Counting Steel Rods Online Using LQV Neural Network in Real-time Images

Although various modifications of learning vector quantization had been presented in many literatures, the applications about them in the industries are relatively few, especially in the steel industries. This paper presents a method of on-line measurement of the amount of steel rods with machine vision based on the learning vector quantization neural network (LVQNN). The basic learning vector quantization network is applied and its learning ratio is slightly regulated in order to meet the need of the actual practice. The input vectors of the LVQNN, which need not been normalized, are acquired directly from the real time images. During training the network, the samples that are cropped from the real time images according to the size of the measured objects are divided into two groups: positive and negative samples, in which the measured object is or not existing respectively. The results of simulation about the network showed that the basic learning vector quantization neural network is adequate for some object measurements because the network is easy to get, meanwhile the number of the training samples and the knowledge about the measured objects need not too much. Also experiments from the online application indicated that the selection and number about training samples, and the process to train the network are relatively arbitrary because an excellent performance could be achieved though only very small sets of data and samples are used.