KLASIFIKASI CITRA FESES UNTUK DIAGNOSA PENYAKIT BURUNG MENGGUNAKAN ALGORITMA CONVOLUTIONAL NEURAL NETWORK (CNN) RESNET-50

Poultry diseases are a serious issue in the livestock industry, as they can reduce productivity and cause significant economic losses. One promising approach for early disease detection is the analysis of bird feces images. This study aims to classify four bird health conditions—healthy, coccidiosis, salmonella, and Newcastle disease—based on feces images using a Convolutional Neural Network (CNN) algorithm with the ResNet50 architecture and transfer learning. The dataset consists of 6,815 images collected from the Kaggle platform. The research was conducted in two scenarios: one without augmentation and the other with data augmentation applied to the Newcastle disease class, which had the fewest samples. The results show that augmentation significantly improved model performance, as reflected by an increase in the F1-score of the Newcastle disease class from 0.88 to 0.95, and an improvement in overall accuracy from 96% to 97%. Additionally, augmentation helped reduce overfitting and enhanced the model's generalization to unseen data. Based on these findings, the ResNet50 architecture with data augmentation proves effective for feces image classification in supporting bird disease diagnosis.