Convolutional Neural Network Regression Model Research Articles

Rapid Detection of Protein and Starch Content in Brewing Wheat Using Hyperspectral Imaging Technology Combined with a Convolutional Neural Network Regression Model

Wheat is the main raw material in liquor brewing. However, the protein content (PC) and starch content (SC) in wheat will affect the quality and flavor of the final liquor. In this study, the rapid, non-destructive determination of the PC and SC in wheat was achieved by combining hyperspectral imaging (HSI) with a convolutional neural network regression (CNNR) model established using the original spectral data in the hyperspectral images. The best preprocessing method was first determined, and then the performance of CNNR, extreme gradient boosting (XGBoost) and partial least squares regression (PLSR) models were compared at full wavelength, and it was concluded that CNNR based on the original spectrum was the optimal model for predicting wheat PC (R-square (R2) = 0.9942, root mean square error (RMSE) = 0.1041, relative percentage difference (RPD) = 13.1306) and SC (R2 = 0.9329, RMSE = 0.8633, RPD = 3.8605) at full wavelength. Finally, to further explore the feature extraction capability of the CNN model, different feature selection and extraction methods are used to build the PLSR model, and the comparison revealed that the PLSR model built by feature extraction using convolutional neural network (CNN_F) performed best in predicting wheat PC and SC. These results showed that HSI combined with a CNNR model could enable the rapid and accurate analysis of the PC and SC in wheat, since the CNNR can extract features from samples better than the traditional machine learning models.

Convolutional neural network regression for low-cost microalgal density estimation

Density of microalgae is critical information for production of algae in a closed cultivation system since it can be used to optimally control their growth rate, biomass concentration and quality of the products. Given advancement in image processing techniques and thanks to low-cost camera sensors, image based methods are increasingly widely utilized to indirectly estimate the density. Advantages of the image based techniques include being less invasive and more nondestructive and biosecured. However, most of the existing techniques rely on averaging all pixels of a microalgae image, which may eliminate crucial information of their spatial correlation. Therefore, in this work we propose to exploit a convolutional neural network (CNN) to efficiently extract information from the microalgae images, which are then employed to regress the density. Interestingly, the proposed deep CNN regression model accepts the whole color image as its input while the density is calculated in the output. The proposed CNN regression architecture was validated in real-world experiments where the microalgal strain Chlorella vulgaris was cultured and their images were captured by our low-cost camera sensor system. The obtained results demonstrate that the averaged estimation accuracy of the proposed model is 99.45% (± 0.68%) while the R2 value between the density predictions and the ground truths is 0.9997, which is highly accurate and practical.

Improved CNN-based model for shape parameter determination of sand particles considering imbalanced dataset

This study presents an artificial intelligence prediction model for determining the roundness and sphericity of sand particles using a convolutional neural network (CNN) regression model, which combines a CNN and a regression model. The dataset used for the model development consisted of 27,000 binary images of coarse grains with various shapes. To develop the prediction model, the validated algorithms, ResNet101 and ResNet152, were utilized and modified into a CNN regression model to predict the shape parameters. The performance of the developed model was evaluated using 3,000 randomly generated binary images, and the mean squared error and root mean squared error were calculated as evaluation metrics. The evaluation results indicated that the model effectively predicted the sphericity and roundness of sand particles, with a relatively low prediction performance observed for roundness.

Deciphering the effect of grain boundary characteristics on fracture toughness of silicon nitride ceramics through a CNN regression model

Fracture toughness of silicon nitride (Si3N4) ceramics with different grain boundary characteristics was determined from their microstructural images via a convolutional neural network (CNN) model. For this purpose, the CNN model was trained using Si3N4 ceramics with sintering additives of Y2O3 and MgO, and then the fracture toughness for those with various lanthanoid oxides and MgO was determined from the microstructural images through the trained model. Herein, the difference between the determined and measured values indicates the effect of the grain boundary characteristics because they little appear in the microstructures. The measured and AI-determined fracture toughness showed that the large ionic radius of Sm2O3 enhanced the fracture toughness of the Si3N4 ceramics sintered at 1850 °C for 24 h. The trained model successfully determined fracture toughness values that closely matched the measured ones from the microstructural images, even for samples with pores and different grain boundary characteristics, except for the Sm2O3 sample sintered for 24 h.

Magnetocaloric properties predicted by deep learning with compositional features for bulk metallic glasses

A convolutional neural network (CNN) is a widely recognized deep-learning model that performs regression or classification from domain knowledge using a kernel function that extracts specific features of the input data. This study demonstrates the prediction performance of the magnetic entropy changes (−ΔSmag) and transition temperatures (Tr) of bulk metallic glasses (BMGs) with magnetocaloric effects using a CNN regression model. Compositional descriptors were generated using the “Magpie” method and were used as input data. The prediction performance was evaluated using the determination coefficient (R2 score) and root mean square error (RMSE), resulting in R2 score of 0.920 and 0.917 and RMSE of 1.285 J/kgK and 45.619 K for −ΔSmag and Tr, respectively. In the case study, we further applied the CNN model to unseen datasets for extrapolated −ΔSmag and Tr prediction and compared the results with the experimental values, resulting in excellent agreement with the reported experimental results.

Estimating the distance to an object from static color stereo images using convolutional neural networks

In issues such as autonomy of movement, the vital information is the vehicle's distance from the obstacle. The research presented is devoted to the phenomenon of disparity in color stereo images. Their goal was to estimate the distance to the obstacle. A convolutional neural network regression model was used as a research method. The data set is a proprietary set of color stereo static images. Testing the learned CNN model gave excellent results. The analysis of the obtained results showed that the developed solution could be used in engineering solutions such as supporting vehicle control.

Multiplex SERS detection of polycyclic aromatic hydrocarbon (PAH) pollutants in water samples using gold nanostars and machine learning analysis.

Polycyclic aromatic hydrocarbons (PAHs) have attracted a lot of environmental concern because of their carcinogenic and mutagenic properties, and the fact they can easily contaminate natural resources such as drinking water and river water. This study presents a simple and sensitive point-of-care SERS detection of PAHs combined with machine learning algorithms to predict the PAH content more precisely and accurately in real-life samples such as drinking water and river water. We first synthesized multibranched sharp-spiked surfactant-free gold nanostars (GNSs) that can generate strong surface-enhanced Raman scattering (SERS) signals, which were further coated with cetyltrimethylammonium bromide (CTAB) for long-term stability of the GNSs as well as to trap PAHs. We utilized CTAB-capped GNSs for solution-based 'mix and detect' SERS sensing of various PAHs including pyrene (PY), nitro-pyrene (NP), anthracene (ANT), benzo[a]pyrene (BAP), and triphenylene (TP) spiked in drinking water and river water using a portable Raman module. Very low limits of detection (LOD) were achieved in the nanomolar range for the PAHs investigated. More importantly, the detected SERS signal was reproducible for over 90 days after synthesis. Furthermore, we analyzed the SERS data using artificial intelligence (AI) with machine learning algorithms based on the convolutional neural network (CNN) model in order to discriminate the PAHs in samples more precisely and accurately. Using a CNN classification model, we achieved a high prediction accuracy of 90% in the nanomolar detection range and an f1 score (harmonic mean of precision and recall) of 94%, and using a CNN regression model, achieved an RMSEconc = 1.07 × 10-1 μM. Overall, our SERS platform can be effectively and efficiently used for the accurate detection of PAHs in real-life samples, thus opening up a new, sensitive, selective, and practical approach for point-of-need SERS diagnosis of small molecules in complex practical environments.

Application of semi-supervised convolutional neural network regression model based on data augmentation and process spectral labeling in Raman predictive modeling of cell culture processes

Raman spectroscopy coupled to chemometric model-based technology have a great potential for real-time monitoring of critical process parameters in animal cell culture. The measurement of off-line reference values required for calibration of monitoring procedures is complex, time- and resource-intensive, whereas the acquisition of spectra is a seconds- or minutes-based process. Usually, less than 5% spectra is labeled with off-line data during calibration steps, thus a large amount of spectra is wasted. In this study, we proposed a Semi-Supervised Convolutional Neural Network Regression (SSCNNR) model framework for Raman model establishment based on data augmentation and process spectra labeling. Compared with the original CNNR model, the size of the calibration dataset was expanded from 132 samples to nearly 8500, and the constructed SSCNNR models obtained a significantly improved accuracy in predicting glucose, glutamine, asparagine, and ammonium, with RMSEP values decreased by 29.1%, 37.3%, 38.3% and 7%, respectively. Compared with traditional modeling approach of PLSR and SVR, the SSCNNR models were proved to accurately predict the trend of substance concentration during frequent feeding process. Based on this framework, potential efforts included improving model prediction accuracy by hyperparameters optimization and accommodating scenarios of culture change by transfer learning.

Finite Element-Based Machine Learning Model for Predicting the Mechanical Properties of Composite Hydrogels

In this study, a finite element (FE)-based machine learning model was developed to predict the mechanical properties of bioglass (BG)-collagen (COL) composite hydrogels. Based on the experimental observation of BG-COL composite hydrogels with scanning electron microscope, 2000 microstructural images with randomly distributed BG particles were created. The BG particles have diameters ranging from 0.5 µm to 1.5 µm and a volume fraction from 17% to 59%. FE simulations of tensile testing were performed for calculating the Young’s modulus and Poisson’s ratio of 2000 microstructures. The microstructural images and the calculated Young’s modulus and Poisson’s ratio by FE simulation were used for training and testing a convolutional neural network regression model. Results showed that the network developed in this work can effectively predict the mechanical properties of the composite hydrogels. The R-squared values were 95% and 83% for Young’s modulus and Poisson’s ratio, respectively. This work provides a surrogate model of finite element analysis to predict mechanical properties of BG-COL hydrogel using microstructure images, which could be further utilized for characterizing heterogeneous materials in big data-driven material designs.

Prediction of compressional wave velocity of cement-reinforced soil from core images using a convolutional neural network regression model

This study aims to predict the compressional wave velocity (Vp) from the photographic images of cylindrically cored cement-reinforced soils using a convolutional neural network (CNN) model. The experimentally measured Vp values and corresponding surficial core images were subjected to the CNN regression model based on a backbone network pre-trained by transfer learning. The model was retrained by fine-tuning and optimized with regularization strategies and data augmentation. The results showed that the trained network model reliably predicted Vp with reasonable performance of R2 = 0.78. Three-dimensional X-ray computed tomographic imaging of both overestimated and underestimated specimens revealed that surficial core images did not sufficiently reflect the internal structures. The slightly scattered prediction seemed attributed to the insufficient dataset size and invisible internal structure. Nevertheless, the proposed approach allowed not only estimating Vp at unmeasured spots in cores based on core images by fully leveraging artificial intelligence but also obtaining consecutive Vp profiles. Furthermore, this study established the hardly seen correlation between core image and Vp by the proposed CNN regression model and can be extended to estimation of other geophysical and geomechanical properties to construct a sufficient dataset for subsurface geostatistical modeling.

Augmented flame image soft sensor for combustion oxygen content prediction

Oxygen content is one of the most critical factors for high-efficiency combustion. Online measurement of oxygen content from flame images is important but still challenging. For construction of an oxygen content prediction model, most current feature extraction methods are not straightforward. Additionally, there are always sufficient data for common operating conditions in practice, while only limited data for other operating conditions. The data collection process for model training is costly and time-consuming. To tackle the problem, this work presents an augmented flame image soft sensor for automated combustion oxygen content prediction. A convolutional neural network (CNN) regression model is designed to predict the oxygen content directly from flame images, without a single feature extraction process. Moreover, a regression generative adversarial network with gradient penalty is proposed to generate flame images with oxygen content labels. It overcomes the imbalanced and insufficient data problem arising in the CNN regression model training. The proposed soft sensor is compared with several common regression methods for oxygen content prediction. Experimental results show that the proposed method can predict the combustion oxygen content with high accuracy from flame images although the original datasets are imbalanced.

DELIGHT: Deep Learning Identification of Galaxy Hosts of Transients using Multiresolution Images

We present DELIGHT, or Deep Learning Identification of Galaxy Hosts of Transients, a new algorithm designed to automatically and in real time identify the host galaxies of extragalactic transients. The proposed algorithm receives as input compact, multiresolution images centered at the position of a transient candidate and outputs two-dimensional offset vectors that connect the transient with the center of its predicted host. The multiresolution input consists of a set of images with the same number of pixels, but with progressively larger pixel sizes and fields of view. A sample of 16,791 galaxies visually identified by the Automatic Learning for the Rapid Classification of Events broker team was used to train a convolutional neural network regression model. We show that this method is able to correctly identify both relatively large (10″ < r < 60″) and small (r ≤ 10″) apparent size host galaxies using much less information (32 kB) than with a large, single-resolution image (920 kB). The proposed method has fewer catastrophic errors in recovering the position and is more complete and has less contamination (<0.86%) recovering the crossmatched redshift than other state-of-the-art methods. The more efficient representation provided by multiresolution input images could allow for the identification of transient host galaxies in real time, if adopted in alert streams from new generation of large -etendue telescopes such as the Vera C. Rubin Observatory.

Predicting Leaf Nitrogen Content in Cotton with UAV RGB Images

Rapid and accurate prediction of crop nitrogen content is of great significance for guiding precise fertilization. In this study, an unmanned aerial vehicle (UAV) digital camera was used to collect cotton canopy RGB images at 20 m height, and two cotton varieties and six nitrogen gradients were used to predict nitrogen content in the cotton canopy. After image-preprocessing, 46 hand features were extracted, and deep features were extracted by convolutional neural network (CNN). Partial least squares and Pearson were used for feature dimensionality reduction, respectively. Linear regression, support vector machine, and one-dimensional CNN regression models were constructed with manual features as input, and the deep features were used as inputs to construct a two-dimensional CNN regression model to achieve accurate prediction of cotton canopy nitrogen. It was verified that the manual feature and deep feature models constructed from UAV RGB images had good prediction effects. R2 = 0.80 and RMSE = 1.67 g kg−1 of the Xinluzao 45 optimal model, and R2 = 0.42 and RMSE = 3.13 g kg−1 of the Xinluzao 53 optimal model. The results show that the UAV RGB image and machine learning technology can be used to predict the nitrogen content of large-scale cotton, but due to insufficient data samples, the accuracy and stability of the prediction model still need to be improved.

Automated porosity estimation using CT-scans of extracted core data

Estimation of porosity at a millimeter scale would be an order of magnitude finer resolution than traditional logging techniques. This enables proper description of reservoirs with thin layers and fine scale heterogeneities. To achieve this, we propose an end-to-end convolutional neural network (CNN) regression model that automatically predicts continuous porosity at a millimeter scale resolution using two-dimensional whole core CT scan images. More specifically, a CNN regression model is trained to learn from routine core analysis (RCA) porosity measurements. To characterize the performance of such approach, we compare the performance of this model with two linear regression models trained to learn the relationship between the average attenuation and standard deviation of the same two-dimensional images and RCA porosity. Our investigations reveal that the linear models are outperformed by the CNN, indicating the capability of the CNN model in extracting textures that are important for porosity estimations. We compare the predicted porosity results against the total porosity logs calculated from the density log. The obtained results show that the predicted porosity values using the proposed CNN method are well correlated with the core plug measurements and the porosity log. More importantly, the proposed approach can provide accurate millimeter scale porosity estimations, while the total porosity log is averaged over an interval and thus do not show such fine scale variations. Thus, the proposed method can be employed to calibrate the porosity logs, thereby reducing the uncertainties associated with indirect calculations of the porosity from such logs.

Using deep learning to predict the East Asian summer monsoon

Accurate prediction of the East Asian summer monsoon (EASM) is beneficial to billions of people’s production and lives. Here, a convolutional neural network (CNN) and transfer learning are used to predict the EASM. The results of the constructed CNN regression model show that the prediction of the CNN regression model is highly consistent with the reanalysis dataset, with a correlation coefficient of 0.78, which is higher than that of each of the current state-of-the-art dynamic models. The heat map method indicates that the robust precursor signals in the CNN regression model agree well with previous theoretical studies and can provide the quantitative contribution of different signals for EASM prediction. The CNN regression model can predict the EASM one year ahead with a confidence level above 95%. The above method can not only improve the prediction of the EASM but also help to identify the involved physical predictors.

Blood Pressure Monitoring System Using a Two-Channel Ballistocardiogram and Convolutional Neural Networks.

Hypertension is a chronic disease that kills 7.6 million people worldwide annually. A continuous blood pressure monitoring system is required to accurately diagnose hypertension. Here, a chair-shaped ballistocardiogram (BCG)-based blood pressure estimation system was developed with no sensors attached to users. Two experimental sessions were conducted with 30 subjects. In the first session, two-channel BCG and blood pressure data were recorded for each subject. In the second session, the two-channel BCG and blood pressure data were recorded after running on a treadmill and then resting on the newly developed system. The empirical mode decomposition algorithm was used to remove noise in the two-channel BCG, and the instantaneous phase was calculated by applying a Hilbert transform to the first intrinsic mode functions. After training a convolutional neural network regression model that predicts the systolic and diastolic blood pressures (SBP and DBP) from the two-channel BCG phase, the results of the first session (rest) and second session (recovery) were compared. The results confirmed that the proposed model accurately estimates the rapidly rising blood pressure in the recovery state. Results from the rest sessions satisfied the Association for the Advancement of Medical Instrumentation (AAMI) international standards. The standard deviation of the SBP results in the recovery session exceeded 0.7.

The Effects of Accelerated Brain Aging on the Antidepressant Response in Late-Life Depression

Introduction Late-Life Depression (LLD) is characterized by functional and structural brain alterations and accelerated brain aging. Our past work utilized a brain-based marker of accelerated aging to examine LLD and demonstrated a cross-sectional association between accelerated brain aging and impaired cognitive ability (poor episodic memory, slow processing speed, and poor executive function) as well as disability (measured by WHODAS). However, it is unclear whether or not accelerated brain aging influences the antidepressant response in depressed elders. Methods We examined depressed adults over 60 years-old with current Major Depressive Disorder (MDD) without psychotic features (N=95) randomized to receive either escitalopram up to 20mg daily or identical placebo (2:1 subject ratio) for 8 weeks. Subjects underwent baseline depression assessment (MADRS) and MRI prior to randomization. After using an automated deep learning tool to estimate brain age from cranial MRI, the brain age gap (BAG) was calculated by subtracting the participant's chronological age from his or her calculated brain age estimation. The brain MRI estimating tool uses ROI volumes derived from multiple atlases and the BrainAGE algorithm - the algorithm applies a deep convolutional neural network regression model trained on 5000 healthy controls. For statistical analysis, the dependent outcome variable was change in MADRS over 8 weeks. Initial models examined the effects of BAG with baseline MADRS, age, sex, and treatment assignment as covariates. Subsequent models tested for a statistical interaction between treatment assignment and BAG to observe whether or not the effects of brain aging on change in MADRS differed between arms. Results In the primary effects model examining change in MADRS, although there was a strong effect of treatment assignment (t = -3.75, p = 0.0003), there was no significant relationship between BAG and antidepressant response (t = -0.02, p = 0.9819). In the subsequent model, we observed a statistical interaction between BAG and treatment assignment on change in MADRS (t = -2.38, p = 0.0196). Analysis of subgroup data demonstrated that a larger BAG (indicating an older appearing brain) was associated with less change in MADRS for those who received placebo. In contrast, a larger BAG (indicating an older appearing brain) was associated with greater change in MADRS for those who received escitalopram. Conclusions The effect of accelerated brain aging, measured with the BAG, on change in MADRS have opposite patterns based on treatment assignment. Although a larger BAG is associated with poorer cognitive performance and greater disability, a larger BAG was associated with greater change in MADRS in the active treatment arm of our study. This is somewhat discrepant with prior work that associates poor cognitive performance with worse antidepressant response. As a global brain metric, BAG is not specific to discrete functional and/or structural brain impairments. Therefore, the elements that jointly contribute to the brain age metric may act in discrete, separate ways, with some influencing cognitive performance, and with others having less of a functional impact, serving as a marker of positive response to antidepressant medication. Further work must examine the complex relationships between accelerated brain aging, cognition, and response to antidepressants. Funding R01 MH102246, K24 MH110598

Determining Spatiotemporal Distribution of Macronutrients in a Cornfield Using Remote Sensing and a Deep Learning Model

Fertilizer misapplications have induced widespread environmental deteriorations, climatic catastrophes, and economic losses; meanwhile, the Precision Agriculture (PA) endorsements have been influential in alleviating these issues. This study intended to tackle the fertilizer consumption inefficiencies by utilizing non-destructive remote sensing technologies, soil macronutrient distribution analysis, and a deep learning model. Specifically, an Unmanned Air Vehicle (UAV) was used in a cornfield to capture the plant’s reflectance information for retrieving the Normalized Difference Vegetation Index (NDVI) during the vegetative and reproductive growth stages. Consequently, the field’s soil samples were examined for their Nitrogen, Phosphorus, Potassium, and Carbon (NPKC) macronutrient constituencies. Finally, a Convolutional Neural Network-Regression model was developed to predict infield NPKC spatiotemporal variations in soil using the NDVI measurements. The deep learning model effectively determined the surpluses or shortages of the NPKC macronutrients within the cornfield throughout the growth stages. The model performed vigorously with R2 values of 0.93, 0.92, 0.98, and 0.83 in predicting N, P, K, and C levels in soil, respectively.

Transfer Learning for Nonrigid 2D/3D Cardiovascular Images Registration.

Cardiovascular image registration is an essential approach to combine the advantages of preoperative 3D computed tomography angiograph (CTA) images and intraoperative 2D X-ray/digital subtraction angiography (DSA) images together in minimally invasive vascular interventional surgery (MIVI). Recent studies have shown that convolutional neural network (CNN) regression model can be used to register these two modality vascular images with fast speed and satisfactory accuracy. However, CNN regression model trained by tens of thousands of images of one patient is often unable to be applied to another patient due to the large difference and deformation of vascular structure in different patients. To overcome this challenge, we evaluate the ability of transfer learning (TL) for the registration of 2D/3D deformable cardiovascular images. Frozen weights in the convolutional layers were optimized to find the best common feature extractors for TL. After TL, the training data set size was reduced to 200 for a randomly selected patient to get accurate registration results. We compared the effectiveness of our proposed nonrigid registration model after TL with not only that without TL but also some traditional intensity-based methods to evaluate that our nonrigid model after TL performs better on deformable cardiovascular image registration.

Deep learning system for Meyerding classification and segmental motion measurement in diagnosis of lumbar spondylolisthesis

Lumbar spondylolisthesis, which is a common disorders caused by the extent of slipping of the spondylolysis vertebra and making the spinal posture unstable, mostly provides no painful symptoms and, consequently, can only be diagnosed by using X-rays. Manual methods currently used for measuring the slipping of the vertebra and segmental motions in X-rays, is not only considered as a high-levelled pre-anatomy process requiring high experienced surgeons to ensure the measurement accuracy but also practically ineffective for in handling a large number of X-ray images. Therefore, this paper mainly concerns the development of a deep learning system with supported by the convolutional neural network (CNN), in which the supplementary CNN model is trained to re-correct the keypoints located on corners of vertebra based on the first CNN regression model, to precisely measure required characteristics. In order to determine the instability of the lumbar spondylolisthesis, the measurement ability of the proposed method is also required to adapt to multiple lateral bending views including flexion and extension postures. Finally, the performance obtained has been validated with comparison between standard references measured from an experienced surgeon and automatic measured values which appreciably performed precise results with the mean deviation of 1.76° within 0.12 s for treating a single X-ray image on the computing configuration utilized.