Nondestructive Detection Method of Egg Quality Based on Multi-Sensor Information Fusion Technology

For the purpose of enhancing the detecting stability and the model adaptability of egg freshness by nondestructive detection method, a sensor fusion was taken by the machine vision and near infrared spectroscopy in the sensor level of characteristics while D-S evidence theory was chosen as the sensor information fusion method and BP neural network as the specific modeling method. First we extract eggs data by machine vision technology and near-infrared spectroscopy. The principal component analysis method was the best method in the spectral feature extraction through the spectroscopy preprocessing. This study compared several different spectral preprocessing methods to preprocess the near infrared spectroscopy in near infrared spectroscopy preprocessing. And from the discriminating rate of BP neural network machine model with the pretreated near infrared spectroscopy, it was concluded that the dimension reduction method was the best spectroscopy preprocessing method. So this study used BP neural network to detect egg freshness based on machine vision technology and near infrared spectroscopy analysis technology. The BP neural network models to discriminate egg freshness were built based on machine vision technology and near infrared spectroscopy analysis technology respectively. The maximum residual errors of haff values (eggs freshness) using these two models were 12.08 and 16.50 in testing sets respectively. To improve the discriminating egg detecting with machine vision and near infrared spectroscopy respectively, a multi-sensor information fusion technique based on near infrared spectroscopy and machine vision technology was used to detect eggs. D-S evidence theory was a good information fusion method, so evidence theory combined with BP neural network was built with image characteristics and spectral characteristics. An improved method was discussed that could remedy for the deficiency of previous two models. Verification results showed that the basic probability assignment of uncertainty was less than 0.01 by sensor fusion optimization. The problem of low detecting range in single sensor method has been well solved. Meanwhile, the accuracy of freshness discriminating has been improved by sensor fusion situation. The discriminating accuracy reached to 90%.