Assessment of hatching properties for identifying multiple duck eggs on the hatching tray using machine vision technique

Abstract

Since the hatching characteristics (fertile and infertile) of multiple duck eggs on the hatching tray were difficult to be recognized intelligently, a method of identifying infertile duck eggs by machine vision was proposed in this study. Based on the image analysis of the duck eggs hatched for 7 days, region of interest (ROI) mask was created to extract the image feature parameter vector with 4 features. This study investigated the optical change of duck eggs during different hatching time, and used Linear Discriminant Analysis (LDA), Naïve Bayes (NB) and Support Vector Machine (SVM) methods to establish discriminant models for different incubation periods. SVM classifier yielded the best results with 4 features for validation set on day 5. Classification accuracy, sensitivity (SE), specificity (SP), positive predictive value (PPV) and negative predictive value (NPV) are all 100%. Compared with artificial candling recognition, the identification time for the fertility and infertility was 2 days earlier using SVM model. The prediction accuracy for the new unknown duck eggs was 92.06% by the established SVM model of day 5. Machine vision techniques combined with the SVM model of high classification accuracy on day 5 provided an appropriate method to identify the infertile duck eggs on the hatching tray.