Over fifty percent of the global population relies on rice as a fundamental grain, making it crucial for worldwide food security. Its nutritional value, versatility, and adaptability make it indispensable for different diets and agricultural economies. Classification and early detection of rice diseases are crucial for timely action, reducing crop damage, and ensuring food security by maintaining high agricultural productivity. Computer-aided methods to detect rice leaf diseases have been introduced to overcome the limitations of manual detection. They enable faster and more accurate identification and timely intervention to prevent significant crop loss and ensure sustainable agricultural practices. Previous studies have identified diseases in rice plants. In our experiment, we attempted to improve the accuracy disease detections and sub-classified them into mild and severe. It can detect and classify the disease according to the following labels: Bacterial Blight, Blast, Brownspot, and Tungro. We tested different convolutional neural network architectures to determine which architecture provided the best accuracy. We used Inception-V3, Xception, Densenet, MobileNet-V2, VGG-16, VGG-19, Inception Resnet, NASNet Mobile, and an ensemble model combining Xception and Inception-V3. We have seen that the ensemble model outperformed all the other models for this dataset, achieving the best results in all evaluation metrics with an accuracy of 93.33%, which has emerged as the best among all the tested methods for this experiment.