Comparative Study of Pre-Trained CNN for Pinkeye Disease Detection in Cattle and Goat Eyes

Keratoconjunctivitis, commonly known as "pinkeye" disease, significantly impacts the health and productivity of cattle and goats. This study conducts a comparative analysis of three pre-trained convolutional neural network (CNN) models - MobileNetV2, VGG16, and DenseNet - for classifying pinkeye disease images in livestock. The research comprises four phases: data collection, data preparation, modeling, and evaluation. Data collection involves a dataset of 1425 images gathered from various sources. Preprocessing includes resizing, augmentation, and dataset stratification. The study creates two CNN models tailored for cattle and goats using pre-trained CNN. MobileNetV2 consistently demonstrates superior generalization, surpassing 95% accuracy. In contrast, VGG16 and DenseNet201, while achieving higher overall accuracy, show overfitting. MobileNetV2 is identified as the most proficient model for pinkeye disease classification, advancing automated disease diagnosis in cattle and goat farming. Further validation in diverse operational contexts is recommended.