Electronic Theses and Dissertation

ABSTRAK
Penyakit pada daun kopi Arabika dapat menurunkan produktivitas panen. Klasifikasi penyakit tersebut telah dilakukan dalam penelitian sebelumnya dengan mengembangkan model berbasis convolutional neural network (CNN), yaitu EfficientNet-B0, ResNet50, dan VGG16, yang masing-masing memperoleh akurasi awal sebesar 97,50%, 96,66%, dan 87,50%. Namun, implementasinya ke dalam aplikasi yang dapat digunakan langsung oleh petani masih minim, dan sebagian besar aplikasi yang telah dikembangkan bergantung pada koneksi internet. Karena itu, dibutuhkan aplikasi yang menampilkan hasil klasifikasi penyakit daun kopi yang dapat dijalankan oleh pengguna akhir tanpa ketergantungan pada internet. Penelitian ini bertujuan mengembangkan aplikasi berbasis Edge AI yang mampu menampilkan hasil klasifikasi empat penyakit daun kopi, yaitu leaf rust, sooty mold, brown eye spot, dan black rot, dari ketiga model CNN tersebut secara real-time tanpa internet. Model dikonversi ke format ONNX agar dapat dijalankan pada perangkat edge Jetson Orin Nano, dengan hasil klasifikasi ditampilkan melalui antarmuka Streamlit. Pengembangan aplikasi dilakukan menggunakan metode system development life cycle (SDLC) dan dievaluasi dengan black-box testing, performance testing, usability testing, dan user acceptance testing (UAT). Hasil pengujian menunjukkan seluruh 8 skenario black-box testing berjalan sesuai rancangan aplikasi. Pengujian usability (SUS) terhadap 10 responden memperoleh rata-rata skor 73,t75, termasuk kategori Baik. Pengujian performance testing menunjukkan aplikasi mampu menjalankan inferensi secara real-time pada perangkat edge dengan rata-rata waktu inferensi dibawah 1 detik per gambar. Pengujian UAT terhadap 15 responden memperoleh rata-rata persentase penerimaan sebesar 87,73%, termasuk kategori Sangat Baik. Hasil ini menunjukkan bahwa aplikasi mampu membantu petani mengklasifikasi penyakit daun kopi secara cepat, akurat, dan efisien di lapangan tanpa ketergantungan pada koneksi internet.
Kata kunci: Daun Kopi, Edge AI, ONNX, SDLC, Streamlit.

ABSTRACT Diseases affecting Arabica coffee leaves can reduce harvest productivity. Classification of these diseases has been carried out in previous studies by developing convolutional neural network (CNN) based models, namely EfficientNet-B0, ResNet50, and VGG16, which achieved initial accuracies of 97.50%, 96.66%, and 87.50%, respectively. However, the implementation of these models into applications that can be directly used by farmers remains limited, and most of the applications developed so far rely on internet connectivity. Therefore, an application is needed that can display coffee leaf disease classification results and be operated by end users without dependence on the internet. This study aims to develop an Edge AI-based application capable of displaying the classification results of four coffee leaf diseases leaf rust, sooty mold, brown eye spot, and black rot using the three CNN models in real time without internet access. The models were converted into ONNX format to enable deployment on the Jetson Orin Nano edge device, with classification results displayed through a Streamlit interface. The application was developed using the system development life cycle (SDLC) method and evaluated through black-box testing, performance testing, usability testing, and user acceptance testing (UAT). The test results showed that all 8 black-box testing scenarios ran according to the application design. Usability testing (SUS) involving 10 respondents obtained an average score of 73.75, categorized as Good. Performance testing showed that the application was able to perform real-time inference on the edge device, with an average inference time of under 1 second per image. UAT involving 15 respondents obtained an average acceptance percentage of 87.73%, categorized as Very Good. These results indicate that the application can help farmers classify coffee leaf diseases quickly, accurately, and efficiently in the field without relying on an internet connection. Keywords: Coffee Leaf, Edge AI, ONNX, SDLC, Streamlit