A Study On Spectral Features And Real-time Detection Of Blood Spots In Eggs

China is a major egg producer as well as a major egg consumer, the detection and grade of egg quality has capacious market future. The absence or presence of internal optically active anomalies blood spots in eggs is an important index to egg quality. This research focused on the detection of the blood spots in chicken eggs by combining spectroscopy technology and chemometrics. After designing the internal blood spots detection device, three analytical methods were used to detect the internal optically active anomalies blood spots, including the traditional discrimination method by detecting the blood value, partial least squares discriminant analysis (PLSDA) and least squares support vector machine (LS-SVM) combining with spectral information and eggshell color information. A real-time discriminant model was trained after online test and a model update method was proved to be useful to improve correct probability of detecting.Main research content, results and conclusions are summarized as follows:(1) After compared the difference between the average spectra of the normal eggs and blood-spot eggs, three characteristic wavelength including two absorption peaks of blood in the eggs at544.45nm,575.58nm and a reference wavelength at633.45nm were found. While the blood values discriminant method couldn’t discriminate the blood-spot eggs from the normal eggs effectively, this results from the influence of surface eggshell color.(2) Four spectral pretreatment methods, first derivative, second derivative, multiplicative scatter correction(MSC), standard normal variate(SNV), were compared to detect the blood spots in eggs using partial least squares discriminant analysis (PLS-DA). The result showed that the multiplicative scatter correction pretreatment method got the highest discriminant rate of81.2%.(3)The discriminant method of partial least squares support vector machine(LS-SVM) method with integration of spectral data and eggshell color information was explored to detect the blood spots in eggs and results showed that due to the use of spectral information within a certain wavelength range, more spectral information was involved than the blood value discrimination method, the discrimination result was better than the blood value discrimination. The overall discriminant accuracy rate was up to88.6%.(4) An online non-destructive inspect system to detect the blood spots was established after the design of optical system, mechanical system, and a real-time classification software, which included the spectral data collection, data storage, data loading and data display modules. (5) The offline model, constructed by the Logistic regression method, was applied for online real-time detection. The results demonstrated that the model update method by adding some similar samples to training set made the discrimination rate reach87.5%.