| Sumario: | One of the characteristics of all living beings is that adequate nutrition has a positive
impact on health. In the case of plants, and specifically in fruit trees, adequate nutrition is also
essential for them to grow healthy and produce fruits in the highest quantity and quality
possible. Therefore, optimal nutrition is key for any farmer. However, excessive use of fertilisers
can harm the environment and be a waste of resources for farmers. One of the keys to achieving
adequate fertilisation is an accurate diagnosis of the nutritional status of the tree. Traditionally,
this diagnosis is made by destructive ionomics analysis, which represents a high economic cost
and a delay in obtaining the results. This work proposes Vis-NIR hyperspectral imaging and
machine learning regression models to estimate the concentrations of macronutrients (N, P, K,
and Ca) and micronutrients (Mn and Fe) in citrus leaves. The methodology involved the
application of several machine learning regression methods (linear regression, partial least
squares, random forest, support vector regression, and Ada Boost). Data were normalised with
standard normal variable (SNV), and principal component analysis (PCA) was used to reduce
dimensionality. The results were promising in estimating nutrients with R2 greater than 0,50 in
all cases, especially nitrogen (R2 of 0.77).
|
|---|