DETECTION OF CRACKS IN EGGS USING COLOR COMPUTER VISION AND ARTIFICIAL NEURAL NETWORKS

A color vision system was used to obtain images of single stationary e ggs. Images of cracked eggs and grade-A eggs were used to generate tra ining, testing, and validating data for an artificial neural network. Three histograms of the primary colors of red, green, and blue were co nstructed from each color image. A composite histogram was constructed by concatenating each primary histogram and, optionally, reducing the number of cells. Various artificial neural networks with different pa rameters and structures were trained using the composite histograms. T he artificial neural networks were tested and validated on independent data sets. The artificial neural networks trained with histograms of color images produced graded samples that exceeded USDA requirements a nd achieved higher accuracy than neural networks trained on data from gray scale images.