Deep Learning Based Detection Of Laying Hen Health Status From Excreta Images Using MobileNetV2: Deteksi Status Kesehatan Ayam Petelur Berbasis Deep Learning Dari Citra Ekskreta Menggunakan MobileNetV2

Muhammad Nidhomun Ni'am (Program Studi Produksi Ternak, Universitas Muhammadiyah Karanganyar)
Sri Widiastuti (Program Studi Magister Ilmu Peternakan, Universitas Gadjah Mada)
Wildan Deni Fahrezi (Program Studi Teknik Komputer, Universitas Muhammadiyah Karanganyar)
Thoriqul Irfah Al-Huda (Program Studi Magister Ilmu Peternakan, Universitas Gadjah Mada)
Abdul Karim Muqofi (Program Studi Peternakan, Universitas Muhammadiyah Karanganyar)
Rizal Aji Mustofa (Program Studi Produksi Ternak, Universitas Muhammadiyah Karanganyar)
Arib Zainul Muafi (Program Studi Produksi Ternak, Universitas Muhammadiyah Karanganyar)

CITIZEN: Jurnal Ilmiah Mulitidisiplin Indonesia
Vol. 6 No. 2 (2026): CITIZEN: Jurnal Ilmiah Multidisiplin Indonesia

Early and accurate disease detection is a critical challenge in modern poultry farming. This study aimed to develop and evaluate a deep learning-based classification system using MobileNetV2 Convolutional Neural Network (CNN) architecture for automated detection of poultry diseases from excreta images only, and to validate model predictions against laboratory microbiological analyses. A total dataset of 8,087 labeled excreta images was compiled across four health categories: Healthy, Salmonella, Coccidiosis, and Newcastle Disease, and subsequently split into training (6,471) and validation (1,616) subsets at an 80:20 ratio. The MobileNetV2 model was trained over eight epochs with data augmentation strategies and evaluated using precision, recall, F1-score, accuracy, and confusion matrix analysis. The model achieved an overall accuracy of 91%, with the highest per-class F1-score for Coccidiosis (0.97) and the lowest for Newcastle Disease (0.75). The CNN MobileNetV2 architecture demonstrates strong potential for real-time, non-invasive poultry disease monitoring.