scholarly journals Estimation of effective calibration sample size using visible near infrared spectroscopy: deep learning vs machine learning

2019 ◽  
Author(s):  
Wartini Ng ◽  
Budiman Minasny ◽  
Wanderson de Sousa Mendes ◽  
José A. M. Demattê
Keyword(s):  
Machine Learning ◽  
Deep Learning ◽  
Soil Properties ◽  
Sample Size ◽  
Near Infrared ◽  
Model Performance ◽  
Clay Content ◽  
Least Squares Regression ◽  
Soil Attributes ◽  
Calibration Dataset

Abstract. The number of samples used in the calibration dataset affects the quality of the generated predictive models using visible, near and shortwave infrared (VIS-NIR-SWIR) spectroscopy for soil attributes. Recently, convolutional neural network (CNN) is regarded as a highly accurate model for predicting soil properties on a large database, however it has not been ascertained yet how large the sample size should be for CNN model to be effective. This paper aims at providing an estimate of how much calibration samples are needed to improve the model performance of soil properties predictions with CNN. It is hypothesized that the larger the amount of data, the more accurate is the CNN model. The performance of two commonly used machine learning models (Partial least squares regression (PLSR) and Cubist) are compared against the CNN model. A VIS-NIR-SWIR spectral library from Brazil containing 4251 unique sites, with averages of 2–3 samples per depth (a total of 12,044 samples), was divided into calibration (3188 sites) and validation (1063 sites) sets. A subset of the calibration dataset was then created to represent smaller calibration dataset ranging from 125, 300, 500, 1000, 1500, 2000, 2500 and 2700 unique sites, or equivalent to sample size approximately 350, 840, 1400, 2800, 4200, 5600, 7000, and 7650. All three models (PLSR, Cubist, and CNN models) were generated for each sample size of the unique sites for the prediction of five different soil properties, i.e. cation exchange capacity, organic matter, sand, silt and clay content. These calibration subset sampling processes and modelling were repeated ten times to provide a better representation of the model performances. Similar results were observed when the performances of both PLSR and Cubist model were compared to the CNN model where the performance of CNN outweighed the PLSR and Cubist model at sample size of 1500 and 1800 respectively. It can be recommended that deep learning is most efficient for spectral modelling for sample size above 2000. The accuracy of the PLSR and Cubist model seemed to reach a plateau above sample size of 4200 and 5000 respectively. A sensitivity analysis was performed on the CNN model to determine important wavelengths region that affected the predictions of various soil attributes.

scholarly journals The influence of training sample size on the accuracy of deep learning models for the prediction of soil properties with near-infrared spectroscopy data

SOIL ◽  
2020 ◽  
Vol 6 (2) ◽  
pp. 565-578
Author(s):  
Wartini Ng ◽  
Budiman Minasny ◽  
Wanderson de Sousa Mendes ◽  
José Alexandre Melo Demattê
Keyword(s):  
Machine Learning ◽  
Deep Learning ◽  
Soil Properties ◽  
Sample Size ◽  
Training Sample ◽  
Calibration Data ◽  
Learning Models ◽  
Data Set ◽  
Calibration Data Set ◽  
Machine Learning Models

Abstract. The number of samples used in the calibration data set affects the quality of the generated predictive models using visible, near and shortwave infrared (VIS–NIR–SWIR) spectroscopy for soil attributes. Recently, the convolutional neural network (CNN) has been regarded as a highly accurate model for predicting soil properties on a large database. However, it has not yet been ascertained how large the sample size should be for CNN model to be effective. This paper investigates the effect of the training sample size on the accuracy of deep learning and machine learning models. It aims at providing an estimate of how many calibration samples are needed to improve the model performance of soil properties predictions with CNN as compared to conventional machine learning models. In addition, this paper also looks at a way to interpret the CNN models, which are commonly labelled as a black box. It is hypothesised that the performance of machine learning models will increase with an increasing number of training samples, but it will plateau when it reaches a certain number, while the performance of CNN will keep improving. The performances of two machine learning models (partial least squares regression – PLSR; Cubist) are compared against the CNN model. A VIS–NIR–SWIR spectra library from Brazil, containing 4251 unique sites with averages of two to three samples per depth (a total of 12 044 samples), was divided into calibration (3188 sites) and validation (1063 sites) sets. A subset of the calibration data set was then created to represent a smaller calibration data set ranging from 125, 300, 500, 1000, 1500, 2000, 2500 and 2700 unique sites, which is equivalent to a sample size of approximately 350, 840, 1400, 2800, 4200, 5600, 7000 and 7650. All three models (PLSR, Cubist and CNN) were generated for each sample size of the unique sites for the prediction of five different soil properties, i.e. cation exchange capacity, organic carbon, sand, silt and clay content. These calibration subset sampling processes and modelling were repeated 10 times to provide a better representation of the model performances. Learning curves showed that the accuracy increased with an increasing number of training samples. At a lower number of samples (< 1000), PLSR and Cubist performed better than CNN. The performance of CNN outweighed the PLSR and Cubist model at a sample size of 1500 and 1800, respectively. It can be recommended that deep learning is most efficient for spectra modelling for sample sizes above 2000. The accuracy of the PLSR and Cubist model seems to reach a plateau above sample sizes of 4200 and 5000, respectively, while the accuracy of CNN has not plateaued. A sensitivity analysis of the CNN model demonstrated its ability to determine important wavelengths region that affected the predictions of various soil attributes.

scholarly journals Machine Learning Algorithms for Soil Properties Prediction with Treated Vis–NIR Spectrums from the Itatiaia National Park

2019 ◽  
Author(s):  
Yuri Andrei Gelsleichter ◽  
Lúcia Helena Cunha dos Anjos ◽  
Elias Mendes Costa ◽  
Gabriela Valente ◽  
Paula Debiasi ◽  
...  
Keyword(s):  
Machine Learning ◽  
Soil Properties ◽  
National Park ◽  
Near Infrared ◽  
Learning Algorithms ◽  
Management Plan ◽  
Machine Learning Algorithms ◽  
Total Carbon ◽  
Least Squares Regression ◽  
Near Infrared Reflectance

Visible and near-infrared reflectance (Vis&ndash;NIR) techniques are a plausible method to soil analyses. The main objective of the study was to investigate the capacity to predicting soil properties Al, Ca, K, Mg, Na, P, pH, total carbon (TC), H and N, by using different spectral (350&ndash;2500 nm) pre-treatments and machine learning algorithms such as Artificial Neural Network (ANN), Random Forest (RF), Partial Least-squares Regression (PLSR) and Cubist (CB). The 300 soil samples were sampled in the upper part of the Itatiaia National Park (INP), located in Southeastern region of Brazil. The 10 K-fold cross validation was used with the models. The best spectral pre-treatment was the Inverse of Reflectance by a Factor of 104 (IRF4) for TC with CB, giving an averaged R&sup2; among the folds of 0.85, RMSE of 1.96; and 0.67 with 0.041 respectively for H. Into the K-folds models of TC, the highest prediction had a R&sup2; of 0.95. These results are relevant for the INP management plan, and also to similar environments. The good correlation with Vis&ndash;NIR techniques can be used for remote sense monitoring, especially in areas with very restricted access such as INP.

scholarly journals Wavelet geographically weighted regression for spectroscopic modelling of soil properties

2021 ◽  
Vol 11 (1) ◽  
Author(s):  
Yongze Song ◽  
Zefang Shen ◽  
Peng Wu ◽  
R. A. Viscarra Rossel
Keyword(s):  
Organic Carbon ◽  
Soil Properties ◽  
Geographically Weighted Regression ◽  
Near Infrared ◽  
Linear Models ◽  
Spatial Information ◽  
Clay Content ◽  
Weighted Regression ◽  
Organic Carbon Content ◽  
Least Squares Regression

AbstractSoil properties, such as organic carbon, pH and clay content, are critical indicators of ecosystem function. Visible–near infrared (vis–NIR) reflectance spectroscopy has been widely used to cost-efficiently estimate such soil properties. Multivariate modelling, such as partial least squares regression (PLSR), and machine learning are the most common methods for modelling soil properties with spectra. Often, such models do not account for the multiresolution information presented in the vis–NIR signal, or the spatial variation in the data. To address these potential shortcomings, we used wavelets to decompose the vis–NIR spectra of 226 soils from agricultural and forested regions in south-western Western Australia and developed a wavelet geographically weighted regression (WGWR) for estimating soil organic carbon content, clay content and pH. To evaluate the WGWR models, we compared them to linear models derived with multiresolution data from a wavelet decomposition (WLR) and PLSR without multiresolution information. Overall, validation of the WGWR models produced more accurate estimates of the soil properties than WLR and PLSR. Around 3.5–49.1% of the improvement in the estimates was due to the multiresolution analysis and 1.0–5.2% due to the integration of spatial information in the modelling. The WGWR improves the modelling of soil properties with spectra.

scholarly journals Estimation of Soil Cohesion Using Machine Learning Method: A Random Forest Approach

2021 ◽  
Vol 2021 ◽  
pp. 1-14
Author(s):  
Hai-Bang Ly ◽  
Thuy-Anh Nguyen ◽  
Binh Thai Pham
Keyword(s):  
Machine Learning ◽  
Random Forest ◽  
Soil Properties ◽  
Clay Content ◽  
Absolute Error ◽  
Experimental Methods ◽  
Liquid Limit ◽  
Support Vector ◽  
Data Set ◽  
Soil Cohesion

Soil cohesion (C) is one of the critical soil properties and is closely related to basic soil properties such as particle size distribution, pore size, and shear strength. Hence, it is mainly determined by experimental methods. However, the experimental methods are often time-consuming and costly. Therefore, developing an alternative approach based on machine learning (ML) techniques to solve this problem is highly recommended. In this study, machine learning models, namely, support vector machine (SVM), Gaussian regression process (GPR), and random forest (RF), were built based on a data set of 145 soil samples collected from the Da Nang-Quang Ngai expressway project, Vietnam. The database also includes six input parameters, that is, clay content, moisture content, liquid limit, plastic limit, specific gravity, and void ratio. The performance of the model was assessed by three statistical criteria, namely, the correlation coefficient (R), mean absolute error (MAE), and root mean square error (RMSE). The results demonstrated that the proposed RF model could accurately predict soil cohesion with high accuracy (R = 0.891) and low error (RMSE = 3.323 and MAE = 2.511), and its predictive capability is better than SVM and GPR. Therefore, the RF model can be used as a cost-effective approach in predicting soil cohesion forces used in the design and inspection of constructions.

Deep Learning-based Soil Property Prediction for Remediation of Radioactive Contamination in Agriculture

2021 ◽  
Author(s):  
Franck Albinet ◽  
Gerd Dercon ◽  
Tetsuya Eguchi
Keyword(s):  
Deep Learning ◽  
Soil Properties ◽  
Radioactive Contamination ◽  
The United States ◽  
Soil Survey ◽  
United States Department ◽  
Least Squares Regression ◽  
Transfer Factors ◽  
Wide Range ◽  
Plant Transfer

&lt;p&gt;The Joint IAEA/FAO Division of Nuclear Techniques in Food and Agriculture, through its Soil and Water Management &amp; Crop Nutrition Laboratory (SWMCNL), launched in October 2019, a new Coordinated Research Project (D15019) called &amp;#8220;Monitoring and Predicting Radionuclide Uptake and Dynamics for Optimizing Remediation of Radioactive Contamination in Agriculture''. Within this context, the high-throughput characterization of soil properties in general and the estimation of soil-to-plant transfer factors of radionuclides are of critical importance.&lt;/p&gt;&lt;p&gt;For several decades, soil researchers have been successfully using near and mid-infrared spectroscopy (MIRS) techniques to estimate a wide range of soil physical, chemical and biological properties such as carbon (C), Cation Exchange Capacities (CEC), among others. However, models developed were often limited in scope as only small and region-specific MIR spectra libraries of soils were accessible.&lt;/p&gt;&lt;p&gt;This situation of data scarcity is changing radically today with the availability of large and growing library of MIR-scanned soil samples maintained by the National Soil Survey Center (NSSC) Kellogg Soil Survey Laboratory (KSSL) from the United States Department of Agriculture (USDA-NRCS) and the Global Soil Laboratory Network (GLOSOLAN) initiative of the Food Agency Organization (FAO). As a result, the unprecedented volume of data now available allows soil science researchers to increasingly shift their focus from traditional modeling techniques such as PLSR (Partial Least Squares Regression) to classes of modeling approaches, such as Ensemble Learning or Deep Learning, that have proven to outperform PLSR on most soil properties prediction in a large data regime.&lt;/p&gt;&lt;p&gt;As part of our research, the opportunity to train higher capacity models on the KSSL large dataset (all soil taxonomic orders included ~ 50K samples) makes it possible to reach a quality of prediction for exchangeable potassium so far unsurpassed with a Residual Prediction Deviation (RPD) around 3. Potassium is known for its difficulty of being predicted but remains extremely important in the context of remediation of radioactive contamination after a nuclear accident. Potassium can help reduce the uptake of radiocaesium by crops, as it competes with radiocaesium in soil-to-plant transfer.&lt;/p&gt;&lt;p&gt;To ensure informed decision making, we also guarantee that (i) individual predictions uncertainty is estimated (using Monte Carlo Dropout) and (ii) individual predictions can be interpreted (i.e. how much specific MIRS wavenumber regions contribute to the prediction) using methods such as Shapley Additive exPlanations (SHAP) values.&lt;/p&gt;&lt;p&gt;SWMCNL is now a member of the GLOSOLAN network, which helps enhance the usability of MIRS for soil monitoring worldwide. SWMCNL is further developing training packages on the use of traditional and advanced mathematical techniques to process MIRS data for predicting soil properties. This training package has been tested in October 2020 with thirteen staff members of the FAO/IAEA Laboratories in Seibersdorf, Austria.&lt;/p&gt;

Nitrogen, phosphorus, and potassium prediction in soils, using infrared spectroscopy

Soil Research ◽  
2011 ◽  
Vol 49 (2) ◽  
pp. 166 ◽  
Author(s):  
Yongni Shao ◽  
Yong He
Keyword(s):  
Infrared Spectroscopy ◽  
Soil Properties ◽  
Least Squares ◽  
Near Infrared ◽  
Soil Analysis ◽  
Support Vector ◽  
Least Squares Regression ◽  
Available Nitrogen ◽  
Phosphorus And Potassium ◽  
Destructive Measurement

The aim of this study was to investigate the potential of the infrared spectroscopy technique for non-destructive measurement of soil properties. For the study, 280 soil samples were collected from several regions in Zhejiang, China. Data from near infrared (NIR, 800–2500 nm), mid infrared (MIR, 4000–400 cm–1), and the combined NIR–MIR regions were compared to determine which produced the best prediction of soil properties. Least-squares support vector machines (LS-SVM) were applied to construct calibration models for soil properties such as available nitrogen (N), phosphorus (P), and potassium (K). The results showed that both spectral regions contained substantial information on N, P, and K in the soils studied, and the combined NIR–MIR region did a little worse than either the NIR or MIR region. Optimal results were obtained through LS-SVM compared with the standard partial least-squares regression method, and the correlation coefficient of prediction (rp), root mean square error for prediction, and bias were, respectively, 0.90, 16.28 mg/kg, and 0.96 mg/kg for the prediction results of N in the NIR region; and 0.88, 41.62 mg/kg, and –2.28 mg/kg for the prediction results of P, and 0.89, 33.47 mg/kg, and 2.96 mg/kg for the prediction results of K, both in the MIR region. This work demonstrated the potential of LS-SVM coupled to infrared reflectance spectroscopy for more efficient soil analysis and the acquisition of soil information.

scholarly journals Deep Learning and Conventional Machine Learning for Image-Based in-Situ Fault Detection During Laser Welding: A Comparative Study

2021 ◽  
Author(s):  
Christian Knaak ◽  
Moritz Kröger ◽  
Frederic Schulze ◽  
Peter Abels ◽  
Arnold Gillner
Keyword(s):  
Machine Learning ◽  
Deep Learning ◽  
Near Infrared ◽  
Learning Algorithms ◽  
Machine Learning Algorithms ◽  
Supervised Machine Learning ◽  
Support Vector ◽  
Detection Rates ◽  
Feature Extraction And Selection ◽  
Welding Defects

An effective process monitoring strategy is a requirement for meeting the challenges posed by increasingly complex products and manufacturing processes. To address these needs, this study investigates a comprehensive scheme based on classical machine learning methods, deep learning algorithms, and feature extraction and selection techniques. In a first step, a novel deep learning architecture based on convolutional neural networks (CNN) and gated recurrent units (GRU) is introduced to predict the local weld quality based on mid-wave infrared (MWIR) and near-infrared (NIR) image data. The developed technology is used to discover critical welding defects including lack of fusion (false friends), sagging and lack of penetration, and geometric deviations of the weld seam. Additional work is conducted to investigate the significance of various geometrical, statistical, and spatio-temporal features extracted from the keyhole and weld pool regions. Furthermore, the performance of the proposed deep learning architecture is compared to that of classical supervised machine learning algorithms, such as multi-layer perceptron (MLP), logistic regression (LogReg), support vector machines (SVM), decision trees (DT), random forest (RF) and k-Nearest Neighbors (kNN). Optimal hyperparameters for each algorithm are determined by an extensive grid search. Ultimately, the three best classification models are combined into an ensemble classifier that yields the highest detection rates and achieves the most robust estimation of welding defects among all classifiers studied, which is validated on previously unknown welding trials.

NIMG-35. MACHINE LEARNING GLIOMA GRADE PREDICTION LITERATURE: A TRIPOD ANALYSIS OF REPORTING QUALITY

2021 ◽  
Vol 23 (Supplement_6) ◽  
pp. vi136-vi136
Author(s):  
Sara Merkaj ◽  
Ryan Bahar ◽  
W R Brim ◽  
Harry Subramanian ◽  
Tal Zeevi ◽  
...  
Keyword(s):  
Machine Learning ◽  
Sample Size ◽  
Model Development ◽  
Model Performance ◽  
Reporting Quality ◽  
Controlled Vocabulary ◽  
Model Specification ◽  
Adherence Rate ◽  
Quality Of Reporting ◽  
Grade Prediction

Abstract PURPOSE Reporting guidelines are crucial in model development studies to ensure the quality, transparency and objectivity of reporting. While machine learning (ML) models have proven themselves effective in predicting glioma grade, their potential use can only be determined if they are clearly and comprehensively reported. Reporting quality has not yet been evaluated for ML glioma grade prediction studies, to our knowledge. We measured published literature against the TRIPOD Statement, a checklist of items considered essential for the reporting of diagnostic studies. MATERIALS AND METHODS A literature review, in agreement with PRISMA, was conducted by a university librarian in October 2020 and verified by a second librarian in February 2021 using four databases: Cochrane trials (CENTRAL), Ovid Embase, Ovid MEDLINE, and Web of Science core-collection. Keywords and controlled vocabulary included artificial intelligence, machine learning, deep learning, radiomics, magnetic resonance imaging, glioma, and glioblastoma. Publications were screened in Covidence and scored against the 27 items in the TRIPOD Statement that were relevant and applicable. RESULTS The search identified 11,727 candidate articles with 1,135 articles undergoing full text review. 86 articles met the criteria for our study. The mean adherence rate to TRIPOD was 44.4% (range: 22.2% - 66.7%), with poor reporting adherence in categories including abstract (0%), model performance (0%), title (1.2%), justification of sample size (2.3%), full model specification (2.3%), participant demographics and missing data (7%). Studies had high reporting adherence in categories including results interpretation (100%), background (98.8%), study design/source of data (96.5%), and objectives (95.3%). CONCLUSION Existing publications on the use of ML in glioma grade prediction have a low overall quality of reporting. Improvements can be made in the reporting of titles and abstracts, justification of sample size, and model specification and performance.

Sign in / Sign up

Export Citation Format

Share Document