Stereo vision controlled variable rate sprayer for specialty crops: part I. Controller development

Authors: Jeon, Hongyoung; Zhu, Heping

Submitted to: Journal of the ASABE
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 9/14/2022
Publication Date: 11/18/2022
Citation: Jeon, H., Zhu, H. 2022. Stereo vision controlled variable rate sprayer for specialty crops: part I. Controller development. Journal of the ASABE. 65(6):1397-1410. https://doi.org/10/13031/ja.15227.
DOI: https://doi.org/10.13031/ja.15227

Interpretive Summary: A machine vision has tremendous potential in agricultural automation and data-driven agriculture as its image data will substantially increases the quality and quantity of the agricultural crop data. The stereo vision which uses two machine visions provides even more opportunities as it will acquire 3-dimensional (D) data of the detection area using two images. A spray controller using a stereo vision was developed and evaluated for its performance with moving 2-D, and 3-D targets and artificial tree. The controller showed good performance in detecting moving targets, calculating their volumes, prescribing necessary spray amount, and activating spray nozzles to apply prescribed amount on the target. The controller made 5 spray decision every second as image capture and processing took a fraction of a second. The outcomes from this research would provide a stereo vision platform for variable rate sprayers and other various agricultural automations. In addition, this research would promote other stereo vision-based research to provide data-based agricultural tools for US growers.

Technical Abstract: A variable rate spray controller using a stereo vision as a target detection sensor was developed and evaluated with moving spray targets at the travel speeds from 3.2 to 8.0 km·h-1 under the laboratory condition. The controller detected the target with the resolution of 480 pixels (vertical) × 640 pixels (horizontal) at the frame rate of 5 Hz and modulated spray nozzles at 10 Hz based on the analysis results of the stereo vision data. A customized graphical user interface was developed to operate the controller, display incoming data and control other components. Evolutions of the controller with a flat, 3-dimensional box column, and artificial tree targets at the travel speeds from 3.2 to 8.0 km·h-1 showed that the controller consistently activated the nozzles 158.7 – 293.6 ms before the spray target arrived on the spray axis. The controller also was capable to repeatedly detect the volume of the spray target with variations less than 12% for the same target at one travel speed although it measured the volume generally higher than manually calculated volume as much as 33% which was resulted from duplicated target detection between detection cycles of the controller. These duplicated detections were resulted in over applications as much as 38% although the controller repeatedly discharged spray volumes with variations less than 11% for the same target at same travel speed. In average, the spray controller required 111.2 to 140.7 ms for data processing in one detection cycle from image acquisition to activate spray nozzles for the travel speeds from 3.2 to 8.0 km·h-1 with potential of reducing it further. The developed controller would provide an economical and versatile platform with the potential of implementing capabilities and benefits of the stereo vision technology by overcoming limitations of other sensors for future variable rate sprayers.