Research On Feature Extraction And Classification Of Organic Contaminants In Drinking Water Using Three-Dimensional Fluorescence Spectra

The drinking water security of urban water distribution system is closely related to residents' health and social stability.The water pollution detection and early warning system are important compositions to ensure the security of urban water supply.Three-dimensional(3D)fluorescence spectroscopy technology has attracted much attention in the field of drinking water quality detection due to its high sensitivity and high selectivity.However,in the process of detecting the organic pollutants in drinking water,several factors,such as environmental fluctuations in the water,changes in the types or the concentrations of the pollutants,will effect on the detection results.In this thesis,we propose a 3D fluorescence spectral feature extraction and qualitative discrimination method for the detection of organic pollutants in urban water distribution system.The algorithm is verified by simulating pollutant experiment.The main work and innovations of the thesis are as follows.(1)Aiming at the problem that the concentration of pollutants in the process of organic detection is generally low,the effective signal of 3D fluorescence spectrum is weak,the data dimension is high,and it is susceptible to interference,the method on morphological gray reconstruction combined with alternating trilinear decomposition(ATLD)is proposed for the feature extraction of the 3D fluorescence spectra,which is sufficient to extract the spectral characteristic information of the pollutant sample,and the threshold method is used for qualitative discrimination of the organic pollutant.Firstly,the 3D fluorescence spectrum of the normal drinking water samples was extracted by ATLD.The sample to be tested is firstly subjected to morphological grayscale reconstruction for fluorescence peak localization,and the data of the characteristic spectral region was then weighted and amplified.Then the samples were used to establish a normal drinking water trilinear model,and the sum of squared residuals of the model and the background water quality model is calculated.Finally,the sample to be tested was qualitatively determined by the threshold method.Through several water pollutants experiments,the detection ability of the algorithm was verified and discussed.(2)Aiming at the influence of background water fluctuation on the detection effect in long-term monitoring,the optimization method of water organic pollutants anomaly detection based on sliding window and dynamic threshold is studied.This method realizes the water quality by continuously updating the decomposition matrix and residual space of the background water model to achieve adaptive optimization of anomaly detection models under the trends of changing spectrum.Firstly,the causes of fluctuations in water quality background and the limits of detection methods under the online monitoring scene of drinking water were analyzed through the three-dimensional fluorescence spectrum data of normal water quality collected over continuous days.Then the feature extraction of the three-dimensional fluorescence spectrum of normal drinking water samples was established by trilinear decomposition method to establish the initial model.Morphological grayscale reconstruction feature extraction and signal amplification were performed on the samples to be tested and a normal drinking water trilinear model was established.Finally,the updating of the trilinear model based on the sliding window ATLD and dynamic threshold was proposed.The verification analysis was carried out by experiments and the optimized method was compared with the fixed ATLD model method.(3)In the process of organic pollutant detection of drinking water quality,the spectral features of similar organic pollutants have high similarity and feature peaks overlap.The existing spectral feature extraction method has high information loss rate and low recognition accuracy.To handle the aforesaid problems,the method for identifying organic pollutants in drinking water based on two-dimensional Gabor wavelet feature extraction and SVM is studied.Two-dimensional Gabor wavelet is proposed,which is also combined with support vector machine(SVM)multi-classifier to identify characteristic pollutants in drinking water.The experimental results show that the two-dimensional Gabor wavelet combined with the block statistic can extract the characteristics of the three-dimensional fluorescence spectrum more effectively and has higher identification accuracy of the characteristic pollutants in drinking water,especially for substances whose characteristic peak overlaps with each other,compared with PCA.The thesis is aimed at the various problems existing in the process of detecting organic pollutants in drinking water security of urban water distribution system.It studies,optimizes and improves the detection and identification methods based on 3D fluorescence spectroscopy.The proposed method can effectively extract the characteristics of 3D fluorescence spectrum data,improve the abnormal detection rate and recognition accuracy of organic pollutant detection,and can be used in online water quality emergency detection system,which has potential application value for ensuring the security of urban water distribution system.