A NEW LOW-COST DETECTION DEVICE FOR EARLY DISCRIMINATION OF EGGS FERTILITY USING ADVANCED STATISTICAL CLASSIFIERS

Abstract

Detecting fertility methods of hatching eggs is getting an importance with the increase in poultry breeding facilities size to remove the non-hatchable eggs which consume time, space and cost without benefits. Early detection of the infertile eggs is a vital economic issue. Fertility detection methods are expensive to be applied widely, hence this investigation aimed to study the possibility of using a low-cost device as light dependent resistor sensors in detecting the fertility of hatching eggs with high efficient at candling process. Mathematical formulas were developed in this study to discriminate the fertile and infertile eggs by the light dependent resistor sensor and interfaced with a personal computer programmed by LabView software package to execute a certain control decision (is a hatchable egg or not?) via these formulas. Different statistical classifiers have been used to classify eggs into fertile and infertile eggs like linear, quadratic and partial least squares discriminant analyses and support vector machine. According to literature three different times were appointed at earlier times of egg incubation process for fertility identification investigation of 6th, 9th and 12th day. For more identification precision, sensor position for light intensity measuring was investigated at three different measuring orientation lines against the investigated eggs at vertical, inclined 45˚ and horizontal orientation line. Classification mathematical models were developed using the previous classifiers. Principal component and partial least squares regression were used to develop multiple linear regression models for each incubation period. determination It was found by the Principal Component Analysis, that the sensor orientation line position for light intensity measuring gives different measured values of the same investigated egg, but all these measurement values have an entirely correlation relationship with the classification efficiency. The highest identification rate of 97% obtained by the classifier of linear discriminant analysis at the 6th day of the incubation period by the light dependent resistor sensor, which confirms the efficient use of this simple low-cost sensor in discrimination at earlier times of incubation period closing to the other sophisticated devices. The developed mathematical model can easily be implemented with Fuzzy logic controller; further research will be needed to accomplish the fully automated system.