1. Generation of maps to localized herbicide application using aerial imaging aerial imaging.
- Author
-
Sacomani, R., Romanelli, T. L., and Marinho, J. L.
- Abstract
The escalating use of agrochemicals in agriculture presents significant challenges to human health, the environment, and agricultural workers. In response, the United Nations' Sustainable Development Goal 2 encourages sustainable farming approaches, like the adoption of eco-friendly farming methods and innovations for localized weed control. Remotely piloted aircraft fitted with imaging sensors have emerged as a viable option for specific weed control. However, addressing technological challenges is necessary to enable widespread adoption. This study aims to evaluate the efficiency of an identification algorithm designed to generate maps for targeted pesticide application. Georeferenced images were acquired through remotely piloted aircraft flights conducted over commercial soybean and second-crop corn areas. Post-processed kinematic corrections of Global Navigation Satellite System coordinates achieved centimeter-level image accuracy. Orthomosaics generated from processed images provided the data for the analyzed algorithm, which produced localized application maps. Field validation data were gathered to create a ground truth map, and the weed identification performed by the algorithm was evaluated using the two-classes confusion matrix method. The performance indicators demonstrated average results of 0.78 for precision, 0.95 for recall, 0.77 for accuracy, 0.80 for F-score, and 0.56 for the Pearson's correlation. These values reveal the algorithm's proficient identification of weeds with surface areas exceeding 400 cm
2 . Utilizing artificial intelligence techniques for aerial images categorization offers a viable strategy for weed recognition and location-specific pesticide application. Nonetheless, further refinement is required to improve the algorithm's exactness and consistency, particularly to recognize weeds smaller than 400 cm2 . [ABSTRACT FROM AUTHOR]- Published
- 2025
- Full Text
- View/download PDF