Detection Of Heavy Metal Stress In Camellia Sinensis And Physiological Indices Changes Based On Spectral Technology

In this study,Camellia sinensis was used as the research object,and the distribution of heavy metal content in plants under lead stress and the effects of different factors on the absorption of heavy metals in plants were investigated.The discrimination analysis of lead pollution in tea was realized by Vis/Near infrared spectroscopy.The changing trends of chlorophyll,ascorbic acid,glutathione and soluble protein in tea leaves under heavy metal stress were explored,and the rapid quantitative detection of these physiological and biochemical indexes was realized by hyperspectral technique.The main research contents are as follows:?1?The distribution of heavy metal content in plants under different stress patterns?leaf lead stress and root lead stress?and the effects of different factors on plant absorption of heavy metals were investigated.It was found that under the leaf lead stress and root lead stress,the heavy metal content in the plant increased significantly with the prolongation of stress time,and there were significant differences in various tissues and organs.The distribution pattern was related to the stress pattern.Under leaf lead stress,the Pb content in each organ was:stem>leaf>root;under the lead stress,the Pb content in each organ was:root>stem>leaf.The distribution of various organs under leaf lead stress is slightly different from that of previous studies on the absorption of heavy metals in the atmosphere,probably because of the difference between foliar spray and actual atmospheric deposition.The content of heavy metals in plants under high concentration stress was also higher than that under low concentration stress,but no significant difference was found.The content of heavy metals in different varieties of Camellia sinensis is also similar.?2?Using Vis/Near infrared spectroscopy combined with chemometrics method,the discrimination analysis of lead pollution in tea was realized,which laid a foundation for realizing rapid detection of heavy metals in the field.In this study,four linear classifiers?LR,SR,Liner SVM and LDA?and two nonlinear classifiers?RF and CNN?were used to establish discriminant models for full spectrum,characteristic wavelengths and vegetation index.The best algorithm based on the full spectrum discriminant analysis model is CNN.The accuracy rate of the leaf lead stress CNN model is 98.9%,and the prediction accuracy is 77.8%.The modeling accuracy of the root stress CNN model is 99.7%,the prediction accuracy rate is 73.3%.At the same time,known from the confusion matrix,the stress samples can be successfully identified from healthy samples,and the false positive rate is only 15%and 5%.However,healthy samples were not well recognized,and the false positive rates of the two stressed health samples were 37%and 30%,respectively.Comparing the classification effects of two supervised learning algorithms?CARS and SPA?and two unsupervised learning algorithms?PCA and AE?,it is found that the optimal discriminant models under leaf lead stress and root lead stress are based on The CNN model of the characteristic wavelength extracted by SPA has a prediction accuracy of 73.0%and 76.4%,respectively.The feature wavelength selection algorithm selects the characteristics of single digits,which greatly reduces the sample dimension and maintains the modeling effect equivalent to full spectrum modeling,indicating that the feature band extraction algorithm can reduce the data dimension while retaining the effective information of the data..Among the five different vegetation indices,the correlation coefficient between(1₁ and the classification label was the largest,and the vegetation index with the largest correlation coefficient was selected for classification modeling.The modeling and prediction accuracy rate reached more than 63%.?3?The trends of chlorophyll?Chl?,ascorbic acid?ASA?,glutathione?GSH?and soluble protein?SP?in tea leaves under heavy metal stress were investigated.Studies have shown that both leaf lead stress and root lead stress can cause changes in the contents of Chl,ASA,GSH and SP in tea samples,and the changes in sample content under high concentration stress tend to be earlier than low concentration stress.The change of Chl content mainly showed the early stability and the slight decrease in the content at the late stage of stress.The overall trend of ASA content was rising,mostly on the 30th day after stress,but the content was slightly higher than the control group.The change in the content of GSH is earlier,and there is a certain increase in the 20th day after the stress,but then there will be a certain degree of reduction.In the stress,the index of SP is a relatively high index,which usually has a significant increase from the 20th to the 30th,and then the content is stable at a level higher than the control group.?4?Rapid quantitative detection of Chl,ASA,GSH and SP under heavy metal stress was achieved using hyperspectral technique combined with chemometrics.In this study,the average spectrum of tea leaves under heavy metal stress was extracted,and linear?MLR?and nonlinear?LS-SVM?models were established based on the characteristic wavelengths selected under optimal pretreatment.The original spectra were pretreated by four pretreatment methods?DT,BC,SNV and MSC?.The optimal pretreatments of the Chl,ASA,GSH and SP models of leaf lead stress were DT,MSC,ORI and ORI,respectively.The optimal pretreatments for the Chl,ASA,GSH and SP models of root lead stress were BC,ORI,MSC and DT,respectively.The characteristic wavelengths of the optimal pre-processed spectra were extracted by CARS,SPA and CARS-SPA.By comparison,it was found that except for SP model of lead leaf stress,GSH and SP model of root lead stress,the prediction effect of the model constructed by CARS-SPA is slightly worse than CARS algorithm,the results of CARS-SPA are all better than other feature extraction methods.It shows that the CARS-SPA feature extraction algorithm can improve the performance of the model while reducing the number of feature bands,and the effect of the model is guaranteed.In order to better explore the correlation between chemical indicators and characteristic wavelengths,improve the robustness and predictive ability of the model,the modeling effects of MLR and LS-SVM algorithms are compared.Comprehensively consideration,the model established by MLR is simple and efficient,and has good prediction ability.The predictive determinant coefficient of the model can reach about 0.6⁰.9,which provides a theoretical basis for the subsequent development of online monitoring equipment.