Currently, in rice production areas, the detection and classification of leaf diseases is often performed manually by technical personnel, which can lead to identification errors and affect crop yield. These drawbacks can be reduced through the use of computer vision systems integrated with deep learning algorithms. In this study, different techniques based on convolutional neural networks (CNNs) were analyzed to classify rice leaf diseases, and two approaches were selected: a custom architecture and the pre-trained Xception model. The classes considered were: leaf with brown spot, healthy leaf, leaf with hispa, leaf with blight, and scald leaf. The performance of both architectures was evaluated under different configurations. For each treatment, their performance in precision, recall, and accuracy were measured. In the end, the best performance was obtained by the Xception model with fine-tuning and the RMSprop optimizer, reaching an accuracy of 93.8 %. These findings may be useful for automated agricultural applications, such as the application of crop protection products.