Detection of Eggshell Defects using Convolutional Neural Networks

Abstract

In commercial egg farming industries, the automatic sorting of defective eggs is economically and healthily important. Nowadays, detect of defective eggs is performed manually. This situation involves time consuming, tiring and complex processes. For all these reasons, automatic classification of defects that may occur on the egg surface has become a very important issue. For this purpose, in this study, classification of egg defects was performed using AlexNet, VGG16, VGG19, SqueezeNet, GoogleNet, Inceptionv3, ResNet18, and Xception architectures, which were developed based on Convolutional Neural Networks (CNN), which provide high performance in object recognition and classification. To test the performance of these architectures, an original data set containing dirty, bloody, cracked, and intact eggs were built. As a result of experimental studies, the highest accuracy score was obtained with VGG19 architecture as 96.25%. In these results, it was observed that ESA methods achieved high success in classifying defective eggs.