Detection of Broken and Good Eggs Using Image Processing and Deep Learning

This paper presents a method for detecting broken and intact eggs using image processing and deep learning. The proposed approach begins with the creation of an image dataset containing both good and broken eggs. A total of 457 images were collected, from which 30 eggs were extracted, annotated, and cropped, resulting in a dataset of 13,710 egg samples.The methodology involves two key pipelines , a deep learning-based classification and a saliency-based analysis. In the classification pipeline, two models were utilized. A TensorFlow-based convolutional neural network (CNN) was trained on the processed dataset, achieving an accuracy of 94.88%. Additionally, YOLOv8, a real-time object detection model, was employed and attained an accuracy of 90.00% for egg classification. In the second pipeline, a zero-shot saliency score technique was applied to highlight and quantify the presence of cracks in eggs without requiring model training. The saliency score correlates with surface damage, where any score greater than 0% indicates a broken egg.This dual-method framework combining supervised classification and unsupervised saliency analysis offers a robust and automated solution for egg quality inspection, with potential applications in agricultural and industrial automation systems.