Validation of an algorithm for processing Red Green Blue (RGB) images for the estimation of crude protein in grasses vs Near Infrared Reflectance Spectroscopy technology (NIRS)

• Autores: Ospina R., Oscar, Anzola Vásquez, Héctor, Ayala Duarte, Olber, Baracaldo Martínez, Andrea
• Formato: artículo
• Fecha de Publicación: 2020
• Institución: Universidad Nacional Mayor de San Marcos
• Repositorio: Revistas - Universidad Nacional Mayor de San Marcos
• Lenguaje: español
• OAI Identifier: oai:ojs.csi.unmsm:article/17940
• Enlace del recurso: https://revistasinvestigacion.unmsm.edu.pe/index.php/veterinaria/article/view/17940
• Nivel de acceso: acceso abierto
• Materia: algorithm; crude protein; drone; RGB; NIRS; algoritmo; proteína cruda; dron

The present work study aimed at evaluating the accuracy of the computerized algorithm included in the TaurusWebs ® software, which allows to calculate the percent of crude protein (% CP) in the dry matter of grasses, from images of grasslands taken by a drone with Red Green Blue – RGB- cameras. The %PC measurements calculated by the algorithm were compared to a reference, Near Infrared Reflectance Spectroscopy (NIRS), from the Corpoica (Agrosavia) Laboratory calibrated for grasses. Forty-two samples were taken for NIRS, 18 of high tropic grasses in Cundinamarca: kikuyo, Pennisetum clandestinum; false poa, Holcus lanatus; Brazilian grass, Phalaris arundinacea and 24 from the low tropics in Tolima, Colombia: pangola, Digitaria decumbens; pará, Brachiaria mutica; Bermuda, Cynodon dactylon and coloswana, Bothriochloa pertusa. The results of the NIRS were compared against the evaluations made with the algorithm to the images of the grasses, coming from the pasture where the samples were taken. The results were compared using nonparametric statistics, the Kendall correlation test and Spearman, rho=0.83 and the Kruskal Wallis test. No differences were found between the result of the %PC of grasses measured by NIRS vs. the %PC measured by the RGB image analysis algorithm. In conclusion, the information generated with the algorithm can be used for analysis jobs of the %PC in grasses.