PENGEMBANGAN APLIKASI WEB UNTUK KLASIFIKASI PENYAKIT DAUN PADI MENGGUNAKAN TRANSFER LEARNING DENGAN ARSITEKTUR MOBILENETV2

This study aims to develop a rice leaf disease classification model using the transfer learning method with the MobileNetV2 architecture and to analyze the effects of increasing the number of disease classes and applying segmentation techniques on model performance. The dataset consists of eight classes of rice leaf diseases, namely Bacterial Leaf Blight, Brown Spot, Leaf Blast, Leaf Scald, Narrow Brown Leaf Spot, Rice Hispa, Sheath Blight, and Healthy Rice Leaf, comprising a total of 6,527 images divided into training, validation, and testing sets. Three pre-processing skenarios were evaluated: Resize, HSV segmentation, and GrabCut segmentation. The results show that the Resize skenario achieved the best performance with 100% validation accuracy and 94% testing accuracy, while HSV and GrabCut segmentations did not significantly improve classification results. The MobileNetV2 architecture proved to be efficient with 2.4 million parameters and a model size of 26 MB, making it suitable for deployment on resource-limited devices. Visualization using Gradient-weighted Class Activation Mapping (Grad-CAM) demonstrated that the model effectively fokused on relevant leaf regions associated with disease symptoms. The best-performing model was integrated into a Streamlit-based web application, achieving an average prediction time of 0.478 seconds per image. These findings indicate that the transfer learning approach using MobileNetV2 is effective, efficient, and practical for real-time rice leaf disease detection in the field.