Comparative Study on Various CNNs for Classification and Identification of Biotic Stress of Paddy Leaf

Rice diseases triggered by exposure to an infectious agent like fungal, or bacterial give rise to crop loss globally. Hence, early disease identification could mitigate this problem. However, it is quite difficult to identify the diseases with eyeball inspection. To overcome this problem, there is needed an automated system that can correctly identify the diseases without human intervention. In the current work, a proposed CNN model is developed and trained with 3799 images of four classes. These four classes include LeafBlast, BrownSpot, Hispa, and Healthy. The proposed CNN model is compared with other state-of-the-art transfer learning models to find its efficiency. The experimental results show that the model outperforms the existing transfer learning models by acquiring accuracy of 99.57%.