LACOMBE, Alta. — Chris Neeser has been studying drones for Alberta Agriculture since 2014 to gauge the usefulness of current technology in agricultural applications.
“We examined the usefulness of imagery from UAVs (unmanned aerial vehicles) for the purpose of weed and disease forecasting, and hopefully the process might answer some of the questions in case you are thinking of getting one for yourself,” Neeser told Murray Hartman’s Science-O-Rama held last month in Lacombe.
Fixed wing and multirotor drones were used to capture images of six crops on two fields each, so 12 fields altogether, and images were captured three times throughout the year.
Neeser said it was easy to see patterns of vegetative growth in the field when scouting for diseases.
“Patterns can be identified individually, or you could also identify them using an algorithm,” he said.
“They worked about equally well with some calibration.”
However, it was difficult to know whether these patterns represented diseases or were caused by features in the landscape.
“We can identify patches. We wouldn’t know what they are, but we can locate them and then we have the options to go with a GPS and walk the area to check it out,” he said.
“Or perhaps you could load up the map in a UAV and have that UAV take pictures of that area.”
Weeds were easily identified when in large patches, but it was difficult to discern weeds from crop when it began to canopy.
“From the point of view of the map, we couldn’t make a proper decision in terms of weed control based on this,” Neeser said.
The resolution limitations of the images that drones take are a major problem when it comes to early weed identification.
The camera Neeser used had a resolution of six centimetres per pixel, and if there was a weed seedling that was one cm across, the background overwhelmed the image and the weed was not picked up.
“My images for a quarter section are about 400 to 500 megabytes,” he said. “If you go to the one cm per pixel resolution, you are up into the 10 to 15 gigabytes. Your normal computer can’t handle that anymore. You need a special work station to handle images that size, and it becomes cumbersome to transfer this kind of file and work with this kind of file.”
Instead of taking high-resolution images of the entire field, Neeser took images and measured particular spots in the field.
“(We) developed an algorithm to take out the crop rows … (and then) we can assume that that is weeds,” he said.
“That would allow us to calculate the ground cover of weeds, based on the number of green pixels compared over the total number of pixels without the crop included.”
This allowed him to calculate the density of weeds per image, which allowed him to create grid samples of the fields and create weed density maps.
However, a large number of sample images need to be taken in order for a grid sampling method to accurately reflect the weed numbers in a field.
“Work that was done in Nebraska in the 90s showed that to do a green map and have some level of confidence in that map, you have to have sample distance of substantially less than 20 metres, so 10 metres or so is what we’d be looking at,” he said.
“Ten metres in a quarter section, which is about 800 by 800 metres, it would take you quite a few samples, when you’ve got 6,400.”
He said a fixed wing drone must be equipped with a high speed camera to perform grid sampling, but a multirotor drone requires lots of stopping and starting to perform grid sampling, which quickly depletes its battery.
There is a steep learning curve when it comes to using drones in an agricultural application, he said.
“Do you have the time and inclination to start working with a UAV?” Neeser said.
“It’s a new piece of technology, can be finicky. It certainly takes some time and there is time involved in using it. It can take more time than you would think.”
It is also crucial that users know how to handle the images once they are taken.
“When you capture all these images, if you can’t process them or store them properly or put them onto a GIS, then they are not really that useful,” he said. “Images become powerful once you can overlay them.”