Identifikasi penyakit pada daun labu (Cucurbita moschata) secara manual masih mengandalkan pengamatan visual siswa di lahan praktik yang bersifat subjektif dan rawan kesalahan, sehingga dapat menyebabkan keterlambatan penanganan serta kerugian hasil panen. Penelitian ini bertujuan merancang dan membangun aplikasi Android berbasis deep learning untuk mengklasifikasikan kondisi daun labu secara otomatis menggunakan algoritma Convolutional Neural Network dengan arsitektur MobileNetV2. Metode pengembangan sistem yang digunakan adalah Prototype. Dataset terdiri dari 5.000 citra yang terbagi ke dalam empat kelas, yaitu Daun Sehat, Bercak Daun, Embun Tepung, Layu Fusarium, dengan proporsi 80% data latih, 10% data validasi, dan 10% data uji. Model dilatih menggunakan Transfer Learning dua fase, yaitu Feature Extraction dan Fine-Tuning, kemudian dikonversi ke format TensorFlow Lite agar dapat berjalan secara luring pada perangkat Android. Hasil evaluasi menunjukkan performa tinggi dengan akurasi mencapai 98,80%. Pengujian Black Box terhadap 16 skenario seluruhnya valid, sedangkan UAT bersama 20 responden siswa SMK Negeri 1 Kuningan memperoleh kelayakan sebesar 85%, sehingga aplikasi dinilai layak digunakan di lapangan.

Manual identification of diseases on pumpkin leaves (Cucurbita moschata) still relies on students visual observation, which is subjective and prone to errors. This condition may lead to delayed treatment and potential crop yield losses. This study aims to design and develop an Android-based deep learning application capable of automatically classifying pumpkin leaf conditions using a Convolutional Neural Network algorithm with the MobileNetV2 architecture. The system development method used in this study is the Prototype method. The dataset consists of 5,000 images divided into four classes: Healthy Leaf, Leaf Spot, Powdery Mildew, and Fusarium Wilt, with an 80% training, 10% validation, and 10% testing split. The model was trained using a two-phase Transfer Learning approach, namely Feature Extraction and Fine-Tuning, and was then converted into TensorFlow Lite format to enable offline operation on Android devices. Evaluation results demonstrate high performance, achieving an accuracy of 98.80%. Functional testing using Black Box Testing on 16 scenarios showed that all scenarios were valid. User Acceptance Testing involving 20 students from SMK Negeri 1 Kuningan obtained a feasibility score of 85%, indicating that the application is feasible for field use.