首页|Heavy metals detection based on deep learning and laser induced breakdown spectroscopy (LIBS)
Heavy metals detection based on deep learning and laser induced breakdown spectroscopy (LIBS)
扫码查看
点击上方二维码区域，可以放大扫码查看
原文链接
NETL
Many pollutants are sprayed in air,water,and soil every day due to the rapid development of industries,mines,transport.Heavy metals are one of these pollutants that can cause damage to animals and plants.They are present in many facets of life and are sometimes essential for human life(iron,cobalt,and zinc).However,some such as chromium,cadmium,mercury,and lead are dangerous for the health.Among those quoted,lead is the most dominant heavy metal contaminant you are probable to meet.There are plenty of techniques to reduce heavy metals in our planet.However,it is quite hard to remove something without detection.Laser-Induced breakdown spectroscopy(LIBS)is known as a tool of detection of several chemical elements.Despite its effectiveness,it is associated with chemometric which make it powerful for quantitative and qualitative analysis.The main objectives of this study were to classify the soil contaminated with a different level of lead(Pb).In order to make the experience as real-life experience,the sample is composed of soil in which tobacco was transplanted.Tobacco is a sensitive plant and can uptake contaminant.The classification using different algorithms such assupport vector machine(SVM),partial least squares discriminant analysis(PLS-DA)and deep learning(deep belief network)was computed.Furthermore,a deep study was also devoted to the tobacco leaves,as we know the tobacco is used to make a cigarette and the leaves are the essential components.It has been demonstrated that there is a rapid way of Pb detection using SVM and PLS. In order to improve our previous algorithms,LIBS has been associated with a new algorithm namely Multilayer-Deep Belief Network(MN-DBN)to see the contents of the tobacco absorption precisely the content of Pb in leaves stems and roots of tobacco.This algorithm has been compared using cross-validation technique to Partial least square(PLS)and support vector machine(SVM).The obtained main results in this study were(1)deep learning(deep belief network)was powerful for classifying soil samples with different level,the accuracy of the cross-validation reaches96.17%for four weeks contamination and84.11%for two weeks contamination,SVM obtained95.90%and82.64%(4weeks and2weeks).The two PLS-DA(LVs=5and LVs=36)received an accuracy of79.26%-90.13%(2weeks and4weeks)and78.63%-90.24%(2weeks and4weeks).(2)PLS and SVM are both efficient and could be used forthe rapid detection of Pb in tobacco leaves,but the comparison of the different mean square error(MSE)gives a better yield to PLS than SVM.(3)Despite their effectiveness,PLS and SVM give a lower result compared to the new algorithm.Indeed,the comparison of mean square error and the coefficient of determination showed MN-DBN more efficient than the two previous.
Mahamed Lamine Guindo
展开 >
Soil contaminated LIBS machine learning deep learning chemometrics
NETL
