<![CDATA[Rice leaf diseases can cause a significant decrease in productivity if not treated early, while currently still using a manual diagnosis process that is often slow, inconsistent, and dependent on extension workers. In this study, the SRIKANDI application was developed, a mobile application for image processing for rice leaf diseases using the NASNetMobile Convolutional Neural Network (CNN) architecture. This system is designed using five labels, namely bacterial, blast, brownspot, leafsmut, and healthy leaves. The dataset used consists of 2500 images collected from Kaggle, Mendeley Data, and taken directly. All images go through preprocessing stages of resizing, pixel normalization, and augmentation, then divided into 80% train, 10% test, and 10% validation. The model training was carried out in two stages, namely, 40 epochs of fine-tuning with a learning rate of 0.0008 followed by 20 epochs of fine-tuning with a learning rate of 1e-5, the results obtained by the model with a test set accuracy rate of 96.40%. The trained model is then saved in TFLite format to be integrated into the SRIKANDI mobile application so that it can help farmers detect rice leaf diseases in real-time via camera or taken from the gallery.]]>
Copyrights © 2026
