An Algorithm For Detection of Nutritional Deficiencies from Digital Images of Coffee Leaves Based on Descriptors and Neural Networks

This work proposes an algorithm for detection of nutritional deficiencies of coffee plantations from the analysis of tonalities and geometric characteristics of the leaves. The algorithm aims to reduce the subjectivity in the analysis by visual perception. Errors in this analysis affect the dosage plan of fertilizers and nutrients applied by producers. The algorithm is formed first by a step of contrast improvement from the luminance followed by a SIFT algorithm that provides the important points for the generation of the corresponding descriptors. In parallel to this, the improved image is subjected to thresholding to obtain Hu and Fourier descriptors. With the three types of descriptors, a specific neuronal network is trained separately according to the nutritional deficiency to be detected. Kappa index was used to compare the results with those taken by visual inspection. The results were satisfactory, obtaining a Kappa coefficient of 0.96 for N and P deficiency, and 0.92 for B.