Duck egg embryonic development classification using transfer learning and CNN

Abstract

Duck eggs are a vital source of food and income for many Filipino households. However, in small to medium-sized poultry farms, farmers manually inspect eggs for quality during incubation, which can be laborious and prone to errors. This study aims to automate the classification process of duck eggs based on their stage of embryonic development (fertilized, unfertilized, or rotten) using image processing and deep learning techniques. A dataset of 9600 images of candled duck eggs were preprocessed using MPSO-CLAHE and applied uniform background transformation. The generated datasets were used to train CNN models based on AlexNet, VGG16, InceptionV3, ResNet50, and Xception. The VGG16 model exhibited superior performance with a training accuracy of 98.85%, validation accuracy of 98.81%, and testing accuracy of 97.40%. These initial results show the potential of this methodology to streamline production process and enhance the quality of duck egg products.