Electronic Theses and Dissertation

Penyakit pada daun padi sering menurunkan kualitas pertumbuhan tanaman, sehingga diperlukan deteksi dini untuk mencegah kerugian hasil panen. Model klasifikasi penyakit daun padi menggunakan arsitektur MobileNetV2 yang dioptimalkan agar dapat diterapkan pada perangkat seluler. Dataset yang digunakan terdiri dari 1.920 citra daun padi yang mencakup enam kelas, yaitu Bacterial Leaf Blight, Brown Spot, Leaf Blast, Leaf Scald, Narrow Brown dan daun sehat. Tahapan penelitian meliputi pre-processing gambar, augmentasi, pembagian data, pelatihan model, serta evaluasi menggunakan metrik akurasi, precision, recall dan F1-score. Eksperimen dilakukan dengan variasi parameter epoch dan learning rate untuk mendapatkan konfigurasi terbaik. Hasil penelitian menunjukkan bahwa MobileNetV2 dengan learning rate 10⁻⁴ dan 100 epoch memberikan performa paling optimal dengan akurasi mencapai 97 persen serta nilai precision, recall dan F1-score yang stabil pada hampir seluruh kelas. Model ini berpotensi diintegrasikan ke dalam aplikasi berbasis TensorFlow Lite sebagai alat deteksi dini penyakit daun padi yang cepat, efisien dan praktis digunakan di lapangan, sehingga mendukung peningkatan produktivitas pertanian.

Diseases on rice leaves often reduce the quality of plant growth, making early detection necessary to prevent crop losses. This study presents a rice leaf disease classification model using the MobileNetV2 architecture optimized for deployment on mobile devices. The dataset consists of 1,920 rice leaf images covering six classes, namely Bacterial Leaf Blight, Brown Spot, Leaf Blast, Leaf Scald, Narrow Brown, and healthy leaves. The research stages include image pre-processing, augmentation, data splitting, model training, and evaluation using accuracy, precision, recall, and F1-score metrics. Experiments were carried out using variations of epoch and learning rate parameters to obtain the best configuration. The results indicate that MobileNetV2 with a learning rate of 10⁻⁴ and 100 epochs provides the most optimal performance, achieving an accuracy of 97 percent along with stable precision, recall, and F1-score values across almost all classes. This model has the potential to be integrated into a TensorFlow Lite–based application as a fast, efficient, and practical early detection tool for rice leaf diseases in the field, supporting increased agricultural productivity.