| Detection and identification of food-borne pathogens is an important part of food safety incidents prevention and a popular research topic in preventive medicine.Traditional detection methods encount several problems such as high cost,long measurement time,and complex sample preprocessing.Therefore,it is necessary to develop a novel,efficient and general food-borne pathogen detection and discrimination technique.In this thesis,spectroscopic techniques,such as laser-induced breakdown spectroscopy(LIBS)and single-cell Raman spectroscopy,were developed to detect pathogens like Salmonella enteritidis,Escherichia coli and Staphylococcus aureus.A classification and discriminant model were established on the fused spectra,and its feasibility for the discrimination of food-borne pathogens was demonstrated.The thesis is divided into three parts:(1)A dark-field microscopy system and sample fixation method were proposed for LIBS analysis of microbial single colony,which allowed precise colony positioning and avoided matrix interference.A variety of trace elements were successfully detected in microbes from the LIBS spectra.After dimension reduction using kernel principal component analysis(KPCA),support vector machine(SVM)and SVM-Adaboost multi-classification models were established on the LIBS spectra to classify food-borne pathogens.The relevant parameters were optimized,and the prediction accuracy rate of both models were above 90%.The enhancement on LIBS spectral line intensities using nano-gold particles as well as its effects on the prediction results of the classification model were also studied.(2)A homemade confocal Raman spectrometer was used to acquire the Raman spectra of five food-borne pathogens at a single-cell level.After dimension reduction via KPCA,SVM and SVM-Adaboost multi-classification models were established,which successfully classified and identified pathogenic bacteria at a single-cell level.The prediction accuracy of the SVM-Adaboost model reached 96%,with a 5%improvement as compared to the SVM model.(3)Data fusion strategies were developed for classification of microorganisms.The strategies combined the elemental information generated from the LIBS spectra and molecular information generated from the Raman spectra.Classification models based on both variable-and feature-level fused spectra were established.The results showed that the accuracy of 10-fold cross-validation and prediction was higher than that of the model based either on the LIBS spectra or the Raman spectra alone.The satisfactory results demonstrated that the spectral data fusion technique is an effective way improving the stability and accuracy of the prediction model.In summary,this thesis focuses on the discrimination of pathogenic bacteria at single-bacteria level,and the single-colony LIBS spectroscopy and single-cell Raman spectroscopy were developed,respectively.Especially,a classification model based on spectral data fusion technique was proposed,which paves a promising way for the multispectral detection and discrimination of pathogens. |
