Intact egg freshness quality inspection using neural networks

This paper discusses two neural network architectures that were used to inspect the freshness quality of eggs for the poultry industry without opening the egg shell. One is the popular multilayer perceptron neural network using error back-propagation training algorithm. The other is a modular multilayer perceptron neural network with a fusion module in the second layer. Training was done module by module using back-propagation algorithm. Images of illumination profile of eggs, provided by a back-lit cold white light source were used as inputs to the system. The images of eggs with different freshness quality were captured by a vision system and the inputs were pre-processed for the neural networks. The results for both methods were encouraging and robust with good accuracy for additive noise. Training time was relatively faster in the modular approach