Study On Pollen Classification And Identification Based On Fluorescence And Raman Spectroscopy

Pollen in the air is potentially harmful to people’s daily activities due to its high sensitization.The types and reactions of pollen allergy vary from person to person,so it is a practical research work to identify and classify pollen.Infrared and Raman spectroscopy detection techniques are not widely used because of their expensive equipment and long testing time.Based on the intrinsic fluorescence principle,this paper designed and developed a fluorescence detection system excited by 375 nm laser light source,which has the advantages of low cost,high sensitivity and real-time detection.Moreover,the deep learning network was constructed to identify and classification the pollen fluorescence data collected by the system.The results of this study have significant reference significance for the detection of pollen in air by fluorescence spectroscopy.According to the comparison of fluorescence intensity emitted by four pollen species at different excitation wavelengths,the optimal excitation wavelength range of the designed fluorescence detection system is 370-380 nm,and the excitation light source of 375 nm is determined.A highly sensitive fluorescence detection system using sin gle photon avalanche diode array(SPAD array)as detection module was developed.The sensitivity of the system was up to 1mg/m L pollen by using orthogonal optical path and sampling through capillary and peristaltic pump.Using this system,43 pollen specie s were measured by fluorescence spectrum,and pollen components were analyzed by fluorescence spectrum and intensity.The fluorescence spectral data of 43 pollen species were preprocessed by multiple scattering correction method,and then principal compone nt analysis method was used to obtain that the fluorescence intensity of the 5 bands at 480,490,470,450 and 460 nm had the highest contribution rate to pollen fluorescence spectrum.Then,the fluorescence spectral data of 43 pollen species were divided into training set and test set,and the results showed that:The support vector machine method achieves the best performance when gaussian kernel is used,and its classification accuracy is 87.85%.The convolutional neural network(CNN)and full convolution al network(FCN)are better than that of SUPPORT vector machine,with the highest accuracy reaching 96.01% and 97.19% respectively.In addition,the minimum loss function of FCN is smaller.Moreover,the same number of iterative training takes less time,so the constructed FCN is more suitable for the classification of fluorescence spectral data in this study.At the same time,the Raman spectra of these 43 pollens were collected,and then the Raman spectra of pollens were successively smootened by the least square polynomial,the piecewise linear fitting method to correct the baseline,and the maximum normalization method to normalize the original signal intensity.The data input format is converted from 1 × 1600 to 40 ×40.The results show that the final accuracy rate of CNN and FCN classification models is 96%,which effectively demonstrates the application of the classification model constructed in this paper.