How can growers best determine their lettuce plants' spacing and when to harvest? That is what researchers at Wageningen University&Research (WUR) in the Netherlands addressed with the help of 3D images and computer vision.
A recently published research paper elaborates on those third Autonomous Greenhouse Challenge results. Researchers proposed two indicators for remote, data-driven, non-invasive decision-making regarding planting distance and harvest timing.
At the close of the latest Autonomous Greenhouse Challenge, the WUR's Anna Petropoulou explained things like determining plant spacing. She is one of the research paper's authors.
Camera and computer vision algorithms
In the study, each greenhouse was fitted with RealSense depth cameras. During the cultivation competition, computer vision algorithms (Deepabv3+ implemented in detectron2 v0.6) based on these images determined the optimal planting distance and harvest time.
The resulting plant height and leaf coverage could be accurately estimated. That was, respectively, an R2 of 0,976 and a mIoU of 98.2. These two characteristics were used to develop a light loss and harvest indicator which support remote decision-making.
You can use the light loss indicator as a decision tool for timely spreading, say the researchers. Several elements were combined for the harvest indicator, leading to a fresh weight estimate with an average 22g absolute error.
The researchers point out the need for spectral indexes describing lettuce growth. Larger data sets than those currently accessible are also "crucial to addressing the existing shortcomings" between academic and more practical production systems.
During the Autonomous Greenhouse Challenge final round, some teams found they had planted the lettuce too close together. That will affect growers' net profit. Three of the five teams chose well and widened the gutters, say to the judges. All the smarter since there is remarkably little theoretical data on this, the judges remarked at the discussion right after the challenge last summer.
Lollo Bionda on movable gutters in a part of the reference crop's greenhouse used during the competition.
The study was published on March 8, 2023, in the 'Sensors' scientific magazine. The researchers were Anna Selini Petropoulou, Bart van Marrewijk, Feije de Zwart, Anne Elings, Monique Bijlaard, Tim van Daalen, Guido Jansen, and Silke Hemming.