Automated leaf disease detection in different crop species through image features analysis and One Class Classifiers

The presented approach demonstrates an automated way of crop disease identification on various leaf sample images corresponding to different crop species employing Local Binary Patterns (LBPs) for feature extraction and One Class Classification for classification. The proposed methodology uses a dedicated One Class Classifier for each plant health condition including, healthy, downy mildew, powdery mildew and black rot. The algorithms trained on vine leaves have been tested in a variety of crops achieving a very high generalization behavior when tested in other crops. An original algorithm proposing conflict resolution between One Class Classifiers provides the correct identification when ambivalent data examples possibly belong to one or more conditions. A total success rate of 95% is achieved for the total for the 46 plant-condition combinations tested.