高光谱技术的水上苯乙烯厚度反演模型研究

Study on inversion model of water styrene thickness based on hyperspectral technology

针对水上发生化学品泄露时传统监测手段不能及时准确地获取泄露污染物的厚度问题,提出了以高光谱数据为基础数据,提取出与污染物厚度相关性较好的特征变量作为预测变量,结合python中的机器学习,通过4种预测模型进行污染物厚度反演。本文以水上泄露常见的化学品苯乙烯为例,测定不同厚度的水上苯乙烯及其对应的高光谱数据集,通过相关系数以及p值极值提取法,筛选出的11个特征变量,分别建立了多元线性回归(MLR)、偏最小二乘(PLSR)、支持向量机(SVM)和随机森林(RF)4种苯乙烯厚度反演模型。结果显示,4种反演模型都可以达到相对较好的反演效果,其中随机森林模型反演效果相对较好,其相关系数(R^2)为0.938 6,均方根误差(RMSE)为20.94,完全可以用于水上苯乙烯厚度反演。

In view of the fact that the thickness of the leaking pollutant cannot be obtained in time and accurately by the traditional monitoring means in the case of chemical leakage on the water,the paper puts forward the data based on the high-spectral data,extracts the characteristic variables with better correlation with the thickness of the pollutant situas as the predictor variable,and combines the machine learning in python.The contaminant thickness inversion was carried out by 4 prediction models.In this paper,taking the common chemical styrene of water leakage as an example,the water styrene of different thicknesses and its corresponding hyperspectral data set(350-2 500 nm) were measured,and the 11 inverse lysions were finally screened by correlation coefficients and p-value extreme extraction methods,respectively,and a multi-linear regression(MLR) was established,respectively.PLSR,support vector machine(SVM) and random forest(RF) four styrene thickness inversion models.The results show that the four inversion models can achieve relatively good inversion results.However,the random forest model has a relatively good inversion effect,with a correlation coefficient(R^2) of 0.938 6 and a root mean square error(RMSE) of 20.94,which can be used for Inversion of styrene thickness on the water.