Abstract:
Feeding the people and preventing the spread of diseases among crops depend on the
early diagnosis of diseases in agricultural crops. This paper employs transfer learning to
automate the classification of rice leaf diseases, particularly those common in the area as
bacterial leaf blight and brown spot. In order to assess the effectiveness of the
aforementioned Deep Learning (DL) models, a dataset including 692 photos of brown
spots, 700 photographs of healthy rice, and 700 images of bacterial leaf blight is
employed in the vicinity of the Pabna Bangladesh Agriculture Development Corporation
(BADC) & Jamalpur district paddy field. Data collection, sample preparation, labeling,
image processing, selecting the best model, training, and testing are all included in the
established technique. The models over fit and produced varying accuracy rates, with
DenseNet201 and MobileNetV2 recording 98.89% and 98.57%, respectively, according
to the findings analysis. The findings are discussed with a focus on model architecture
and choice as crucial elements in the success of disease classification. The anticipated
future efforts will focus on investigating imbalanced categorization and ensemble
learning. To sum up, this study contributes pertinent knowledge to the advancement of
automated disease identification in agriculture, which is important for global food
security
