Cabai (Capsicum annuum) merupakan komoditas hortikultura strategis Indonesia yang produksinya terganggu oleh penyakit daun, diperparah oleh metode deteksi manual yang subjektif dan rentan terhadap kesalahan identifikasi akibat kemiripan gejala antar penyakit (overlapping symptoms), serta terbatasnya akses petani di Kabupaten Kuningan terhadap pakar patologi tanaman. Penelitian ini bertujuan merancang dan membangun aplikasi Android berbasis Convolutional Neural Network (CNN) dengan arsitektur MobileNetV3-Small yang diintegrasikan mekanisme Squeeze-and-Excitation Coordinate Attention (SCA) untuk mendeteksi penyakit daun cabai (Capsicum annuum) secara otomatis dan luring (offline). Pengembangan sistem menggunakan metode Rapid Application Development (RAD) dengan data primer berupa 1.200 citra per kelas dari lahan petani di Kuningan dan kelas bukan daun 1.200 citra, mencakup empat kelas: Daun Sehat, Keriting Daun, Virus Gemini, dan Bercak Daun (Cercospora leaf spot). Pra-pemrosesan citra meliputi resizing 224×224 piksel, normalisasi, dan augmentasi data. Pelatihan model menggunakan teknik transfer learning berbasis bobot ImageNet dengan Adam Optimizer, batch size 32, dan 30 epoch. Evaluasi dilakukan menggunakan Confusion Matrix serta metrik performa. Hasil penelitian menunjukkan bahwa model mencapai nilai akurasi sebesar 99,67%, dengan nilai Precision sebesar 99,67%, Recall sebesar 99,67%, dan F1-Score sebesar 99,67%. Aplikasi yang dihasilkan berupa file APK yang beroperasi mandiri pada perangkat Android, menjadikannya solusi diagnosis dini yang praktis bagi petani dalam menekan risiko gagal panen.

Chili (Capsicum annuum) is a strategic horticultural commodity in Indonesia whose production is hampered by leaf diseases, exacerbated by subjective manual detection methods prone to misidentification due to overlapping symptoms, and limited access to plant pathologists for farmers in Kuningan Regency. This study aims to design and develop an Android-based application using Convolutional Neural Network (CNN) with MobileNetV3-Small architecture integrated with the Squeeze-and-Excitation Coordinate Attention (SCA) mechanism to detect chili leaf diseases automatically and offline. The system development utilizes the Rapid Application Development (RAD) method with primary data consisting of 1,200 images per class from farmers' fields in Kuningan and class not leaf, covering four classes: Healthy Leaves, Leaf Curl, Gemini Virus, and Cercospora Leaf Spot. Image pre-processing includes resizing to 224×224 pixels, normalization, and data augmentation. Model training employs transfer learning techniques based on ImageNet weights with the Adam Optimizer, a batch size of 32, and 30 epochs. Evaluation is conducted using a Confusion Matrix and performance metrics. The results show that the model achieved an accuracy value of 99.67%, with Precision of 99.67%, Recall of 99.67%, and F1-Score of 99.67%. The resulting application, provided as an APK file, operates independently on Android devices, offering a practical early diagnosis solution for farmers to reduce the risk of crop failure.