Implementasi Model MaxViT untuk Deteksi Penyakit Daun Cabai Berbasis Mobile

Nur Puspita Amalia (Universitas Negeri Surabaya, Surabaya)
Salamun Rohman Nudin (Universitas Negeri Surabaya, Surabaya)

TIN: TERAPAN INFORMATIKA NUSANTARA
Vol 7 No 1 (2026): June 2026

Chili is an important horticultural commodity in Indonesia, but its productivity often decreases due to leaf diseases such as Cercospora leaf spot, Powdery Mildew, Mites and Thrips, and Nutritional deficiency. Manual disease identification requires more time and may produce less accurate diagnoses. This study aims to develop a mobile-based chili leaf disease detection system using the Multi-Axis Vision Transformer (MaxViT) model. The dataset consisted of 10,987 chili leaf images divided into five classes and split into training, validation, and testing data with a ratio of 70:15:15. Model training was carried out using four optimizer scenarios, namely a standard baseline model, SGD, Adam, and AdamW. The result showed that the Adam optimizer archieved the best performance with 99,45% accuracy, 99,44% precision, 99,45% recall, and 99,45% F1-Score. The best model was converted into TensorFlow Lite format with a file size of 32.0 MB and deployed in a mobile application. The application can detect diseases through camera capture or gallery images and provide prediction results along with disease descriptions and treatment recommendations. Testing results indicate that the system performs well under various usage conditions. This system is expected to help users indentify chili leaf diseases quickly, practically, and accurately.