CNN Optimization Using Dropout Regularization in Chicken Egg Fertility Detection

This research investigates the application of Convolutional Neural Networks (CNNs) to automate fertility detection in chicken eggs, a task traditionally performed through manual candling. Utilizing a dataset of 292 candled egg images, we initially employed a CNN-basic model, which, while achieving high training accuracy, exhibited substantial overfitting with only 75% accuracy and a high loss of 0.9525 on validation datasets. Various Dropout rates were systematically tested to enhance the model's generalization capabilities, with 0.3 identified as the optimal rate. This adjustment significantly improved the model's performance, achieving a testing accuracy of 97.92% and reducing the loss to <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{0.2095}$</tex>. The findings highlight the efficacy of tailored CNNs in accurately classifying egg fertility, offering significant advancements for poultry farming. Future efforts will refine the CNN architecture through advanced preprocessing, expanded data augmentation, and sophisticated segmentation techniques to increase robustness and extend the model's applicability across diverse agricultural environments.