Development Of Novel Raman Imaging Methodologies For Safety Inspection Of Dairy Products

Dairy products represent one of the main food sources of human being,whose consuming amount is huge.Unfortunately,some criminals made fake milk powder for maximum economic benefits,which will greatly threat the consumer health.The detection of adulteration presented in milk powder becomes the research focus in dairy safety area,providing the last threshold to control the safety of dairy products.However,the relatively single testing item and time consuming sample preparation combined to limit applications of the conventional detection methods in dairy safety detection,and it is therefore very important to develop highly efficient detection methods for identification of adulterations presented in milk powder.Milk powder is an inhomogeneous solid system consisted of complex multiple components.The demand of analysis of milk powder is usually high,since the sources of dopants are uncertain and uneven as well.This requires a perfect combination of efficient microdomain scanning and large scale area screening for analysis of milk powder.Among the detection methods,Raman imaging represents one of few techniques that satisfy the high demand of analysis of milk powder,which possesses the unique advantages of high resolution and throughput.This paper aims to apply Raman imaging technique for rapid detection of adulterated milk powder.In Raman imaging analysis of milk powder,it relies on chemometrics methods to extract the dopant information presented in highly overlapped Raman imaging signals with strong fluorescence background.The unexpected volatility in Raman imaging requires sophisticated methods for a reliable estimation.In this regards,a novel strategy,named as data-driven Multiscale modeling(DDMM),is then proposed to extract the essential information of dopants in Raman imaging signals adaptively through constructing the optimal wavelet basises for data sets at hands.DDMM is capable of simultaneous estimation of presence of adulteration and its concentration level.As a result,a spectral diagnosis model is finally constructed for identification of adulterated milk powder,avoiding the leakage of important information through data fusion.This thesis also develops another nonlinear algorithm based on least square support vector regression,named as modified discrete wavelet transform – elimination of uninformative variables – least square support vector regression(MDWT-UVE-LSSVR)for suppression of fluorescence background,matrix effect and nonlinear interferences introduced in the Raman imaging signals,respectively.MDWT-UVE-LSSVR was utilized for identification of adulterated milk powder,and the calibration results were encouraging.The results illustrated that MDWT-UVE-LSSVR model was capable of discrimination of different dopants in milk powder.As a result,the total accuracy rate is 99.74%,and the semi-quantitative analysis of adulteration in milk powder was performed simultaneously.This thesis aims to apply Raman imaging for rapid detection of adulterated milk powder.Two novel algorithms were developed to extract the essential characteristics presented in the Raman imaging signals.The calculation results reveal that a reasonable combination of algorithms is capable of coding with different interference presented in Raman imaging signals of complex systems.The two algorithms developed here is supposed to encode the unexpected volatilities successfully without any statistical hypothesis.This would greatly suppress the unexpected interference presented in Raman imaging signals to enhance the stability and accuracy of Raman imaging analysis,thus providing a promising tool for rapid discrimination of adulterated milk powder.