A framework for crack detection of fresh poultry eggs at visible radiation

Abstract

Visible radiation technology is increasingly being used for inspection of crack in the eggshell. However, the complexity of the eggshell together with the heterogeneity of the egg internal structures, resulted in very textured image especially when view under candling light. Hence, the inspection process is technically very challenging. In an attempt to improve the inspection performance, a refined anisotropic diffusion filter together with the double thresholding algorithm are used to morphologically segment crack pixels from the background. In this case a novel technique based on Radon transform is developed for feature extraction while the classification is established via a multiclass Support Vector Machine (SVM). Experimental results indicate that the proposed framework performs well on eggs from same or different poultry houses with sensitivity and specificity, averaging at 89.2% and 94.6% respectively. The ROC analysis produced 100% correct classification of intact or unbroken eggs. However, the performance decreased slightly when inspecting different types of cracked eggs with false positive ranging from 3 to 11% due to high degree of similarity between groups.