Egg Weight Estimation by Machine Vision and Neural Network Techniques (A case study Fresh Egg)

Abstract

Egg weight measurement is one of the most important parameters in marketing this product. Information regarding egg weight is not only vital for grading systems based merely on weight, but it is also necessary for assessing quality indices such as yolkalbumen ratio, shell thickness and hatchability. In the present study a machine vision system combined with artificial neural network technique was used for estimating egg weight. The system hardware consists of a CCD camera, a capture video, an illumination system and a mirror. As an egg is introduced into the frame, grabber two perpendicular images are grabbed. These
images are then processed in MATLAB and pixel data corresponding to each image edge is extracted. Once center of gravity of each image edge is obtained, twelve size features can be calculated for each image. These features are then classified into three categories named as input vectors (1-3). Each input vector along with its real weight data (measured) is exported to three parallel training algorithms of a Multi Layer Perceptron (MLP) Network. The training algorithms are variable learning rate (MLP-GDX),resilient back propagation (MLP-RP) and scaled conjugate gradient (MLP-SCG). These training algorithms were optimized to
estimate egg weight. Evaluation results showed that MLP-SCG was superior to other two algorithms in estimating egg weight at high accuracy (R=.96). In other words, MLP-SCG was capable of egg weight estimation at an absolute error of no more than 2.3g for the average egg size of 60 g.