Rock and Soil Classification Using PLS-DA and SVM Combined with a Laser-Induced Breakdown Spectroscopy Library

Laser-induced breakdown spectroscopy （LIBS） has become a powerful technology in geological applications. The correct identification of rocks and soils is critical to many geological projects. In this study, LIBS database software with a user-friendly and intuitive interface is developed based on Windows, consisting of a database module and a sample identification module. The database module includes a basic database containing LIBS persistent lines for elements and a dedicated geological database containing LIBS emission lines for several rock and soil reference standards. The module allows easy use of the data. A sample identification module based on partial least squares discriminant analysis （PLS-DA） or support vector machine （SVM） algorithms enables users to classify groups of unknown spectra. The developed system was used to classify rock and soil data sets in a dedicated database and the results demonstrate that the system is capable of fast and accurate classification of rocks and soils, and is thus useful for the detection of geological materials.

The classification of plants by laser-induced breakdown spectroscopy based on two chemometric methods

The applications of laser-induced breakdown spectroscopy(LIBS) on classifying complex natural organics are relatively limited and their accuracy still requires improvement.In this work,to study the methods on classification of complex organics,three kinds of fresh leaves were measured by LIBS.100 spectra from 100 samples of each kind of leaves were measured and then they were divided into a training set and a test set in a ratio of 7:3.Two algorithms of chemometric methods including the partial least squares discriminant analysis(PLS-DA) and principal component analysis Mahalanobis distance(PCA-MD) were used to identify these leaves.By using 23 lines from 16 elements or molecules as input data,these two methods can both classify these three kinds of leaves successfully.The classification accuracies of training sets are both up to 100% by PCA-MD and PLS-DA.The classification accuracies of the test set are 93.3% by PCA-MD and 97.8% by PLS-DA.It means that PLS-DA is better than PCA-MD in classifying plant leaves.Because the components in PLS-DA process are more suitable for classification than those in PCA-MD process.We think that this work can provide a reference for plant traceability using LIBS.

Comparative study on identi¯cation of healthy and osteoarthritic articular cartilages by fourier transform infrared imaging and chemometrics methods

Two discriminant methods,partial least squares-discriminant analysis(PLS-DA)and Fisher's discriminant analysis(FDA),were combined with Fourier transform infrared imaging(FTIRI)to differentiate healthy and osteoarthritic articular cartilage in a canine model.Osteoarthritic cartilage had been developed for up to two years after the anterior cruciate ligament(ACL)transection in one knee.Cartilage specimens were sectioned into 10μm thickness for FTIRI.A PLS-DA model was developed after spectral pre-processing.All IR spectra extracted from FTIR images were calculated by PLS-DA with the discriminant accuracy of 90%.Prior to FDA,principal component analysis(PCA)was performed to decompose the IR spectral matrix into informative princi pal component matrices.Based on the different discriminant mechanism,the discriminant accuracy(96%)of PCA-FDA with high convenience was higher than that of PLS-DA.No healthy cartilage sample was mis assigned by these two methods.The above mentioned suggested that both integrated technologies of FTIRI-PLS-DA and,especially,FTIRI-PCA-FDA could become a promising tool for the discrimination of healthy and osteoarthritic cartilage specimen as well as the diagnosis of cartilage lesion at microscopic level.The results of the study would be helpful for better understanding the pathology of osteoarthritics.

Classification and comparison of physical and chemical properties of corn stalk from three regions in China

Corn stalk samples from Anhui,Jiangxi and Shanghai were used as test materials.Their physical,chemical and thermo-chemical engineering characteristics were analyzed.The similarities and differences in properties of corn stalk from the three regions were determined using SIMCA-P and SPSS software in order to obtain a proper energy utilization method of corn stalk.The results show that the corn stalk from Shanghai has significant differences from the samples of Jiangxi and Anhui.In particular,the following properties of corn stalk from Shanghai such as the contents of cellulose,calcium(Ca),iron(Fe),crude ash,volatile matter,carbon(C),nitrogen(N),and oxygen(O)are significantly different from those of Jiangxi and Anhui samples(P<0.05).While other properties such as the contents of magnesium(Mg),copper(Cu),zinc(Zn),moisture,hydrogen(H),and sulfur(S)have no significant difference among samples of three regions.Compared with the corn stalk in Anhui and Jiangxi,the Shanghai samples are more suitable for the production of ethanol because of their higher ratio of cellulose to hemi-cellulose content.Because of its high content of ash and low calorific value,the Shanghai corn stalk is suitable for the gasification process instead of for direct combustion or bio-oil production.The research can provide a reference for raw material selection for biomass energy production and utilization.

Detection of explosives with laser-induced breakdown spectroscopy

Our recent work on the detection of explosives by laser-induced breakdown spectroscopy （LIBS） is reviewed in this paper. We have studied the physical mechanism of laser-induced plasma of an organic explosive, TNT. The LIBS spectra of TNT under single-photon excitation are simulated using MATLAB. The variations of the atomic emission lines intensities of carbon, hydrogen, oxygen, and nitrogen versus the plasma temperature are simulated too. We also investigate the time-resolved LIBS spectra of a common inorganic explosive, black powder, in two kinds of surrounding atmospheres, air and argon, and find that the maximum value of the O atomic emission line SBR of black powder occurs at a gate delay of 596 ns. Another focus of our work is on using chemometic methods such as principle component analysis （PCA） and partial least squares discriminant analysis （PLS-DA） to distinguish the organic explosives from organic materials such as plastics. A PLS-DA model for classification is built. TNT and seven types of plastics are chosen as samples to test the model. The experimental results demonstrate that LIBS coupled with the chemometric techniques has the capacity to discriminate organic explosive from plastics.

On Splitting Training and Validation Set:A Comparative Study of Cross-Validation,Bootstrap and Systematic Sampling for Estimating the Generalization Performance of Supervised Learning

Model validation is the most important part of building a supervised model.For building a model with good generalization performance one must have a sensible data splitting strategy,and this is crucial for model validation.In this study,we con-ducted a comparative study on various reported data splitting methods.The MixSim model was employed to generate nine simulated datasets with different probabilities of mis-classification and variable sample sizes.Then partial least squares for discriminant analysis and support vector machines for classification were applied to these datasets.Data splitting methods tested included variants of cross-validation,bootstrapping,bootstrapped Latin partition,Kennard-Stone algorithm(K-S)and sample set partitioning based on joint X-Y distances algorithm(SPXY).These methods were employed to split the data into training and validation sets.The estimated generalization performances from the validation sets were then compared with the ones obtained from the blind test sets which were generated from the same distribution but were unseen by the train-ing/validation procedure used in model construction.The results showed that the size of the data is the deciding factor for the qualities of the generalization performance estimated from the validation set.We found that there was a significant gap between the performance estimated from the validation set and the one from the test set for the all the data splitting methods employed on small datasets.Such disparity decreased when more samples were available for training/validation,and this is because the models were then moving towards approximations of the central limit theory for the simulated datasets used.We also found that having too many or too few samples in the training set had a negative effect on the estimated model performance,suggesting that it is necessary to have a good balance between the sizes of training set and validation set to have a reliable estimation of model performance.We also found that systematic sampling method such as K-S and SPXY generally had very poor estimation of the model performance,most likely due to the fact that they are designed to take the most representative samples first and thus left a rather poorly representative sample set for model performance estimation.