Use of smartphone images in the evaluation of weed control with glyphosate: A computational approach using Random Forest

The evaluation of herbicide control efficacy is essential in integrated weed management, and this information is fed back into the database for future crop protection actions. The evaluation of weed control using herbicides is carried out based on on-site visual evaluations, which can be time-consuming and imprecise. In this study, models capable of evaluating the efficacy of Urochloa brizantha control by glyphosate were proposed using computational image analysis with Random Forest. The images were obtained with a smartphone in an area with U. brizantha infestation desiccated by glyphosate. To stratify the control response, glyphosate doses of 0, 144, 288, 576, 864, 1152, and 1440 g a.e. ha −1 were applied. For model estimation, 27 vegetation indices were used as input variables, with the objective of estimating control based on visual evaluation data and on control percentage data determined by the plant mass killed by the action of the herbicide. The model developed with data from visual control scores presented excellent performance, both in training (r 2 of 0.92) and in validation (r 2 of 0.92). The model trained with the percentage of dead plant mass also presented satisfactory performance, with an r 2 of 0.88 for the training sample and 0.90 for the validation sample. The proposed evaluation method presents advantages over the usual methods, due to the ease of obtaining smartphone images and the possibility of inserting the indices and models developed in mobile applications.