Pathogen-based Classification of Plant Diseases: A Deep Transfer Learning Approach for Intelligent Support Systems

The national economy's key pillar, agriculture has a significant influence on society. Plant health monitoring and disease detection are essential for sustainable agriculture. To protect plants against pathogen damage, farmers must be able to detect an infection prior to its obviousness. Effective plant disease detection technique can greatly lessen the use of toxic chemicals thereby aiding a better environment. For diseases to be managed effectively, plant pathogens must be accurately detected. The pathogens that cause plant diseases include bacteria, fungi, viruses, oomycetes, nematodes, phytoplasmas, protozoa, and parasitic plants. In this paper pathogen-based plant disease detection is done. An automated plant disease detection and its classification are done along with identifying the pathogen responsible for it using keras transfer learning models. This is done by considering Agri-ImageNet dataset as well as images of leaves, bulb, and flowers of sunflower and cauliflower captured in a natural realistic environment. This dataset overcomes the drawback of Plant Village dataset in which images are captured in homogeneous backgrounds and controlled settings. However, it can be time and resource-consuming to train deep learning models. These problems can be solved by reusing knowledge representations through deep transfer learning. Main objective of this paper is to explore and analyze all the deep transfer learning models, to identify which model is best suited for plant disease dataset. This work has been carried out by using 38 deep transfer learning models to obtain best classification accuracy. EfficientNetV2B2 and EfficientNetV2B3 models gives highest accuracy in comparison with all other deep transfer learning models for sunflower, cauliflower and Agri-ImageNet datasets. Classification report is generated from the best deep transfer learning model.