A method and system for rapid estimation of soil vanadium content using mid-infrared FTIR spectral data analysis and a multi-model machine learning framework.

IN605087BActive Publication Date: 2026-10-06INDIAN INST OF TECH INDIAN SCHOOL OF MINES DHANBAD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
IN202531078724
Authority / Receiving Office
IN · IN
Patent Type
Patents
Current Assignee / Owner
Filing Date
2025-08-19
Publication Date
2026-10-06

AI Technical Summary

Technical Problem

Existing methods for determining vanadium content in soil, such as LIBS and XRF, are time-consuming, costly, and lack accuracy in detecting trace amounts, while traditional IR spectroscopy methods do not effectively optimize vanadium predictions.

Method used

A method combining mid-infrared FTIR spectroscopy with machine learning models to identify specific absorption bands associated with vanadium complexes, using a multi-model regression approach to predict vanadium content in soil.

Benefits of technology

Provides a rapid, non-destructive, and cost-effective method for accurately estimating vanadium content in soil, suitable for real-time analysis using handheld or laboratory FTIR spectrometers, reducing operational costs and time.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

A method and system for rapid estimation of soil vanadium content using mid-infrared FTIR spectral data analysis and a multi-model machine learning framework. The present invention is a method to predict vanadium content in soil. Soil samples from agricultural field (204) and soil samples of unknown vanadium content (206) are collected, processed and prepared (208) for FTIR analysis (210) and x-ray fluorescence (XRF) analysis (212) as reference (VXRF, ppm) (214). Samples are processed and prepared at a known dilution factor (216), according to concentration increment for mid-infrared Fourier transform infrared spectroscopy analysis (218) and the resulting spectral signature for all samples (soil + KBr) between 1200 to 350 cm-1 are recorded (220). K-fold cross-validation of the soil samples (222) is used to split the sample into training (224) and test sets (226). A data analyzer (230) is used to conduct the analysis using multiple regression models to analyze the selected data. A multi-model estimation method is applied using the average of these best-performing models (230). The training set is used to predict the vanadium of the solid test pellets using FTIR data (234). The predicted and observed vanadium content in soil samples are compared (238) to prove that the observed and predicted vanadium of soil samples are statistically similar (240). The methodology yields vanadium content in soil samples (242). Fig. 5(a) & (b)
Need to check novelty before this filing date? Find Prior Art

Description

FIELD OF INVENTION:The present invention relates to a system and method to predict the soil vanadium (V) contentutilizing a multi-model regression framework and mid-infrared Fourier transform infraredspectroscopy (FTIR). Mid-infrared FTIR spectral examination of soil was performed to detectand measure thirty-nine specific absorption bands that are sensitive to the content of functionalgroups with respect to vanadium, within the range of 1200-350 cm-1. The area under the curve(AUC) derived for these spectral zones are used as parameters in estimating and forecastingvanadium content in soil through machine learning methods.BACKGROUND OF THE INVENTION AND PRIOR ART:Vanadium (V), one of the important elements with atomic number 23, is a silver-grey transitionmetal known for its strength, durability, and resistance to corrosion. Vanadium is distributedwidely in sedimentary and igneous rocks and is found all over Earth's crust, but is low inabundance (Pyrzyska, 2004; Yang et al., 2017). Vanadium in soils occurs within a range of 20-110 mg kg-1. The average abundance of vanadium is 135 ppm (Clarke's value) (Misra et al.,2020) (Reijonen et al., 2016). However, anthropogenic vanadium emissions have significantlyincreased over the past few decades due to industrial use and the burning of fossil fuels.Leachates from mine tailings, municipal waste, slags from the steel industry, and fertilizerscontaining vanadium get released, leading to contamination (Kabata-Pendias, 2004).Constantly redistributed in nature, vanadium hardly occurs in its natural phase. It naturallyoccurs in over 65 different mineral forms (mostly found in varnotite, patronite, roscoelite, andvandinite) and is the 22nd most abundant element, more common than copper and nickel(Amorim et al., 2007; Y.-M. Zhang et al., 2011). Studies have shown that vanadium is possiblycarcinogenic to human beings (ATSDR, 2012). Heavy metal pollution in soil, particularly ofvanadium, disrupts ecological equilibrium and is a cause of serious health concern in humanbeings.The vanadium content in soil is determined by different analytical techniques which commonlyincludes X-ray fluorescence spectrometry (XRFS), laser-induced breakdown spectroscopy(LIBS), atomic absorption spectrophotometry (AAS), inductively coupled plasma atomicemission spectroscopy (ICP-AES), inductively coupled plasma mass spectroscopy (ICP-MS),and atomic absorption spectrophotometry (AAS). Although these laboratory techniques offerprecise quantitative estimations with good detection limits, they necessitate time-consumingmeasurement procedures and costly equipment. To reduce operating expenses and time,efficient and rapid analytical techniques are required in mine facilities (Verma & Deb, 2012).Recently, the latest analytical methods, like laser-induced breakdown spectroscopy, have beenexplored to develop vanadium content models of soil. LIBS combined with different machinelearning models like PLSR, MLR, RF, GBDT, ANN, SVR, LS-SVR, and different sampleforms, including coal, petroleum coke, and soil samples, were used to predict the vanadiumcontent (Du et al., 2024; Lu et al., 2019; Turovets et al., 2024; W. Zhang et al., 2021). Thoughrapid LIBS suffers from the limitation that it cannot detect elements present in trace amounts,and hence estimating vanadium in soil using LIBS is challenging.IR spectroscopy is a key molecular analysis technique in which molecules absorb IR radiationand enter higher energy states, causing bond vibrations that produce absorbance peaks atvarious wavenumbers in the spectrum. These peaks reveal molecular details quickly, safely,and without much sample preparation, making IR spectroscopy a valuable tool for soil analysis.In predictive modeling, assessing the importance of predictors and evaluating the correlationsamong variables are crucial for achieving accurate predictions. However, these aspects werenotably absent in most of the reviewed literature. While advanced analytical techniques suchas LIBS and XRF (Najafabadipour et al., 2025) have shown potential, they stand alone and donot accurately optimize vanadium predictions, giving unsatisfactory results. Therefore, there isa need for a technique that can simultaneously optimize the analysis of vanadium and otherheavy metals present in soil using a single, efficient analytical method. Table 1 compares thestatistical values found in earlier studies using different methods / equipment / inputdata / algorithms to estimate the vanadium content of soil and different samples.Novelty: (Describe how your invention is novel / new as compared to the existing state of theart). Please see Annexure-B for generating Novelty Search Report.The proposed invention is a novel method, a vanadium estimating system using FTIRspectroscopy that is not available to the public and has not been published in any document.The present invention offers unique features that distinguish it from existing standard vanadiummeasuring methods in the market.The present invention has identified unique absorption bands in the FTIR spectra of soilsamples associated with vanadium-related functional groups and bonds, particularly vanadium-oxygen complexes. A regression model was then developed using machine learning techniquesto capture the relation of these unique absorption bands (area under the curve) in the FTIRspectra with the vanadium quantification in soil. It is a novel and first attempt to measure thevanadium content of soil utilizing FTIR spectroscopy, which is not present in any currentlyavailable solutions.The present invention represents a method to predict the vanadium content of soil using FTIRspectroscopy combined with machine learning models. This innovative prototype can befurther designed to be incorporated into a standalone device or integrated into handheld orlaboratory FTIR spectrometers, equipped with a computer or machine, enabling users toperform near real-time estimation of vanadium in the field or industrial units such as mines,ensuring prompt and precise data collection for extraction and uses.The present invention using FTIR spectroscopy addresses a significant market gap as it is arapid and non-destructive optical method. It is also easier to handle and is safe, unlike othertraditional analytic methods, which require long sample preparation, precise temperaturecontrol, calibration, and acid digestion, which may lead to contamination. Moreover, predictingthe vanadium of soil using traditional or advanced analytical methods like LIBS and XRF oftenrequires costly, time-consuming data. This highlights the need for a single, cost-efficientanalytical technique to predict vanadium content.OBJECTS OF THE INVENTION:Therefore, it is an object of the present invention to utilize easily obtainable mid-infrared FTIRspectroscopy combined with machine learning models to predict the vanadium content of soil,compared to the existing state-of-art machine learning models using laser induced breakdownspectroscopy (LIBS), atomic absorption spectrophotometry (AAS) and few others existingspectroscopy, which are far more complex.Another object of the present invention is to perform mid-infrared FTIR spectral examinationof soil to detect and measure thirty nine specific absorption bands (ranges and peaks) that aresensitive to the content of functional groups in the soil with respect to vanadium, within therange of 1200-350 cm-1and to develop a multi-model machine learning approach that relatesvanadium content of soil with the identified spectral parameters.A still another object of the invention is to determine a multi-model estimation approachthrough a computer-readable storage unit by averaging the predictions of the three best-performing models to generate a final output to predict the vanadium content of soil, and tocompare the same with the observed vanadium of the soil sample.SUMMARY OF THE INVENTION:The present invention utilizes easily obtainable mid-infrared FTIR input data sets, comparedto the existing state-of-the-art machine learning models using laser-induced breakdownspectroscopy (LIBS), atomic absorption spectrophotometry (AAS), inductively coupledplasma atomic emission spectroscopy (ICP-AES), inductively coupled plasma massspectroscopy (ICP-MS) and X-ray fluorescence spectrometry (XRFS) combining two or moreof the said methods, which is far more complex. LIBS, being an atomic spectroscopy technique,provides only elemental composition data, limiting its ability to predict trace-levelquantification due to its high detection limits. Soil's complex, heterogeneous compositionamplifies matrix effects, complicating analysis. Optimizing experimental parameters andadvanced data processing are needed to improve vanadium detection. While advancements inLIBS instruments may mitigate these issues, significant progress is expected to takeconsiderable time. NIR spectroscopy detects molecular information, and the reflectance spectraof soil in the visible to near-infrared range have been utilized to estimate vanadium. However,soil properties exhibit varying correlations with elements and molecules, making spectralinterpretation complex. While NIR captures overtones and combination bands of molecularvibrations, its spectra often feature broader and less distinct bands, complicating theidentification of different vanadium bonds. Mid-infrared spectroscopy, on the other hand,targets the basic vibrational modes of molecular bonds and provides more specificity. Theseresult in sharp, well-defined absorption bands that can be directly associated with differentbonds, making mid-infrared spectroscopy more effective for analyzing different bonds ofvanadium present in the soil. FTIR spectroscopy can optimize the analysis of multiple heavyminerals present in the soil using a single analytical technique, offering a more efficient andaccurate approach to soil pollution. It is a highly efficient, rapid, and low-cost method.Furthermore, integrating modern machine learning algorithms to process the mid-infraredFTIR input spectral variables makes it easier to suitably modify the invention depending on thesoil type, which is one of the significant advancements compared to the existing state-of-the-art. The present invention can be suitably designed to be incorporated into handheld FTIRspectrometers, equipped with computer integration, enabling users to perform real-time, on-site analysis of soil samples, thereby reducing the need for extensive laboratory testing anddecreasing operational costs and time delays. The present invention is a novel, innovativeintegration of FTIR technology and machine learning techniques, which presents a unique andnon-obvious solution for predicting the vanadium content of soil.- FTIR spectroscopy: Fourier Transform Infrared Spectroscopy (FTIR) is a technique thatuses an optical method to acquire the molecular composition of a sample by measuring itsabsorption of infrared light.- Vanadium: Vanadium is a silvery white soft transition metal with atomic number 23,regarded as a strategic element due to its critical role in both traditional industries and theemerging green energy transition- X-ray fluorescence spectroscopy: An analytical technique that uses X-ray radiation tocharacterize materials. It works by exciting core electrons in atoms, leading to the emissionor absorption of X-rays with energies characteristic of the elements present and theirchemical environment.BRIEF DESCRIPTION OF THE ACCOMPANYING DRAWINGS:Table 1 compares the statistical parameter values found in earlier studies using differentmethods / equipment / input data / algorithms and the present study, which were used to model andestimate the amount of vanadium content.Figure 1 shows an embodiment of a method for predicting the vanadium content in soilsamples as per the present invention.Figure 2 shows the systematic variations in the absorbance observed for analyzed samples ateight known concentrations of 0.45%, 0.68%, 0.90%, 1.12%, 1.34%, 1.79%, 2.22%, sampleKDL05. V1 to V39 represent the absorbance peak locations related to the different functionalgroups or bonds of vanadium in soil.Figure 3 shows the XRF spectroscopy analysis report providing the elemental composition ofthe soil samples.Figure 4 shows the peak (V) assignment details (for peaks 1 to 39) along with the assignednature of the bond and 2.65% of soil in soil + KBr pellets, using FTIR spectroscopy in therange 1200 to 350 cm-1for theFigure 5 shows the scatter plot displaying the correlation between the observed vanadiumusing XRF (VXRF) and model-estimated vanadium by FTIR for the soil samples using (a)VFTIR.ANN, (b) VFTIR.PLSR, (c) VFTIR.SVR, (d) VFTIR.PCR methods along with the error measures (R2:coefficient of determination; RMSE: root mean square error; MBE: mean bias error).Figure 6 shows the scatter plot displaying the correlation between the observed vanadiumusing XRF (VXRF) and model-estimated vanadium by FTIR for the soil samples using (a)VFTIR.RF, (b) VFTIR.XGB, (c) VFTIR.KNN, (d) VFTIR.MME, methods along with the error measures (R2:coefficient of determination; RMSE: root mean square error; MBE: mean bias error).Figure 7 shows the error measures for vanadium prediction in soil samples determined throughANN, PLSR, SVR, PCR, XGB, RF, KNN and MME methods.Figure 8 shows a Taylor diagram representation of the performance evaluation of differentmodels (ANN, PLSR, SVR, PCR, XGB, RF, KNN and MME) for vanadium estimation in soil.Figure 9 shows a boxplot comparing the variation in the vanadium content measured usingXRF (VXRF) and estimated values using FTIR combined with machine learning methods(VFTIR.ANN, VFTIR.PLSR, VFTIR.SVR, VFTIR.PCR, VFTIR.XGB, VFTIR.RF, VFTIR.KNN, VFTIR.MME) in soilsamplesFigure 10 shows the distribution of the mean bias error of all the models (VFTIR.ANN, VFTIR.PLSR,VFTIR.SVR, VFTIR.PCR, VFTIR.XGB, VFTIR.RF, VFTIR.KNN, VFTIR.MME) used to estimate vanadiumcontent in soil samples.Figure 11 shows the results of a two-tailed paired t-test to compare the difference in the meanof vanadium content obtained via XRF (VXRF, ppm) and model-estimated values using FTIR(VFTIR.ANN, VFTIR.PLSR, VFTIR.SVR, VFTIR.PCR, VFTIR.XGB, VFTIR.RF, VFTIR.KNN, and VFTIR.MME, ppm)at a 99% confidence level and α = 0.01 for soil samples.Figure 12 Results of a two-tailed F-test to compare the difference in the variance of vanadiumobtained via XRF (VXRF, ppm) and model-estimated values using FTIR (VFTIR.ANN, VFTIR.PLSR,VFTIR.SVR, VFTIR.PCR, VFTIR.XGB, VFTIR.RF, VFTIR.KNN, and VFTIR.MME, ppm) at a 99% confidencelevel and α = 0.01 for soil samples.Figure 13 shows a paired bar chart comparing the observed (VXRF) and multi-model predicted(VFTIR.MME) vanadium content of soil samples.DETAILED DESCRIPTION OF A PREFERRED EMBODIMENT OF THEINVENTION:Vanadium is a silvery white soft transition metal with atomic number 23, regarded as a strategicelement due to its critical role in both traditional industries and the emerging green energytransition. The present invention is a method and a vanadium predicting system for predictingthe trace amounts of vanadium content in soil (ppm). The Fourier transform mid-infraredspectra data of soil may be used to predict the vanadium content of soil samples. FourierTransform Infrared Spectroscopy (FTIR) is a technique that uses an optical method to acquirethe molecular composition of a sample by measuring its absorption of infrared light.Specifically, the area under the curve (AUC) (or peaks) of the thirty-nine mid-infraredabsorption bands sensitive to the functional groups and bonds related to vanadium in soil setas independent variables may be fitted using ANN, PLSR, SVR, PCR, RF, KNN, and XGBregression models. The data from mid-infrared FTIR spectra can be utilized to predict thevanadium content in the preferred soil using the generated regression models. It is possible toadapt and use the generated models to predict the vanadium content of soil from differentlocations.Referring to Figure 1, the present invention describes how the vanadium prediction may beachieved. Figure 1 represents the implementation of the vanadium prediction technique 202.Computers, digital video discs (DVD), flash memory disks, onboard read-only memory(ROM), random access memory (RAM), storage devices (SSD / HDD / USB), and compact discs(CD) are a few examples of components of vanadium forecasting systems. Personal digitalassistants (PDAs), mobile standalone analyzers, centralized and / or localized analyzers, andother similar equipment can all be used in method 202 to predict the vanadium content of soilsamples. Soil samples collected from agricultural field 204 and soil samples of unknownvanadium content 206 may be collected, processed, and prepared 208 for FTIR analysis 210and X-ray fluorescence (XRF) analysis 212. X-ray fluorescence spectroscopy is an analyticaltechnique that uses X-ray radiation to characterize materials. It works by exciting core electronsin atoms, leading to the emission or absorption of X-rays with energies characteristic of theelements present and their chemical environment. The vanadium content obtained from the X-ray fluorescence (XRF) analysis may be used as the reference (VXRF, ppm) 214. Users mayprepare the processed soil samples at a known dilution factor 216, according to concentrationincrement for mid-infrared Fourier transform infrared spectroscopy analysis 218, and theresulting spectral signature for all samples (soil+KBr) between 1200 to 350 cm-1 may berecorded 220. The users must make sure that whatever concentration range is chosen for onesample, it should be followed by the same for the other samples to maintain consistency. Allthe soil samples demonstrate well-defined absorbance peaks corresponding to bonds of eachchemical component present in the soil. The FTIR spectra of soil in Figure 2 show thecharacteristic absorption bands produced by functional groups and bonds in the range of 1200and 350 cm-1. These absorbance peaks are a fundamental marker for identifying the chemicalbonds involving vanadium and oxygen in soil and estimating their abundance.K-fold cross-validation of the soil samples 222 may be used to split the samples into training224 and test sets 226. The AUC of the identified absorption bands sensitive to the vanadiumcomplexes in soil can be used as independent variables 228 to analyze the dependent variable214.A data analyzer 230 may be used to conduct the analysis. In general, the present invention useseight regression models to analyze the selected data, including ANN, PLSR, SVR, PCR, RF,KNN, and XGB. A multi-model estimation (MME) method was used in the present invention,using the average of the three best-performing models 230. The regressor model in the dataanalyzer may predict the vanadium from the training set of soil samples 232. The training setis used to predict the vanadium of the soil test pellets using FTIR data 234. The averagevanadium content in each sample (08 pellets), considering the known dilution factor, may becomputed 236.Artificial Neural Networks (ANNs) are a type of machine learning algorithm inspired by thearchitecture and operations of the human brain. They are composed of layers of connectednodes, or "neurons," that handle input data via weighted links. ANNs can be envisioned aslayers of linked computation units called "neurons", where each link has a specific "weight"based on the type of relationship. Some advantages of using ANN are that it does not requireany prior assumption of the underlying process of the functional form of the dependency, andalso works by offering numerical models capable of establishing relationships betweencomplex nonlinear data problems and can reduce noise in the data (Otchere et al., 2021).Neurons transfer values from input to output neurons, using the weights as intermediateparameters and learning through an iterative process of updating weights connecting theneurons (Aggarwal, 2018). The ANN regression model was built usingthe MLPRegressor implementation from scikit-learn, with hyperparameters optimizedthrough Optuna.Partial least squares regression model (PLSR) is a novel multivariate statistical data analysisthat combines methods of principal component analysis (PCA) and multiple linear regression.This method is based on factor analysis that can effectively eliminate multiple correlations. Ithas several advantages, particularly when dealing with multiple correlations between variablesand small sample sizes, in linear regression analysis. PLSR is often used for spectral dataanalysis (Jia et al., 2024; Wold et al., 2001). The model was made with the help of the cross-decomposition module of the scikit-learn library in the Python programming environment(Mishra et al., 2023; Shukla et al., 2023; Vinod et al., 2024).Support Vector Regression (SVR) was developed by Vapnik and his group in 1990's, SVR hasthe strongest mathematical model for classification and regression (Fung et al., 2020). SVR isa mathematical extension of the Neural Network (Chandra & Bedi, 2021). It uses only a subsetof the training data (support vectors) to construct a model that is most generalizable and can belinear or nonlinear, depending on the kernel function used (Boucher et al., 2015). TheGridSearchCV function and the SVR module of the scikit-learn library in the Pythonprogramming environment were used for model optimization and training.Principal Component Regression (PCR) is a multivariate technique that analyzes a data tablein which observations are described by several intercorrelated quantitative dependent variables(Abdi & Williams, 2010) and is used by almost all scientific disciplines. PCR reduces thenumber of features in the dataset while keeping the most important information. Linear algebrais used to transform data into new features called principal components, by calculatingeigenvectors and eigenvalues from the covariance matrix (Mackiewicz & Ratajczak, 1993).PCR also represents the pattern of similarity of the observations and the variables by displayingthem as points in maps (Karamizadeh et al., 2013). Model optimization and training have beenperformed in the Python programming language, using the sci-kit learn.Random forest (RF) algorithm is a supervised machine learning algorithm that is extremelypopular and is used for Classification and Regression problems in machine learning. RF usesmany decision trees to make better predictions. Each tree looks at different random parts of thedata, and their results are combined by voting for classification or averaging for regression; themore trees it has, the more robust a forest is, to improve accuracy and reduce errors (Meenal etal., 2021; Salman et al., 2024). The present invention uses a random forest regressor modelmade by the ensemble module of the scikit-learn library in the Python programmingenvironment. The hyperparameter tuning may be done with the help of RandomizedSearchCVand GridSearchCV; upon comparison with the default hyperparameter setting of the model, thebest estimator was used for the modeling.XGBoost or Extreme Gradient Boosting regression model, tree boosting is a highly effectiveand widely used machine learning method. In the present invention, we used a tree boostingsystem called XGBoost, based on the Gradient Boosting Decision Tree (GBDT) algorithm.The most important factor of XGBoost is its scalability, it is able to solve real-world scaleproblems using a minimal amount of resources (T. Chen & Guestrin, 2016). XGB appliesregularization, iteratively adds decision trees, and performs feature splitting by summing leafnode scores to generate new features, reducing prediction variance and avoiding overfitting.Synthetic algorithm that combines basis functions and weights to obtain good data fittingresults. Due to its advantages of computational efficiency, feature importance analysis, andhandling of missed values, XGB is widely used in data regression and classification (Wang etal., 2024). It is a novel implementation method of the GBDT algorithm to increase robustnessand prevent overfitting. XGB exploits the principle of boosting weak learners (CARTsgenerally) using the gradient descent algorithm (Abbasimehr et al., 2023). The GridSearchCVfunction and the randomized SearchCV function with XGBRegressor class of the xgboostlibrary in scikit-learn in the Python programming environment were used for modeloptimization and training.k-Nearest Neighbors regression model, the algorithm proposed in 1967, is a simple buteffective method that is used both for classification and regression. In regression, KNN predictscontinuous values by averaging the values of the k-nearest neighbors. The algorithm works byidentifying the "k" nearest data points (neighbors) to a given input and making a predictionbased on the majority class for classification or the average value for regression. It only requiresspecifying the number of neighbors, making it convenient for on-site implementation (Xiong& Yao, 2021). Since KNN makes no assumptions about the underlying data distribution, itmakes it a non-parametric and instance-based learning method, and is a preferred algorithmdue to its simplicity, easy understanding (Deng et al., 2016).The predicted and observed vanadium content in soil samples may be compared, 238 and modelvalidation statistics and test of significance (t & F tests) may be conducted at a 99% confidencelevel, 240 which proves that the observed and predicted vanadium of soil samples are similar242.Flowchart (Figure 1) Number Reference Table202 Vanadium (V) content predictor for soil samples204 Soil samples collection from agricultural field206 Soil sample(s) (Unknown vanadium content)208 Sample(s) processing & preparation for XRF & FTIR analysis (size -74μm)210 Sample(s) processing & preparation for FTIR analysis (size -74μm)212 X-ray fluorescence analysis214 Vanadium content measured by XRF (VXRF, ppm)216 Preparation of sample + KBr pellets at 08 known "dilution factor" foreach soil sample218 Fourier Transform Infrared Spectroscopy (FTIR) Analysis220 Spectral signature recorded for all sample+KBr pellets at 1200-350 cm-1222 Model development and K-fold cross-validation (CV) using 23:01 split224 Modeling using a "training set"226 Validation using "test set"228 FTIR spectral response (sensitive groups). Area under the curve (AUC)calculations in specific zones230 Modeling: ANN, PLSR, SVR, PCR, RF, KNN, XGB, MME (MME:multi-model estimation of Vanadium content by taking the average of thebest 3 models)232 Prediction of vanadium content using a regressor model from the"training set"234 Prediction of vanadium content for the sample test pellets using FTIRdata236 Average vanadium content in each soil sample (08 pellets) consideringthe known dilution factor238 Comparison of predicted vanadium content (VFTIR.MME, ppm) withobserved vanadium content (VXRF, ppm)240 Model validation statistics (R2, RMSE, RMSE%, MBE, MBE%, MAE),test of significance (t & F test) at 99% Confidence level242 Results (vanadium content of soil samples)Case study ExamplesDhapa agricultural fieldDhapa is a solid waste disposal area heavily impacted by toxic waste, extending an area of21.47 ha, located in the eastern fringes of Kolkata, West Bengal, India (Paul et al., 2014). Theregion is surrounded by extensive agricultural, with vegetable cultivation. This makes the areaparticularly significant, as there is a high risk of heavy metals like vanadium leaching into thesoil and crops. Such contamination poses serious health hazards to consumers.Following IOS standards (International Organization for Standardization, 2006, 2017, 2018) atotal of 24 soil samples were collected for analysis. For each sample, five subsamples ofapproximately 1 kg each were taken from a 1m x 1m area-one from each corner and one fromthe center. These subsamples were thoroughly mixed to create a single composite sample. Fromthis composite, a 1 kg portion was used for further analysis. The samples were then dried andsieved to 74 μm for FTIR and XRF analysis.X-ray analysis has been established as an important method for elemental analysis; it is knownas a non-destructive analysis method that is capable of analyzing materials in variousaggregated states. The X-ray fluorescence spectroscopy of all the samples used in the studywas conducted in the Bruker S8 Tiger, Germany, XRF, DST-FIST facility, IISER Kolkata. Thesamples were prepared by the press pellet method, in this method the samples weighing 4 gramswere mixed with 1 gram of Boric Acid (Merck) were combined well with the agate mortar thenfilled into the aluminium mold and pressed for roughly 10 seconds at 15-ton pressure to preparethe pellet for the major and minor element analysis. Figure 3 includes the data range ofvanadium content for all soil samples (reported in ppm). The vanadium content obtained fromthe X-ray fluorescence analysis may be used as the reference (VXRF) for the vanadium contentof soil.FTIR spectroscopy has been widely utilized as a fast and accurate qualitative or quantitativeanalytical tool since 1970 (Gong et al., 2024). FTIR spectroscopy is based on the atoms'vibration and rotation, an absorption-based optical analysis that identifies the energy ofmolecular vibrations after the sample is subjected to infrared radiation. FTIR spectroscopy maybe used to obtain the mid-infrared spectrum of samples and identify the valid absorption peaksassociated with the different vanadium species in soil. FTIR spectral analysis was performedusing an INVENIO S model (BRUKER OPTIK GmbH, Bremen, Germany) (Mishra et al.,2023; Shukla et al., 2023; Vinod et al., 2024). Sample pellets used for analysis were preparedby mixing a soil sample (size -74 μm) and potassium bromide powder of IR spectroscopygrade. Eight pellets for each of the 24 samples were prepared with a known dilution factor. TheKBr may be fixed at 220 ± 0.20 mg, and the soil content in the pellet may be varied in 0.45,0.68, 0.90, 1.12, 1.34, 1.79, 2.22, and 2.65 percent. One hundred ninety-two sample pellets ofsoil have been prepared, and their mid-infrared spectral signatures were recorded using BrukerFTIR. The recorded spectra have been trimmed to hold the frequency range of interest from1200 to 350 cm-1. The resulting absorbance values may be plotted as an absorbance versusfrequency plot (Y-X plot). Baseline correction is a necessary preprocessing step for quantitativespectra analysis to enhance the absorbance features. The spectra of pure KBr pellets werededucted from the sample pellet spectra to correct the baseline and eliminate the KBr spectralsignature.A. VariablesSoil is a heterogeneous material mainly composed of carbonaceous and inorganic components.Soil consists of elemental vanadium, which is a redox-sensitive metal that is released to soilsby weathering and anthropogenic emissions (Larsson, 2014). In environmental matrices likesoil, free, uncomplexed vanadium ions are highly unlikely to exist in any significant quantity.Vanadium is a transition metal with a strong affinity for oxygen and other donor atoms (likenitrogen, sulfur, or carbon in organic matter). It readily forms coordination complexes withthese atoms. In soil, vanadium is primarily governed by the vanadate sorption ontoFe / Al(hydr)oxides and vanadyl binding with the natural organic matter (Trujillo-GonzAlez etal., 2024). Vanadium-oxygen bonds have specific electronic transitions that result incharacteristic absorption features in FTIR spectroscopy. Figure 2 depicts the systematicvariations in the FTIR absorbance spectra observed for the soil + KBr pellets of the sampleKLD05 at eight concentrations (0.45%, 0.68%, 0.90%, 1.12%, 1.34%, 1.79%, 2.22%, and2.65%). The position and intensity of infrared absorption peaks correlate with the molecularcomposition and the concentration of the identified bonds. All the identified peaks, their onset,and offset in wavenumber (cm-1), along with the corresponding functional group, are given inFigure 4. According to the literature (AbdelAziz et al., 2014; W. Chen et al., 2004), the FTIRspectra of vanadium compounds are typically classified based on the types of vanadium-oxygen bonds present. It can be categorized into four distinct absorption bands: the 700-400cm-1band relating to V-O bond; the 850-500 cm-1 band, corresponding to bridging V-O-Vbonds; the 1050-900 cm-1 band, associated with terminal V=O bonds (vanadium oxygendouble-bond); 1062 cm-1corresponding to vanadyl complex.The spike at 365 cm-1is ascribed to the stretching modes of the VO2 band, and 374 cm-1is forthe V2O3. These peaks are given in notations V01and V02 (Botto et al., 1997) (Figure 4). Thecomplex band in the range 700-400 cm-1is associated with the bridging V-O-V bonds withinthe soil and is thought to play a crucial role in catalytic reactions. The peaks at 411.919,478.789, and 480 cm-1are attributed to the V2O5 bonds, which are identified as V03, V05, andV06 (Botto et al., 1997; Iordanova et al., 1996, 1996; Liu et al., 2006; Ortega-Reyes & Avila-Garcia, 2015). The spike at 423.083 cm-1is of the VO2 bond, and 498.69 cm-1corresponds tothe V3O5 bond identified as V04 and V07 (Ortega-Reyes & Avila-Garcia, 2015), respectively.The band ranging from 850-500 cm-1 majorly represents the bridging V-O-V bond. The peaks531, 730, 763, 790, 823 and 825 cm-1 represents the V-O-V symmetric stretch bridging bondsof vanadium, denoted as V09, V20, V22, V23, V24 and V25 (Botto et al., 1997; Chary et al.,2003; Farahmandjou, 2017; Kianfar, 2015; Kumawat et al., 2021; Liu et al., 2006; Malta &Torresi, 2005; Miyata et al., 1990; Ortega-Reyes & Avila-Garcia, 2015; Pinna et al., 2003; Xueet al., 2010). Peaks denoted as V08, V10, V15, V17 and V18, corresponding to thewavenumbers 520, 550, 638, 669, and 682 cm-1, respectively, represent the V2O3 compound(Botto et al., 1997; Liu et al., 2006; Malta & Torresi, 2005; Ortega-Reyes & Avila-Garcia,2015; Xue et al., 2010; H. Zhang et al., 2015). V-O stretching band of vanadium and oxygenat 562, 578 and 885 cm-1are identified as V11, V12 and V27 (Rammah et al., 2020; Xue etal., 2010) respectively. The absorption bands at 599, 620, 652.358, 721.499 cm-1 are for theVO2 bond and identified as V13, V14, V16 and V19 (Botto et al., 1997; Kianfar, 2015;Kumawat et al., 2021; Liu et al., 2006; Ortega-Reyes & Avila-Garcia, 2015). V2O5 bonds aregiven by the absorption band at 750 and 830 cm-1indicated as V21 and V26, respectively(Aurbach et al., 2001; Liu et al., 2006; Malta & Torresi, 2005). The stretching vibration modeof vanadyl group structures in the FTIR spectrum of soil samples appears in the range 1200-900 cm-1. The peaks represented as V29, V30, V31, V33, V34, V36, V37, V38, correspondsto the vanadyl group with the presences of both single and the double stretching bond and haveabsorption band of 921, 944, 957, 983, 997, 1022, 1034, 1040 cm-1respectively (Bosco et al.,2015; Botto et al., 1997; Farahmandjou, 2017; Iordanova et al., 1996; Kianfar, 2015; Liu et al.,2006; Malta & Torresi, 2005; Miyata et al., 1990; O'Dwyer et al., 2007; Ortega-Reyes & Avila-Garcia, 2015; Pinna et al., 2003; Rammah et al., 2020; Schramlmarth, 1990; Xue et al., 2010)(Figure 4). The peaks V28, V32, V35 and V39 correspond to the V=O bonds with wavenumber909, 967, 1010, and 1062 cm-1(Aurbach et al., 2001; Bosco et al., 2015; Farahmandjou, 2017;Malta & Torresi, 2005; O'Dwyer et al., 2007; Ortega-Reyes & Avila-Garcia, 2015; Xue et al.,2010; H. Zhang et al., 2015). Using a detailed analysis of peak assignments corresponding totheir respective chemical bonds, 39 peaks were identified, as illustrated in Figure 4.B. ResultsThe vanadium content of the soil samples may be predicted using ANN, PLSR, SVR, PCR,RF, KNN, and XGB methods. The results obtained for all the methods with the observed andpredicted vanadium of soil are plotted as a simple linear fit along with the coefficient ofdetermination (R2), root mean square error (RMSE), and mean bias error (MBE) (Figures 5 (ad) and 6 (a-d)). The present invention results in good correlation with the observed andpredicted vanadium for the soil samples. The ANN model delivered an R2of 0.951, RMSE%of 11.985, MBE% of -0.544, and MAE of 0.173 ppm. PLSR model attained an R2value of0.950, RMSE% of 12.000, MBE% of 0.047, and MAE of 0.171 ppm; SVR yielded an R2 valueof 0.950, RMSE% of 12.009, MBE of 0.446%, and MAE of 0.170 ppm; PCR produced an R2value of 0.949, an RMSE% of 12.208, an MBE% of 0.214, and an MAE of 0.173ppm; RFachieved an R2of 0.942, an RMSE% of 12.955, an MBE% of 0.413, and an MAE of 0.185ppm;XGB attained R2of 0.949, RMSE% of 12.218, MBE% of 0.287, and MAE of 0.178 ppm; KNNdelivered an R2of 0.933, RMSE% of 14.094, MBE% of 1.222, and MAE of 0.205 ppm. Amulti-model estimation approach is introduced in this invention, taking the mean of the threetop models (ANN, PLSR, SVR). Figures 6(d) show the scatter plot displaying the correlationbetween the observed vanadium content using X-Ray fluorescence (XRF) and multi-modelpredicted vanadium using FTIR (VFTIR.MME), including the error measures. Figure 7 comparesthe vanadium predicted using the soil spectra combined with machine learning models (ANN,PLSR, SVR, PCR, XGB, RF, KNN, and MME) and the vanadium content obtained throughthe XRF analysis of the soil. Figure 8 depicts the Taylor plot diagram depicting the comparativeperformance of ANN, PLSR, SVR, PCR, XGB, RF, KNN, and MME models used forvanadium content estimation in soil, highlighting their correlation, standard deviation, andcentered root mean square error. Models plotted nearer to the reference XRF point performbetter with higher correlation, similar standard deviation, and lower RMSE. Figure 9 shows aboxplot comparing the variation in the vanadium content measured using XRF (VXRF) andestimated values using FTIR combined with machine learning methods (VFTIR.MME, VFTIR.ANN,VFTIR.PLSR, VFTIR.SVR, VFTIR.PCR, VFTIR.RF, VFTIR.KNN, VFTIR.XGB) in soil samples. This supports thesimilarity between the interquartile range and means of the vanadium estimated using XRF andmodeled using the present invention. Figure 10 shows the distribution of the mean bias errorof all the models (VFTIR.MME, VFTIR.ANN, VFTIR.PLSR, VFTIR.SVR, VFTIR.PCR, VFTIR.RF, VFTIR.KNN,VFTIR.XGB) used to estimate vanadium content in soil samples. The mean bias error of all themodels indicates acceptable values of -0.010, 0.001, 0.008, 0.004, 0.005, 0.008, 0.023, and0.000 for vanadium using ANN, PLSR, SVR, PCR, XGB, RF, KNN, and MME models,respectively. The three best-performing models (ANN, PLSR, and SVR) were averaged for amulti-model estimation approach. The developed MME estimation method presents greaterprecision and robustness than individual models, as evidenced by an R2value of 0.952, anRMSE% value of 11.867, an MBE% value of -0.017, and an MAE value of 0.169 ppm (Figure6h).Statistical hypothesis testing conducted (two-tailed paired t-test for the mean and two-tailed Ftest for variances at the confidence level (CL) of 99% (where α = 0.01)) establishes that thevanadium predicted from soil spectra is not significantly different from that observed withXRF. Figures 11 and 12 provide the details of the two-tailed paired t-test for the mean and two-tailed F-test for variances, respectively, where μ represents the mean, σ² is the sample variance,n is the number of observations, and df denotes the degrees of freedom; the tstat is the calculatedt-statistic, which is compared with the tcritical, the critical value for the test. The μd is thehypothesized mean difference, and the p-value indicates the probability distribution for the test.Based on the p-value and the α, which is the significance level, we compute the confidenceinterval (CI) at a given confidence level (CL). The p-values greater than α indicate that the nullhypothesis (Ho: μd = 0) cannot be rejected, implying that there is no significant differencebetween the predicted and observed values.Figure 13 shows the observed (XRF) and predicted (FTIR with MME) vanadium content ofsoil samples. The present invention indicates similar and comparable results between the multi-model estimated vanadium content (VFTIR.MME, ppm) and the vanadium content in soilmeasured by XRF (VXRF, ppm).In summary, FTIR spectroscopic data of soil have been used to identify the presence ofvanadium in soil and to derive relations to quantify the vanadium content of soil. Theabsorption bands related to different bonds, incorporating vanadium and oxygen in theirframework in soil and their AUC obtained from the FTIR spectra of soil, have been fitted toANN, PLSR, SVR, PCR, XGB, RF, and KNN regression models. A multi-model estimationapproach utilizing the best of the three utilized models (ANN, PLSR, and SVR) can yield goodprediction results because of the good correlation coefficient and low error bias.As more data from different soils will be input, more realistic results can be generated.However, the vanadium content in soil is greatly influenced by the parent rocks found nearby,redox potential, pH, organic matter content, mining, industrial processes, and the use offertilizers and pesticides (L. Chen et al., 2021; Guagliardi et al., 2018; Tambat et al., 2024).The present invention may employ readily obtainable FTIR data sets instead of moresophisticated and expensive analytical tests and procedures. It can be extended to compute thevanadium content of soil samples from other regions.The descriptions given of the preferred embodiments of this invention are meant to serve asexamples and clarifications. They are not intended to restrict the invention's use to the preciseforms that are shown. Based on the information given, alternatives, changes, or equivalentsmay be used, allowing other experts in the field to successfully apply the invention in a varietyof ways and with adjustments tailored to applications.ADVANTAGES OVER THE EXISTING METHODS:The standard analytical techniques used for vanadium detection and quantification includestandard atomic spectroscopic techniques and separation methods coupled with sensitivedetectors. Separation methods include capillary electrophoresis (CE) and liquidchromatography (LC). Atomic spectroscopic methods used for the determination of vanadiuminclude atomic absorption spectroscopy (AAS) with flame and graphite tube atomizers,inductively coupled plasma optical emission spectrometry (ICP-OES), inductively coupledplasma mass spectrometry (ICP-MS), and spectrophotometric methods (Z. L. Chen & Owens,2008; Pyrzyska, 2004). Vanadium content estimation through these methods is labour-intensive, destructive, and economically not feasible. Moreover, atomic spectroscopic methodsrequire complex sample preparation methods, and direct analysis of vanadium is not feasibledue to the relatively low concentrations of vanadium found in samples such as soil, as comparedto other metals. Methods like AAS and ICP-OES are tedious due to high-pressuredecomposition of samples in corrosive chemicals (HF, aqua regia) or fusion necessary totransfer the analyte into the solution (Pouzar et al., 2011). Thus, several practical and economicreasons invoke the current trend towards fast and non-destructive methods like LIBS and XRF.Though non-destructive, estimating vanadium concentrations of soil samples using LIBS andXRF is challenging due to its complex matrices, leading to sensitivity issues due to matrixeffect, that can significantly impact its accuracy (Lucena et al., 1998; Racusan Ghircoias et al.,2023; Ravansari et al., 2020). The low sensitivity and high cost of these methods compromisepracticality for large sample analysis.FTIR spectroscopy presents a promising, rapid, and efficient method for estimating thevanadium content of soil samples. The present invention utilizes easily obtained mid-infraredFTIR input data sets, avoiding complex sample preparation and procedures. Moreover, it iscost-effective, uses fewer consumables, and is user-friendly, requiring simpler instrumentation,and can predict vanadium in low concentrations (in ppm) when combined with machinelearning regression techniques.Table 1

Claims

1. A method for the rapid estimation of soil vanadium content using mid-infrared FTIR spectral data analysis and a multi-model machine learning framework, the said method (202) comprising; - collecting soil samples from agricultural field (204); - collecting soil samples with unknown vanadium content (206); - processing and preparing the samples for x-ray fluorescene (XRF) and Fourier Transform infrared spectroscopy (FTIR) analysis (208); - processing and preparing sample for FTIR analysis (210); - processing and preparing sample for XRF analysis (212); - obtaining vanadium content from the XRF analysis for future reference (VXRF, ppm) (214); - preparing and processing soil samples with KBr pellets for each soil sample at a known dilution factor (8) according to concentration increment for mid-infrared fourier transform infrared spectroscopy analysis (216, 218); - recording the resulting spectral signature for all samples (soil+KBr pellets at 1200-350 cm1 ) (220); - maintaining consistency of concentration range of one sample followed by the same for other samples; - developing a model with K-fold cross-validation (CV) of the soil samples providing 23:01 split (222), splitting the samples into training (224) and test set (226) for validation; - calculating Area under the curve (AVC) of the identified absorption bands sensitive to the vanadium complexes in soil to be considered as independent variables (228) to analyze the dependent variable (214); wherein, - a data analyzer (230) conducts the analysis, employing eight regression models to analyze the selected data, wherein a multi-model estimation method estimates the vanadium content by taking the average of the best three models (230), - wherein the regressor model in the data analyzer predicts vanadium content from the training set of soil samples (232); - wherein prediction of vanadium content for the sample soil test pellets made employing FTIR data (234), wherein average vanadium content in each soil sample (08 pellets) is computed, considering the known dilution factor (236), wherein the predicted VFTIR, MME, ppm and observed VXRF, ppm vanadium content in soil samples is compared (238) and model validation statistics and test of significance (t and F test) is conducted at a 99% confidence level (240) to prove that the observed and predicted vanadium of soil samples are similar (242).

2. The method for the rapid estimation of soil vanadium content as claimed in claim 1 wherein multiple regression models employed to analyze the selected data, includes ANN, PLSR, SVR, PCR, RF, KNN and XGB.

3. A system for rapid estimation of vanadium content in soil using mid-infrared FTIR spectral data analysis and a multi-model machine learning framework, characterised in that said system comprises: a processor, a computer-readable storage unit, wherein the processor being configured to calculate a multi-model estimation approach by averaging the predictions of the three best-performing models to generate a final output to predict the vanadium content in soil, a data collector unit associated with the processor for receiving data, including the AUC (or height) of spectral absorbance peaks sensitive to the vanadium bonds in soil, a data analyzer unit associated with the processor for identifying and training a 'artificial neural network regression' model (ANN), a 'partial least square regression' model (PLSR), a 'support vector regression' model (SVR), ), a 'principal component regression' model (PCR), a 'XGBoost' regression model (XGB), a 'random forest regression' model (RF) a 'k-nearest neighbor' model (KNN), for determining a multi-model estimation approach by averaging the predictions of the three best-performing models (ANN, PLSR, and SVR) to generate a final output to predict the vanadium content in soil.