Automatic classification of nutritional deficiencies in coffee plants

Classification of nutritional deficiencies, in coffee plants, is a problem for coffee farmers, since they do not have the knowledge to identify nutritional deficiencies neither receive technical assistance. Moreover, the performance of automatic classification of nutritional deficiencies based on digital image processing is affected by changes in image appearance such as: illumination, noise and acquisition conditions. In addition, some nutritional deficiencies have similar visual characteristics, such as: magnesium, manganese and iron, doing difficult to distinguish among them. In this paper, Random Forest, a machine learning technique based on decision trees, is used to classify automatically nutritional deficiencies in coffee plants, using local and global features to build the classification model. The Scale-Invariant Feature Transform (SIFT) algorithm is used to extract local features. Global features are defined based on shape and color characteristics of nutritional deficiencies. Experimental evaluation was performed using 335 images of coffee leaves with one nutritional deficiency. Results showed that global features have better performance that local features with an accuracy of 67,5%.