Semantic-Search: A Knowledge-Driven Classification Method for Plant Diseases

To efficiently detect plant diseases, A-CPS have been integrated with image processing mechanisms built on deep learning and CNNs. But these methods are Data Driven and need lots of labeled data per class. These models learn features specific to each class and need to be retrained to add any new class. In contrast, we present a Knowledge-Driven approach where the CNNs are used for cognitive computing so that it can semantically understand the patterns of the disease on the leaf. These semantics (features) will be used to query a manually created database containing semantic descriptions of diseases for disease classification. Since the model has cognitive abilities to describe the presented disease, adding a new class will need a simple database update instead of retraining on numerous images. The proposed Semantic-Search was validated on the PlantVillage database with 2000 images of 20 different diseases from 5 types of plants and achieved an accuracy of 94%.