Average Coefficient Of Determination Research Articles

Explainable highway performance degradation prediction model based on LSTM

With the dense and huge highway network in China, the highway maintenance management system has become the most concerned issue for Chinese highway managers in recent years. Therefore, based on the highway network in Guizhou Province of China, this paper established the semi-rigid asphalt pavement performance multi-output Long Short-Term Memory (LSTM) prediction model with regional applicability, including Pavement Surface Condition Index (PCI), Pavement Rutting Depth Index (RDI), Pavement Riding Quality Index (RQI) and Pavement Skidding Resistance Index (SRI) and the hyperparameters of the model were optimized by Bayesian optimization algorithm and the casual analysis of the model was conducted based on SHapley Additive exPlanations (SHAP), providing a reference for China's highway network-level preventive maintenance decision-making. The results prove that the established model can effectively predict the performance of the following year through the data of the previous two years. The established model has a better comprehensive prediction effect in comparison to similar multi-output models. The average coefficient of determination (R2) of the four prediction indexes can reach 0.823. Besides, the prediction effect among the indexes is stable and multiple pavement performance indexes can be effectively and reliably predicted at the same time. In addition, the contribution of input feature variables to output is quantified based on SHAP. According to quantified the contribution, the association relationship of the neural network model is set-valued mapped to the causality relationship in the post hoc. The input feature variables with the main contribution are extracted for causal analysis. The causal analysis shows that the degradation of the four pavement performances all have obvious time memory characteristics and have significant differences, but their pavement performance degradation is the result of nonlinear changes caused by the coupling effects of traffic load, road age, climate, and pavement structure.

Representative Sample Size for Estimating Saturated Hydraulic Conductivity via Machine Learning: A Proof‐Of‐Concept Study

AbstractMachine learning (ML) has been extensively applied in various disciplines. However, not much attention has been paid to data heterogeneity in databases and number of samples used to train ML models in hydrology. In this study, we addressed these issues and their impacts on the accuracy and reliability of ML models in the estimation of saturated hydraulic conductivity, Ks. We selected 17,990 soil samples from the USKSAT database and created random subsets N = 2,000, 4,000, 6,000, 8,000, 10,000, 12,000, 14,000, 16,000, and 17,990, 80% of which were used for training. The random subset selection was repeated 50 times. The extreme gradient boosting (XGBoost) algorithm was used to estimate Ks from other soil properties, such as bulk density, soil depth, texture, and organic content. For each subset, we conducted the learning curve analysis on the training and cross‐validation data sets. Results showed that for all training sample sizes the number of samples was not enough for the training and cross‐validation curves to reach a plateau. We also applied the concept of representative elementary volume by plotting the average coefficient of determination, R2, and root mean square log‐transformed error, RMSLE, against the training sample size. For the testing data set, as the number of training sample size increased from 1,600 to 14,392 the average R2 value increased from 0.74 to 0.90, while the average RMSLE value decreased from 1.08 to 0.69. Either the learning curve or representative sample size analysis is required to investigate whether the number of samples is enough or not.

Reconstruction of Baltic Gridded Sea Levels from Tide Gauge and Altimetry Observations Using Spatiotemporal Statistics from Reanalysis

A method for reconstructing weekly Baltic gridded sea levels was developed and tested. This method uses input data from tide gauge and altimetry observations. The reconstruction is based on sea level empirical orthogonal function (EOF) modes, calculated as spatiotemporal statistics from daily model reanalysis results available from the Copernicus Marine Service for the 1993–2021 period. In the semi-enclosed, tideless Baltic Sea, the four leading EOF modes cover 99% of the sea level variance. Several experiments with different combinations of input data were carried out. This method was validated against coastal tide gauges and altimetry data. The best reconstruction was obtained when both the tide gauges and altimetry data were used as inputs. An assessment of the centered root-mean-square difference (cRMSD) of the reconstruction versus the tide gauges revealed a value of 0.05 m, and a result of 0.10 m was revealed versus altimetry. The average coefficient of determination (R2) was determined to be 0.93 for the tide gauges and 0.82 for the altimetry data. In the cases where only one type of input data was used, the reconstruction worsened with respect to other data sources. The reconstruction method demonstrated its usefulness for the reconstruction of coastal sea levels in unsampled locations and the calculation of changes in sea volume.

Proton dose deposition matrix prediction using multi-source feature driven deep learning approach

Proton dose deposition results are influenced by various factors, such as irradiation angle, beamlet energy and other parameters. The calculation of the proton dose deposition matrix (DDM) can be highly complex but is crucial in intensity-modulated proton therapy (IMPT). In this work, we present a novel deep learning (DL) approach using multi-source features for proton DDM prediction. The DL5 proton DDM prediction method involves five input features containing beamlet geometry, dosimetry and treatment machine information like patient CT data, beamlet energy, distance from voxel to beamlet axis, distance from voxel to body surface, and pencil beam (PB) dose. The dose calculated by Monte Carlo (MC) method was used as the ground truth dose label. A total of 40 000 features, corresponding to 8000 beamlets, were obtained from head patient datasets and used for the training data. Additionally, seventeen head patients not included in the training process were utilized as testing cases. The DL5 method demonstrates high proton beamlet dose prediction accuracy, with an average determination coefficient R 2 of 0.93 when compared to the MC dose. Accurate beamlet dose estimation can be achieved in as little as 1.5 milliseconds for an individual proton beamlet. For IMPT plan dose comparisons to the dose calculated by the MC method, the DL5 method exhibited gamma pass rates of γ(2 mm, 2%) and γ(3 mm, 3%) ranging from 98.15% to 99.89% and 98.80% to 99.98%, respectively, across all 17 testing cases. On average, the DL5 method increased the gamma pass rates to γ(2 mm, 2%) from 82.97% to 99.23% and to γ(3 mm, 3%) from 85.27% to 99.75% when compared with the PB method. The proposed DL5 model enables rapid and precise dose calculation in IMPT plan, which has the potential to significantly enhance the efficiency and quality of proton radiation therapy.

Quantitative Structure–Activity Relationship Models for the Angiotensin-Converting Enzyme Inhibitory Activities of Short-Chain Peptides of Goat Milk Using Quasi-SMILES

The inhibitory activity of peptides on angiotensin-converting enzyme (ACE) is a measure of their antihypertensive potential. Quantitative structure–activity relationship (QSAR) models obtained based on the analysis of sequences of amino acids are suggested. The average determination coefficient for the active training sets is 0.36 ± 0.07. The average determination coefficient for validation sets is 0.79 ± 0.02. The paradoxical situation is caused by applying the vector of ideality of correlation, which improves the statistical quality of a model for the calibration and validation sets but is detrimental to the statistical quality of models for the training sets.

Assessment of post-pandemic economic recovery in Beijing, China, using night-time light data

ABSTRACT The coronavirus disease 2019 (COVID-19) pandemic posed a significant impact on human society, resulting in substantial changes in night-time light (NTL). This study developed a novel NTL index to estimate the quarterly Gross Domestic Product (GDP) at the county level in Beijing. This study aimed to investigate the economic recovery status in Beijing by establishing regression models of economic development before and after the pandemic. The estimated quarterly GDP at the county level was found to have a strong correlation with actual quarterly GDP statistics, with an average coefficient of determination (R 2) of 0.790. The regression models of the economic development of Beijing county-level districts before and after the pandemic were established after seasonal decomposition of the quarterly GDP. Based on the economic recovery framework for the pandemic in Beijing, most districts in the city had exhibited low economic resilience in terms of resistance and recoverability before and after the pandemic. Although the economy of Beijing is recovering gradually, it has not yet restored to pre-pandemic normal levels. These findings imply that NTL data can help assess the impact of emergency events on the disturbance of economic activity.

Comparative evaluation of four actual evapotranspiration models over different ecosystems and climate zones in China

ABSTRACT To better understand the discrepancies in evapotranspiration (ET) simulations between ET models, we intercompared four models in China: Priestley–Taylor Jet Propulsion Laboratory (PT-JPL), Penman–Montieth–Leuning Version 2 (PML-V2), Sigmoid Generalized Complementary Function (SGCF), Mapping Evapotranspiration at High Resolution with Internalized Calibration (METRIC). Data from 18 flux sites were used to evaluate the model performance at daytime (when incident shortwave radiation is greater than 20 W/m2) scales. To compare more fairly, we took the intersection of the outputs from four models for the analyses in the main text. All models yielded acceptable results, with PML-V2 or SGCF performing best at most sites. The average coefficient of determination and root mean square error among all sites of LE (latent heat of ET) were 0.72 and 51.71 W/m2 for PT-JPL, 0.80 and 46.65 W/m2 for PML-V2, 0.79 and 41.13 W/m2 for SGCF, 0.70 and 51.09 W/m2 for METRIC. PT-JPL and PML-V2 underestimated ET at most sites, whereas SGCF overestimated, potentially due to uncertainties in the vegetation indices and ET constraint parameters. Compared to measurements, PT-JPL underestimated the proportion of transpiration to evapotranspiration (0.81 versus 0.59), while PML-V2 overestimated (0.81 versus 0.90). Furthermore, all models performed best in the semi-arid zone dominated by grassland sites.

Modeling rutting depth on RIOHTrack asphalt pavement using Circle LSTMs

In response to the critical need for accurate pavement rutting prediction in pavement design, this study introduces a novel deep-learning model, Circle LSTM, designed for the estimation of rutting depths in asphalt pavements. Leveraging data from a full-scale field accelerated road testing facility in Tongzhou, Beijing, which spans 2.038 kilometres and features 19 distinct asphalt pavements, the Circle LSTM model utilizes variables such as temperature and load axis for its predictions. Methodologically, the Circle LSTM outperforms traditional machine learning approaches by cycling through different LSTM models to enhance short-term dependence modeling, showcasing its strength in addressing nonlinear problems inherent in road engineering. Evaluation of the model on the test dataset yielded promising results, with error metrics showing an average RMSE of 2.4170, an average MAE of 2.9454, an average MAPE of 2.61%, and an average coefficient of determination (average R2) of 0.8253 for 19 different asphalt pavements. These findings underscore the Circle LSTM model’s potential not only for fast and reliable rutting depth prediction but also for broader applications in multivariate time-series forecasting tasks in road engineering and beyond.

Decision tree-based approach to extrapolate life cycle inventory data of manufacturing processes

Life cycle assessment (LCA) plays a crucial role in green manufacturing to uncover the critical aspects for alleviating the environmental burdens due to manufacturing processes. However, the scarcity of life cycle inventory (LCI) data for the manufacturing processes is a considerable challenge. This paper proposes a novel approach to extrapolate LCI data of manufacturing processes. Taking advantage of LCI data in the Ecoinvent datasets, decision tree-based supervised machine learning models, namely decision tree, random forest, gradient boosting, and adaptive boosting, have been developed to extrapolate the data of GHG emissions, i.e., carbon dioxide, nitrous oxide, methane, and water vapor. Initially, a correlation analysis was conducted to derive the most influential factors on GHG quantities resulting from manufacturing activities. First, the collected data have been preprocessed and split into train and test sets (70% and 30%, respectively). Second, a five-fold cross-validation method was applied to tune the hyperparameters of the models. Then, the models were re-trained using the best hyperparameters and evaluated using the test set. The results reveal that the Gradient Boosting model has a superior predictive performance for extrapolating the GHG emission data, with average coefficients of determination (R2) on the test set <0.95. Moreover, the model predictions involve relatively low values of the average root mean squared error and an average mean percentage of error on the test set. The correlation and feature importance analyses emphasized that the workpiece material and manufacturing technology have a considerable effect on natural resource consumption, i.e., energy, material, and water inflows into the process. Meanwhile, energy consumption, water usage, and raw aluminum depletion were the most influential factors in GHG emissions. Eventually, a case study to extrapolate the inflows and the outflows for new manufacturing activities has been conducted using the validated models. The proposed GraBoost model provides a computational supplementary approach to estimate and extrapolate the GHG emissions for different manufacturing processes when LCI data are incomplete or don't exist within LCI databases.

Effect of different constraining boundary conditions on simulated femoral stresses and strains during gait

Finite element analysis (FEA) is commonly used in orthopaedic research to estimate localised tissue stresses and strains. A variety of boundary conditions have been proposed for isolated femur analysis, but it remains unclear how these assumed constraints influence FEA predictions of bone biomechanics. This study compared the femoral head deflection (FHD), stresses, and strains elicited under four commonly used boundary conditions (fixed knee, mid-shaft constraint, springs, and isostatic methods) and benchmarked these mechanics against the gold standard inertia relief method for normal and pathological femurs (extreme anteversion and retroversion, coxa vara, and coxa valga). Simulations were performed for the stance phase of walking with the applied femoral loading determined from patient-specific neuromusculoskeletal models. Due to unrealistic biomechanics observed for the commonly used boundary conditions, we propose a novel biomechanical constraint method to generate physiological femur biomechanics. The biomechanical method yielded FHD (< 1 mm), strains (approaching 1000 µε), and stresses (< 60 MPa), which were consistent with physiological observations and similar to predictions from the inertia relief method (average coefficient of determination = 0.97, average normalized root mean square error = 0.17). Our results highlight the superior performance of the biomechanical method compared to current methods of constraint for both healthy and pathological femurs.

Precision Irrigation Soil Moisture Mapper: A Thermal Inertia Approach to Estimating Volumetric Soil Water Content Using Unmanned Aerial Vehicles and Multispectral Imagery

To address the issue of estimating soil moisture at a hyper-resolution scale, a methodology referred to as Precision Irrigation Soil Moisture Mapper (PrISMM), that includes three key components, is developed: high-resolution remotely sensed optical and thermal data, surface energy balance modeling, and site-specific soil analysis. An Unmanned Aerial Vehicle/System (UAV or UAS) collects high-resolution multispectral imagery in the Dallas–Fort Worth metropolitan study area. Orthomosaics are converted to thermal inertia estimates in a spatially distributed format using the remotely sensed data combined with a set of surface energy balance modeling equations. Using thermal and physical properties of soil gained from site-specific soil analysis, thermal inertia estimates were further converted from thermal inertia to daily volumetric soil water content (VSWC) with a horizonal resolution of 8.6 cm. A ground truthing dataset of measured VSWC values taken from a Time Domain Reflectometer was compared with model results, producing a reasonable correlation with an average coefficient of determination of (R2) = 0.79, an average root mean square error (RMSE) = 0.0408, and mean absolute error (MAE) = 0.0308. This study highlights a practical approach of estimating VSWC for irrigation purposes while providing superior spatio-temporal coverage over in situ methods. The authors envision that PrISMM can be implemented in water usage management by relating VSWC with weather forecasts and evapotranspiration rates to develop time-based spatially distributed irrigation management plans.

Data-driven prediction of inter-layer process condition variations in laser powder bed fusion

The pressing problem of laser powder bed fusion of metals (PBF-LB/M) is the inability to predict and mitigate undesired process conditions in a build. This study aims to develop an efficient and geometric-agnostic prediction tool for the variations of the average meltpool temperature of each layer given a laser scanning path. The approach is a data-driven model based on defined physics-based features and measurement data from a two-color coaxial photodiode system. Although the coaxial photodiode system cannot measure absolute meltpool temperature, it can measure relative changes in the meltpool temperature if the process is performed in the shallow or no-keyhole regimes, the focus of this study. Once trained, the model can predict meltpool temperature variations before the start of the build process for a part geometry which had not yet been printed. The results show that variations in the layer average meltpool temperature can be predicted with the average coefficient of determination (r2) score of 0.65. Furthermore, the method shows the significance of considering the effect of surrounding parts on the average meltpool temperature profile of an observed part. For part geometries investigated in this study, the model shows the information from at least 40 previous layers is needed to enable sufficient and stabilized prediction performance.

Improving canopy transpiration model performance by considering concurrent hot and dry conditions

CONTEXTThe Jarvis-Stewart model has commonly been used to estimate transpiration. However, its assumption regarding independent effects from different influencing factors has seldom been explored. OBJECTIVEThe aim is to test whether the Jarvis-Stewart model can well estimate transpiration under concurrent hot and dry conditions with strong environmental interactions. A secondary objective was to test was to propose a new coupling method without the environmental-independence assumption for transpiration estimation. METHODSThis study tested and compared 24 Jarvis-Stewart models composed of various constraint functions to determine the optimal model structure for an orange orchard under concurrent hot and dry conditions. Meanwhile, a new coupling equation that considers interactive environmental effects on transpiration was proposed, and a total of 48 configurations were examined to determine the most effective configuration. A comprehensive comparison was made between the best Jarvis-Stewart model and the best coupling model based on a Bayesian framework. RESULTS AND CONCLUSIONSResults showed that with fewer parameters, the best new coupling model significantly improved model performance compared with the best Jarvis-Stewart model. Specifically, estimation accuracy was slightly improved, with the average mean relative error decreased from 11.79% to 11.06%, and the average coefficient of determination increased from 0.67 to 0.72. More importantly, estimation uncertainty was significantly decreased, with the average uncertainty band width reduction of 42%. Moreover, the new model generated some water use information that was much more consistent with measured data and did not suffer from over-fitting problems as the Jarvis-Stewart model suffered from. Furthermore, the new model did not require additional data or computational cost. SIGNIFICANCEThe results of our study emphasize the necessity for studying environmental interaction effects on plant water consumption, particularly under the possibility of concurrent hot and dry conditions that are likely to occur under future climate conditions.

An optimal ensemble of the CoLM for simulating the carbon and water fluxes over typical forests in China

Stomatal conductance (gs) and compensatory water uptake (CWU) are crucial processes in land surface models, as they directly influence the exchange of carbon and water fluxes between terrestrial ecosystems and the atmosphere. In this study, we integrated a new stomatal scheme derived from optimal stomatal theory (Medlyn's gs model), and an empirical CWU scheme into the Common Land Model (CoLM). Assessing the impacts on modeling gross primary productivity (GPP) and latent flux (LE) through observations obtained from eddy covariance (EC) measurements at three forest sites in China. Our results show that replacing the Ball-Berry's gs model (termed BB) with Medlyn's gs model (termed MED) did not bring about significant changes (had neutral impacts) in the performance of CoLM simulations at three forest sites. Considering the climate factors of annual mean precipitation to optimize key fitting parameters in gs exhibited improvement in model simulations. The average coefficient of determination (R2) achieved to 0.65 for GPP and LE at three sites, and the normalized root mean squared error (NRMSE) decreased from 0.83 to 0.77 at those sites. Besides, incorporating CWU into the model improved its performance. The R2 increased to 0.84 and RMSE decreased to 4.84 μmol m−2 s−1 for GPP, and the R2 increased to 0.62 and RMSE decreased to 55.64 W m−2 for LE. Therefore, modifying the model process of both contributed more to enhancing the model simulations than relying solely on one of these functions. Our study highlights that the response of plant functional types (PFTs) to water stress can be effectively represented in gs models when coupled with biochemical capacity to quantify carbon and water fluxes in forest ecosystems or other ecosystems.

Vibration response characteristics of Jujube trees based on finite element method and structure-from-motion

It is an indispensable step to choose optimal vibration parameters in jujube fruits vibration harvesting. Excessive vibration frequency will cause jujube tree damage, otherwise, the separation of fruit and fruit stalk cannot be achieved, resulting in harvesting failure. In our work, finite element method and vibration tests are applied to find optimal vibration parameters. To improve the accuracy of finite element method, a low-cost 3D reconstruction method is proposed to build jujube tree model. Experimental result showed that the phenotypic information of the reconstructed jujube tree and the real jujube tree have a high similarity, and their determination coefficient R2 reach to 0.984. Moreover, the natural frequencies of jujube tree are analyzed by modal analysis, transient analysis is applied to analyze vibration transfer characteristics. The simulation and experimental results show that the optimal vibration frequency range of jujube tree is 25∼30 Hz, the optimal vibration excitation location is on the main trunk and as high as possible close to the secondary branches, which can obtain maximum vibration response for optimum harvesting efficiency. The average determination coefficient is 0.66 between test and simulation results at different frequencies. This study can provide references for accurate 3D modeling of jujube trees, selection of appropriate vibration parameter range and the development of jujube harvesting machinery.

Interpretation of fatigue lifetime prediction by machine learning modeling in piston aluminum alloys under different manufacturing and loading conditions

Various input variables, including corrosion time, fretting force, stress, lubrication, heat-treating, and nano-particles, were evaluated by modeling of stress-controlled fatigue lifetimes in AlSi12CuNiMg aluminum alloy of the engine pistons with different machine learning (ML) techniques. Bending fatigue experiments were conducted through cyclic loading with zero mean stress, and then experimental data was predicted by five different ML-based models. Moreover, when the optimal ML prediction model was found, it was analyzed using the Shapley additive explanation (SHAP) values method. Results illustrated that extreme gradient boosting (XGBoost) had superior data for estimations, with average training coefficients of determination of at least 63% and 90%, respectively for fatigue lifetime and its logarithmic value. The SHAP values interpretation of the XGBoost model revealed that fretting force, stress, and corrosion time were the most significant inputs in estimating the logarithm values of fatigue lifetimes, respectively.

Aboveground Biomass Mapping and Analysis of Spatial Drivers in the Qinghai-Xizang Plateau Permafrost Zone: A Case Study of the Beilu River Basin.

Aboveground biomass (AGB) serves as a crucial measure of ecosystem productivity and carbon storage in alpine grasslands, playing a pivotal role in understanding the dynamics of the carbon cycle and the impacts of climate change on the Qinghai-Xizang Plateau. This study utilized Google Earth Engine to amalgamate Landsat 8 and Sentinel-2 satellite imagery and applied the Random Forest algorithm to estimate the spatial distribution of AGB in the alpine grasslands of the Beiliu River Basin in the Qinghai-Xizang Plateau permafrost zone during the 2022 growing season. Additionally, the geodetector technique was employed to identify the primary drivers of AGB distribution. The results indicated that the random forest model, which incorporated the normalized vegetation index (NDVI), the enhanced vegetation index (EVI), the soil-adjusted vegetation index (SAVI), and the normalized burn ratio index (NBR2), demonstrated robust performance in regards to AGB estimation, achieving an average coefficient of determination (R2) of 0.76 and a root mean square error (RMSE) of 70 g/m2. The average AGB for alpine meadows was determined to be 285 g/m2, while for alpine steppes, it was 204 g/m2, both surpassing the regional averages in the Qinghai-Xizang Plateau. The spatial pattern of AGB was primarily driven by grassland type and soil moisture, with q-values of 0.63 and 0.52, and the active layer thickness (ALT) also played a important role in AGB change, with a q-value of 0.38, demonstrating that the influences of ALT should not be neglected in regards to grassland change.

A Novel Fusion-Based Methodology for Drought Forecasting

Accurate drought forecasting is necessary for effective agricultural and water resource management and for early risk warning. Various machine learning models have been developed for drought forecasting. This work developed and tested a fusion-based ensemble model, namely, the stacking (ST) model, that integrates extreme gradient boosting (XGBoost), random forecast (RF), and light gradient boosting machine (LightGBM) for drought forecasting. Additionally, the ST model employs the SHapley Additive exPlanations (SHAP) algorithm to interpret the relationship between variables and forecasting results. Multi-source data that encompass meteorological, vegetation, anthropogenic, landcover, climate teleconnection patterns, and topological characteristics were incorporated in the proposed ST model. The ST model forecasts the one-month lead standardized precipitation evapotranspiration index (SPEI) at a 12 month scale. The proposed ST model was applied and tested in the German federal states of Brandenburg and Berlin. The results show that the ST model outperformed the reference persistence model, XGBboost, RF, and LightGBM, achieving an average coefficient of determination (R2) value of 0.845 in each month in 2018. The spatiotemporal Moran’s I method indicates that the ST model captures non-stationarity in modeling the statistical association between predictors and the meteorological drought index and outperforms the other three models (i.e., XGBoost, RF, and LightGBM). Global sensitivity analysis indicates that the ST model is influenced by a combination of environmental variables, with the most sensitive being the preceding drought indices. The accuracy and versatility of the ST model indicate that this is a promising approach for forecasting drought and other environmental phenomena.

Spatial distribution model of thermal load for high-speed dry hobbing tool based on probability density function

High-speed dry hobbing is a clean and environmental-friendly gear machining process, but thermal dissipation inefficiency accelerates tool wear and shorten tool life which is difficult to be effectively inhibited due to its complex distribution characteristics. This paper proposes a thermal load distribution model based on probability density function, describing the spatial distribution of thermal load on tool and providing theoretical support for predicting the distribution of thermal load. Firstly, conducting a finite element simulation analysis based on a mathematical model of high-speed dry gear hobbing process; Secondly, the influence of various parameters on the constructed model is analyzed, and the Bayesian optimization algorithm is utilized to optimize the model parameters. Finally, the model is validated at the micro and macro level. The results show that the average coefficient of determination of the model is greater than 0.86 in micro level. And the average fitting error is less than 6 °C in macro level, besides, the average relative error is less than 15%. The description of the uneven thermal distribution on tool is obtained with probability density function, it provides a theoretical support for the prediction and further optimization of thermal load on tool under multiple parameters in the future.

Detecting Multiple Mixed Bacteria Using Dual-Mode Hyperspectral Imaging and Deep Neural Networks

Identifying and analyzing mixed pathogenic bacteria is important for clinical diagnosis and antibiotic therapy of multiple bacterial infection. In this paper, a dual-mode hyperspectral microscopic detection technology with hybrid deep neural networks (DNNs) was proposed for simultaneous quantitative analysis of four kinds of pathogenic bacteria in mixed samples. To acquire both transmission and fluorescence spectra regarding the mixed pathogens, we developed a dual-mode hyperspectral detection system with fine spectral resolution and wide wavelength range, which can also generate spatial images that can be used to calculate the total amount of mixed bacteria. The dual-mode spectra were regarded as mixed proportion characteristics and the input of the neural network for predicting the proportion of each bacterium present in the mixture. To better analyze the dual-mode spectral data, we customized a mixed bacteria measurement network (MB-Net) with hybrid DNNs architectures based on spectral feature fusion. Using the fusion strategy, two DNNs frameworks applied for transmission/fluorescence spectral feature processing were stacked to form the MB-Net that processes these features simultaneously, and the achieved average coefficient of determination (R2) and RMSE of validation set are 0.96 and 0.03, respectively. To the best of our knowledge, it is the first time of simultaneously detecting four types of mixed pathogenic bacteria using spectral detection technology, showing excellent potential in clinical practice.