Research On Infrared Spectrum Recognition Method Of Hazardous Gas Based On Boosting Algorithm

With the rapid expansion of our country's industrial scale,hazardous gas leakage accidents caused by industrial production,transportation and other links occur more and more frequently,causing huge losses to the safety of people's lives and property.The use of infrared spectroscopy analysis technology to analyze the leakage of hazardous gas components can provide decision-making basis for fire decontamination at the accident site and reduce the losses caused by the accident.The qualitative analysis of the collected spectral data quickly is the key link in the infrared spectral analysis technology.Traditional spectral recognition methods,such as spectral library search,have extremely high requirements for spectral signal-to-noise ratio,making it difficult to use widely.With the rapid development of computer science and technology,especially machine learning algorithms have shined in various classification and recognition fields,the combination of machine learning algorithms and infrared spectrum recognition has become one of the research hot spots in the field of infrared spectrum analysis.Among the many machine learning algorithms,the integrated learning algorithm has attracted much attention because of its powerful learning ability and stability,but it is rarely used in the field of spectrum recognition.In this paper,based on the Boosting algorithm in the integrated learning algorithm,the infrared spectrum identification method of hazardous gases is researched.The main content of this article includes the following aspects:(1)In view of the phenomena of baseline drift,noise interference and excessive differences in spectral dimensions in the infrared spectrum data of the five measured hazardous gases of acetone,p-xylene,chloroform,trichloroethane,and tetrachloroethylene,use "Polynomial fitting spectrum automatic baseline correction+Savitzky-Golay smoothing filter+ Normalization" method preprocesses the original data.By analyzing the characteristics of the selected hazardous gas infrared spectrum data,the parameters such as full width at half maximum,window correlation coefficient,kurtosis,skewness,signal-to-noise ratio,maximum position and other parameters are selected to construct the spectral feature vector for model recognition training.(2)Using the Adaboost.M2 multi-classification algorithm,respectively using the BP neural network and the CART classification tree model as the base learner,two spectral recognition models,Adaboost-BP and Adaboost-CART,are established.Six pre-processed infrared spectral data and spectral feature vectors are used to train the model and verify its performance.The experimental results show that the recognition accuracy of the two spectral recognition models of Adaboost-BP and Adaboost-CART has reached more than 96%,and the correlations of the recognition accuracy between the training set and the test set are 96.17%and 98.39%,respectively.The model is generalized strong ability.(3)Using the latest research results in Boosting algorithm-XGBoost algorithm,using the CART regression tree model as the base learner,an infrared spectrum recognition model is built based on the XGBoost algorithm,and the model is trained and performance verified using spectral feature vectors.The experimental results show that the recognition accuracy of the infrared spectrum recognition model based on the XGBoost algorithm reaches 96.87%,and the correlation between the training set and the test set reaches 98.53%,indicating that the XGBoost algorithm has good application prospects in the field of infrared spectrum recognition.