CNN-Based Rice Leaf Disease Classification Applying Augmentation Techniques for Improved Accuracy: A Deep Learning Approach

Early detection of rice leaf diseases is critical for crop protection and yield optimization. This paper presents an automated classification system for three prevalent rice diseases—Bacterial leaf blight, Brown spot, and Rice—blast using Inception V3 convolutional neural network with comprehensive data augmentation. A dataset of 330 RGB images collected from Central Luzon rice fields was partitioned 80:20 for training and testing. Data augmentation techniques (rotation$\pm 20^{\circ}$, flipping, zoom$0.9-1.1 \times$, brightness$\pm 20 \%$) expanded the effective training set threefold. The Inception V3 transfer learning model achieved 96% classification accuracy after 50 epochs, compared to 93% at 30 epochs, with training loss reduced from 0.19 to 0.10 and validation loss from 0.54 to 0.40. Correct predictions exhibited confidence scores$>0.99$, while misclassifications showed notably lower confidence (0.61-0.79), enabling uncertainty-aware decision support. The model's 87 MB footprint and 0.15 s inference time enable mobile deployment for field use. Results demonstrate that transfer learning with systematic augmentation provides an efficient, accurate solution for rice disease diagnosis in resource-constrained agricultural settings.