Research On Classification And Recognition Of Water Quality Spectra Based On Machine Learning

The safety of water quality is related to the health of people.The reliable and convenient water quality detection technology can guarantee the safety of drinking water for residents.The traditional water quality detection technology has complex operation and low efficiency,while machine learning has excellent performance in the field of classification and recognition.Therefore,the water quality spectral detection technology based on machine learning has the characteristics of fast speed,easy operation and high sensitivity.In this paper,three kinds of pharmaceuticals and personal care products(PPCPs)water pollutants including etamsylate,promethazine hydrochloride and chlorpromazine hydrochloride were selected as the detection objects,and the sample solutions with different concentrations were configured for fluorescence spectra collection.This paper has studied in depth the classification and recognition of water quality spectra based on machine learning,focusing on the improvement of model classification and recognition capabilities by spectral preprocessing technology,and the impact of different machine learning algorithms on the performance of spectral classification.In order to study the effect of spectral preprocessing on the classification performance of the algorithm,Savitzky-Golay(SG)convolution smoothing method and median filtering method were selected to denoise the collected 490 fluorescence spectra respectively.For each of the three water quality pollutants,the original spectra,the SG smoothed spectra and the median filtered spectra of three concentrations were selected for preliminary analysis.Among them,the signal-to-noise ratio of the median filtering method is better than that of the SG convolution smoothing method.In order to study the influence of different machine learning algorithms on the performance of spectral classification,in this paper,support vector machine(SVM),random forest(RF)and extreme gradient boosting(XGBoost)algorithms were selected for research.The original spectra,the spectra smoothed by SG and the spectra filtered by median were used for training and testing,and the grid search algorithm was used to optimize the model.The experimental results show that the machine learning technology has a high performance for the classification and recognition of spectra of water quality pollutants,and the noise reduction pretreatment can improve the accuracy of the algorithm classification to a certain extent,and the median filtering method has a better performance.The SVM models have the highest performance with 100 %accuracy.The experimental speed of the SVM models is also fast,which can reach0.18 s.The RF models can be faster than the SVM models,but not as accurate as the SVM models and the XGBoost models.Based on the experimental results,this paper builds the application software for the detection of water pollutants through Python.This paper constructs water quality spectral identification models based on machine learning algorithms to realize the rapid detection of water pollutants,which has a certain value for water quality detection and has a certain significance for ensuring the safety of water quality.