Rice is the main staple food in many Asian countries and the quality of these crops play an important part in security. Therefore, the ability to timely identify and accurate diagnose common rice diseases such as bacterial leaf blight, rice brown spot and leaf smut is highly desirable. This paper compares automatic plant disease identification methods using deep convolutional neural networks and targeting for mobile platforms. A dataset of 120 images of the three common diseases samples was collected, and the model was trained to identify them on rice paddy. As a result it can effectively detect and recognize those classes of rice diseases with an accuracy of 81.87% and 81.25% on the training set and the validation set respectively.