Desarrollo de una herramienta de aprendizaje automático que estime el estado de nitrógeno presente en las hojas de gulupa usando imágenes multiespectrales

The present work aims to develop a tool based on machine learning that allows the estimation of the nitrogen status in the leaves of a gulupa crop using multispectral images, for which a methodology was carried out that includes sampling, image capture with the help of a dark box and the multispectral camera Parrot Sequoia that delivers five images, four of them multispectral (red, green, near infrared and red edge), The samples were taken to the laboratory so that their nitrogen percentage could be evaluated by means of the Kjeldahl method. Segmentation techniques were used to obtain data only from the object of interest, i.e., the leaf. This segmentation was useful to find NDVI , GNDVI and OSAVI vegetation averages indices. Texture features were also extracted from the four multispectral images generated by the camera. Five classical machine learning algorithms were implemented and dataset obtained from feature extraction was used to achieve prediction between two classes, which were assigned as two nitrogen percentages ranges, based on chemical tests results.