Automatic detection of nutritional deficiencies in coffee tree leaves through shape and texture descriptors

Nutritional deficiencies in coffee plants affect production and therefore it is important its early identification. The current research is focused on the automatic identification of nutritional deficiencies of Boron (B), Calcium (Ca), Iron (Fe) and Potassium (K), by using shape and texture descriptors in images of coffee tree leaves. After the acquisition of images containing coffee tree leaves, they are subjected to a segmentation process using Otsu's method. Afterwards, for the resulting images they are applied the descriptors Blurred Shape Model (BSM) and Gray-Level Co-occurrence Matrix (GLCM) for extracting characteristics of shape and texture. Finally, the obtained image representation is used for training KNN, Naïve Bayes and Neural Network classifiers by using the extracted features, in order to infer the type of deficiency presented in each analyzed image. The experimental results show that the developed procedure has a high accuracy, being the better results associated to the identification of Boron (B) and Iron (Fe) deficiencies.