Deep Learning-Based Egg Crack Detection: Benchmarking CNN, Resnet, and Xception Models

Mrs.B RUPA DEVI SINGH 1
Associate Professor,Department ofAI&DSAnnamacharya Institute of
Technology and Sciences, Tirupati –517520, A.P.
rupadevi.aitt@annamacharyagroup.org
ORCID: 0009-0005-1298-737X
G SIVATEJA 4
UG Scholar, Department of AI&DSAnnamacharya Institute ofTechnology and Sciences, Tirupati –
517520, A.P.gaddesivateja@gmail.com
B SWETHA 2
UG Scholar, Department ofAI&DSAnnamacharya Institute of
Technology and Sciences,Tirupati – 517520, A.P.
swetha9000519@gmail.com
K NIKHITHA 5
UG Scholar, Department ofAI&DSAnnamacharya Institute of
Technology and Sciences,Tirupati – 517520, A.P.
ketheneninikhitha@gmail.com

C VISHNUVARDHAN 3
UG Scholar, Department of AI&DSAnnamacharya Institute ofTechnology and Sciences, Tirupati –
517520, A.P.vishnuvardhanchitteti7262@gmail.com

Abstract— Egg grading and quality inspection have a crucial role in maintaining food safety and avoiding any economic losses in poultry rearing. Manual inspection of eggs for cracks cannot be very effective as this is a time consuming, laborious, and erratic method, especially inscenarios where large quantities of eggs have to be inspected regularly. To deal with this problem, this manuscript proposes a framework for automatic crack detection of eggs using deep learning and image processing techniques. Images of eggs are taken, and appropriate preprocessing techniques are applied to these images to enhance the features of the egg shell surface. A classification system using Convolutional Neural Network (CNN) classifier, ResNet, and Xception models is designed to classify the eggs as normal or cracked eggs. Anadditional module using a YOLOv-based detection model to find the cracked location in eggs with a bounding box around the particular region of interest is also included. The suggested method is assessed using appropriate performance measures suc as accuracy, precision, recall, and F1-score. Experimental study results showed that the recognition ability of the proposed transfer learning model is enhanced compared to the basic CNN, while the YOLOv model ensures fast localization of the cracks. Such a system ensures a non-destructive, economical, and automated method for smart poultry farms and the egg processing industry.Index Terms— Egg inspection, Crack detection, Deep learning, CNN, ResNet, Xception, YOLOv, Computer vision.