Evaluating the quality of pellet component by using image processing technique with neural networks

A new method is proposed to evaluate the quality of pellet component. Unlike other methods which only rely on the binarized image to solve this problem, ours rely on both the binarized image for measuring and other features of the image. In this paper, first of all, a model is constructed. The size of the model depends on the designed requirement of dimension and its allowable tolerance. By using this model, the sizes of the component are inspected roughly. Then, we propose the backpropagation (BP) neural networks for evaluating the ring of central part based on the features of the tested component. Because there are measurement errors such as systematic and random errors, so the dimension of the component has to be measured several times. Finally, the final decision method is proposed. This method can evaluate the quality of pellet component more precisely than the traditional methods.