Development of standard protocols for leaf image analysis and evaluation of machine learning methods for nitrogen diagnosis in tropical grass pastures

In Brazil, despite the relevance of tropical grasses as food source for livestock, the poor management alongwith minimal or no nutrient replenishment to the soil are key drivers of pasture degradation. The N isrecognized as the most limiting nutrient for pasture growth, and the current lack of research focused on itsnutritional diagnosis is a critical point for pastures persistency and productivity, highlighting the need for thedevelopment of robust diagnostic methods. An effective artificial vision system using image analysis shouldconsider procedures and methods capable of capture physiological, structural, and biophysical plant responsesthrough the selected relevant features. In this way, the present project aims i) to define a standard protocol forimage segmentation when using leaf images of tropical grass pastures; ii) to evaluate the impacts of RGBindices and texture features of tropical grasses aiming to identify the most promising features for optimalmodeling N diagnosis from individual leaf images; and iii) to compare the generalization performance andidentify the most promising machine learning models (and their configurations) for diagnosis of N status intropical grass pastures.