Paddy crops are particularly vulnerable to many illnesses that can negatively affect output and quality, endangering both food security and farmers’ livelihoods. Conventional disease detection techniques rely on the observations of agricultural specialists; these procedures are labor-intensive, time-consuming, and frequently inaccurate. To overcome these issues, this study suggested a deep learning framework to automatically identify paddy crop illnesses. We implemented the EfficientNet B5 model for paddy crop disease detection. We created a novel real-time dataset using images from the paddy crop field in Khammam district of Telangana. All the images were resized and normalized to maintain the same feature level. We accurately identified and categorized various diseases affecting rice crops based on digital images using the state-of-the-art CNN architecture. The EfficientNet B5 model was then evaluated and compared with the traditional models in terms of accuracy, precision, recall, and loss value. The suggested model attained an accuracy of 98.35% and a loss of 0.46%, which is superior to the other existing models.