive drones when coupled with proper data processing and data interpretation.
“We’ve had results slightly better than more-expensive NIR (near-infrared) multispectral cameras and as good as the GreenSeeker,” said LSU AgCenter agricultural engineer Randy Price.
Price is experimenting with lower-cost drones readily accessible at big-box stores that capture images in standard RGB — red, green and blue. The images are then processed with software to strip out background pixels of the soil. In the case of rice, water is taken out as well.
“This leaves only the plant spectral properties for evaluation,” Price said. “The resulting image can also be evaluated for red absorption, leading to a very accurate assessment of plant chlorophyll production and health, as well as variances in the field.”
“The more absorption in the red, the healthier the plant; the more reflective the red, the less healthy the plant,” he said.
The expensive NIR cameras capture images that use Normalized Difference Vegetation Index (NDVI) calculations (used since the 1970s with satellites) and scale them to more well-known health indices. The difficulty is that these cameras are harder to use, generate four times as much data and require special processing techniques and high-end computers to operate.
NIR cameras are typically only useable for one purpose — field mapping — and do not allow purposes such as checking fence lines and cattle. In addition, the NDVI indices can be much more susceptible to shadows from the sun and clouds that change and alter the images.
Price has, to this point, conducted his tests on oat, wheat and soybean fields at the Dean Lee Research Station as well as additional tests at offsite corn and rice fields. Post-processing of the RGB images helps remove pixels, and therefore unnecessary noise, that could limit the analysis of the plant itself.
“With all these cameras, we want to see small changes in the field, not just big changes,” Price said. “This system we’re using produces an image that is similar to a Hollywood greenscreen, except we use a red background, making it easier to see the plants and the plant density in a field without the soil background color.”
In order to strip the soil from RGB images, special software (currently unavailable in most software packages) and histograms are required to complete the function. Price is working on methods to take out the soil from the background, using home-written software and special histograms to evaluate fields and the green pixels left in it.
The current problem with standard RGB cameras is some pre-processing may be needed to help the cameras attain sensitivity levels similar to an NIR camera. There is not as much software currently available to help farmers with RGB camera images, while plenty of software is available for multispectral cameras, Price said.
Price has been working to produce a free software toolbox to help farmers process images and interpret the data from these cameras.
“There is a definite learning curve with this technology, and it requires a systems approach that involves multiple steps,” Price said. “The real challenge lies in the correct analysis and interpretation of the data.”
“We’re continuing research on these indices, which are all linearly related to each other,” he added. “In the future, farmers may find more incentive to buy these less-expensive drones to evaluate their crops and detect crop variances that indicate plant health and biomass.”
This story is featured in the Louisiana Soybean and Grain Research and Promotion Board 2020 Report.
Close-up of soybean plants with soil stripping taken with a standard RGB camera drone. Photo by Randy Price
Color threshold on rice plots taken with a standard RGB camera drone. Photo by Randy Price