Egg Defects Detection Using Bidirectional Long Short-Term Memory Network

Abstract

Eggs are a vital nutritional resource globally, making their inspection crucial for maintaining quality and ensuring food safety. Motivated by this need, a machine vision system incorporating deep learning techniques was developed to detect egg defects. The machine vision system has a rotating mechanism that allows for comprehensive visualization of the egg surface from various angles, leading to a more accurate assessment. The proposed system leverages deep feature extraction using a pre-trained convolutional neural network and analyses these features with a Bidirectional Long Short-Term Memory (BiLSTM) network. The types of egg defects that this study aimed to detect were bloodstained, cracked and dirty eggs. A total of 400 eggs sample were used with 6 images per egg resulting with a dataset of 2400 images for the proposed deep learning method. The performance evaluation of the model revealed an accuracy of 97.71% in detecting egg defects, with a recall score of 0.9788, a specificity score of 0.9926, a precision score of 0.9770, and a F1 score of 0.9770. Comparisons with other state-of-the-art deep learning and machine learning methods like (SVM, VGG16, YOLOv5) indicate that the proposed model has certain advantages and does not differ much in terms of accuracy for defect detection. The result from this study demonstrates the potential of sequential feature modelling for robust egg defect detection.