Development of a DGCI threshold model for real-time weed detection and herbicide application in turfgrass

Abstract Precision applications are gaining interest as a sustainable approach to managing turfgrass pests. For instance, controlling turfgrass weeds with precision application could effectively reduce herbicide volume without sacrificing weed control. Machine learning models have been a common method for precision application, but machine learning requires intensive labor and expertise to collect and label imagery. The objective of this study was to develop and test a new system utilizing the Dark Green Color Index (DGCI) for precision application of glyphosate to detect and spray winter weeds in dormant bermudagrass systems. For this study, a sprayer prototype was constructed that utilized machine vision and DGCI. The prototype consisted of three primary components: 1) a camera that streamed video frames, 2) a control system that stored computer code focused on the integration of DGCI, and 3) solenoid valves that activated upon detection of winter weeds in dormant bermudagrass. Four field trials with different weed species and weed densities were established to test the DGCI system amongst traditional application methods (i.e., broadcast application and manual spot application with a backpack sprayer). In the lowest weed-density scenario, the DGCI system accurately detected and sprayed 90% of the weed population, reducing herbicide volume by 62% compared to a broadcast application. Additionally, the DGCI system required less time for treatment than the spot application with a backpack sprayer. The results from these trials suggest that vegetative indices, such as DGCI, have potential in dormant bermudagrass systems to optimize herbicide volume.