Enhancing Rice Disease Prediction Through Pre-trained CNN Architectures

Rice is a basic food for more than half of the world's population, which highlights the crop's significance in global agriculture. But growing rice is a difficult undertaking, with the most significant obstacle being the constant barrage of biotic stresses-most notably diseases-that jeopardize crop output and health. This research paper examines how well-trained Convolutional Neural Network (CNN) architectures can improve rice leaf disease prediction. In this paper, multiple pre-trained CNN models (VGG16, Inception, MobileNetV2, DenseNet, and Xception) are assessed using transfer learning on a dataset of images of rice leaves with different illnesses. Main goal of this paper is to determine which model works best for correctly categorizing rice leaf diseases from their visual symptoms. Authors have determined, via thorough testing and analysis that Xception performs better than other architectures in terms of correctly forecasting rice leaf illnesses with 86.90% and 75.00% accuracy on training and testing data, respectively. This research provides important insights into the use of pre-trained CNN architectures for better rice disease diagnosis.