Abstract

Zhihui Zhu and Meihu Ma

The ability to automatically identifying fertile eggs prior to incubation would allow timely removal of the infertile eggs, which could bring high profits to hatcheries with better chick quality and lower pathogen contamination of chicks. A method based on machine vision and least square support vector machine (LS-SVM) for fertile eggs identification prior to incubation was proposed. Digital images were acquired by high-resolution digital cameras with cold light back illumination, and egg shapes (e.g. egg shape index, roundness, elongation, geometric moment) and color mean information of the egg yolk region such as hue (H), intensity (I), saturation (S) from image characters were extracted. LS-SVM algorithm was used to establish fertile egg classification model from infertile eggs. The test results obtained from the 40 testing sets showed that the best classification accuracy was 92.5%. With using a same data set, the performance comparison between LS-SVM classifier with different kernel functions and the other different classifiers was conducted. Compared with other kernels, LS-SVM classifier with radius basis functional (RBF) kernel was found to obtain the best accuracy and provide better accuracy, higher speed compared with support vector machine (SVM) and back-propagation (BP) artificial neural networks classifier.