Semi-Field Data Collection

Data collection of plants at seedling stage at the field is a challenging task, mainly for weeds which are hard if not impossible to identify when they are too small. To overcome this problem we have developed a high throughput outdoor setup where we grow the plant species of interest in pots and image them using an automated imaging system called the BenchBot. The range of plant sizes for this setup goes from seedling up to 45 cm or until a point where we estimate that their morphology could be affected by the restriction in growth that the pots could pose depending on the species. By growing plant in pots we know exactly what each pot has and we can automatically label the images just by knowing pot position without the need of visual identification.

Three of these setups have been imaging weeds, cash crops and cover crops since 2022 one on each of the three core partner institutions TexasA&M in TX, NC State University in NC and SASL-BARC-ARS in MD.

This semi-field setup allows us to collect thousand of images in a semi-automated way, these images are then uploaded to our database in the cloud and lastly fed into the semi-field data processing pipeline.

Open

AgIR Semi-Field System Diagram. The potted plants are arranged on a grid and are imaged by the high resolution camera which is mounted to the robot. On the host, a python script lunches the user interface, controls the robot and the camera, and organizes the incoming data (JPG and RAW files) and metadata. All the user needs to do on a typical day is run the script from the command line, enter some simple metadata and then upload a dated folder to the cloud using our web uploader. The robot y and x movement, camera plus flash triggering and files organization is fully automated.