Detecting Fertility of Hatching Eggs Using Machine Vision II: Neural Network Classifiers

The ability to automatically detect dead embryos or infertile eggs during early incubation would allow timely removal of such eggs from incubation thereby contributing to overall hatchery profits. Neural networks were trained to recognize shape differences in gray level histograms of images of fertile and infertile eggs at early incubation stages. Images of eggs at three incubation stages (days 4, 3, 2) were obtained using a machine vision system with backlighting. The gray level histograms of the images were enhanced and used as inputs to the neural networks. The learning and classification behavior of different configurations of networks were studied. Selected networks were tested on pattern sets not used in training. The best classification accuracies obtained on the testing sets were 93.9% at day 4, 93.5% at day 3 and 67.6% at the day 2 stage of incubation.