The quality and purity of industrial packaged foods is today of fundamental importance, given the level of expectation of consumers and the current laws imposing serious liabilitis on produces. This paper presents a novel method for automatic recognition of pollutants in packaged foods fo industrial applications. To maximize the contrast between foods and pollutants a dual acquisition method has been applied to obtain a pair of images taken at two differen x-ray source voltages. Taking advantage from the wavelength dependence of absorption coefficient for different materials. In order to further increase the classification potential of the algorithms, the HΣ color spectrum was adopted, for its high dicrimination capabilities. The analysis of images is performed on-line utilizing three independent methods. Over a series of experiments each of the three strategies have given a correct classification rate of pollutants ranging from 83% to 95%. To further increase the degree of reliability of the automatic recognition process, the three methods have been combined into a pollution coefficient. The confidence achieved on the experimental set resulted in a 92% correct classifications, for pollutants larger than 2mm.
Automatic Recognition of Pollutants in Packaged Foods from x-ray Imaging
GRASSO, Giorgio Mario;
2005-01-01
Abstract
The quality and purity of industrial packaged foods is today of fundamental importance, given the level of expectation of consumers and the current laws imposing serious liabilitis on produces. This paper presents a novel method for automatic recognition of pollutants in packaged foods fo industrial applications. To maximize the contrast between foods and pollutants a dual acquisition method has been applied to obtain a pair of images taken at two differen x-ray source voltages. Taking advantage from the wavelength dependence of absorption coefficient for different materials. In order to further increase the classification potential of the algorithms, the HΣ color spectrum was adopted, for its high dicrimination capabilities. The analysis of images is performed on-line utilizing three independent methods. Over a series of experiments each of the three strategies have given a correct classification rate of pollutants ranging from 83% to 95%. To further increase the degree of reliability of the automatic recognition process, the three methods have been combined into a pollution coefficient. The confidence achieved on the experimental set resulted in a 92% correct classifications, for pollutants larger than 2mm.Pubblicazioni consigliate
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.