Rice is a major food commodity in Indonesia that plays a vital role in maintaining national food security. However, rice productivity often declines due to pest and disease attacks, especially when the disease is not detected early. Currently, the process of identifying rice diseases is generally still carried out manually by farmers or experts through direct observation, which is subjective, time-consuming, and prone to identification errors. To overcome these limitations, a technology-based solution is needed that is able to detect rice diseases automatically, quickly, and accurately. This study aims to develop a rice disease detection system based on leaf images using a deep learning approach with the Visual Geometric Group (VGG-19) architecture. The research method used is experimental by comparing the performance of the VGG-19 architecture using three different types of optimizers, namely Adaptive Moment Estimation (ADAM), Root Mean Square Propagation (RMSProp), and Stochastic Gradient Descent (SGD), to obtain the best accuracy in rice disease classification. The findings show that the combination of VGG-19 with the ADAM optimizer produces the highest accuracy of 96.45%, followed by RMSProp at 95.96% and SGD at 87.08%. These findings indicate that the selection of optimizers plays an important role in improving the performance of deep learning models, especially in detecting rice diseases based on leaf images.