Non-destructive Fertility Detection of Multiple Chicken Eggs Using Image Processing and Convolutional Neural Network

Abstract

This paper presents a non-destructive fertility detection of multiple chicken eggs in incubation industry based on image processing technique and convolutional neural network (CNN). The aim of this research were to design and implement the system for simultaneously distinguish multi eggs infertility from fertility one. LED light source setting up for illumining the 48 eggs, consisting of both egg types and randomly placed on a tray in dark box. In addition, a pre-trained CNN is performed to classify fertile and infertile eggs. All eggs is captured and processed to extract a region of interest (ROI) for each egg, generating large ROI egg images dataset which used to train and test the CNN model. A designed system is programmed using Python, operating on Windown7-64bit supported by OpenCV and Keras. Experimental results showed the accuracies for fertile incubated eggs detection between day 7 and day 9 reaches 100%. Meanwhile, eggs location has 100% of accuracy is also observed. Hence, the proposed technique are high reliability, high accuracy system and suitable to use in real application.