Flexible Neural Network Classifier for the Automated Detection of Bones in Chicken Breast Meat

This paper discusses the use of neural networks for the classification of potential defects found in x-ray images of chicken breast meat after the de-boning process. The chicken meat is passed under a solid-state x-ray sensor which acquires a two dimensional image of the chicken breast. A series of image processing operations are applied to the acquired image, which identify certain pixel groupings (blobs) as potentially containing a bone. The image processing task is a difficult one and the resulting segmented blobs represent not only correctly identified bones but also areas caused by overlapping muscle regions in the meat which appear very similar to bones in the resulting x-ray image. A number of image processing measurements were made on each blob and these features were used as the input into a neural network classifier whose function was to differentiate between bones and non-bone segmented regions. A standard Single Multi-Layer Perceptron network was used as the initial Neural Network Architecture. Although this performed reasonably well in the classification task it lacked flexibility in the sense that all bones were treated equally. A second classification scheme used a Two Stage Neural Network (TSNN) classifier and this is shown to be far more flexible with regard to its ability to be optimized for different bone classes.