Neural Network Classification Model Research Articles

Accurate identification of low-resistivity gas layer in tight sandstone gas reservoirs based on optimizable neural networks

In tight sandstone reservoirs, low resistivity gas reservoirs have a certain gas production capacity, but their reservoir resistivity value is less than 20 Ω m, or their ratio to water layer resistivity is less than 2. Therefore, the identification of low resistivity gas reservoirs is a prominent problem that needs to be resolved. In this paper, on the basis of traditional experimental analysis, NMR experiments and wettability experiments were conducted. The high bound water saturation and pore fractal dimension have a significant impact on reservoir resistivity. After various experimental analysis results were preprocessed by the image pool and data pool, an optimized neural network (ONN) classification model was established based on deep learning theory. The confusion matrix result (training datasets) was 87.13%, and the ROC area of low resistivity gas layers was close to 1, indicating good recognition performance for low resistivity gas layers. The model was used to identify the properties of the J4 well reservoir section. The identification result was consistent with the actual logging resistivity values (low resistivity) and testing conclusions (gas layer). This model can accurately identify low resistivity gas reservoirs in the research area, and has important guiding significance for identifying the reservoir properties of tight sandstone gas reservoirs in the Sichuan Basin. It can provide reference value for the exploration and development of similar gas reservoirs in other regions.

Ensemble learning landslide susceptibility assessment with optimized non-landslide samples selection

Non-landslide samples influence the outcomes of landslide susceptibility assessment. Existing studies did not fully consider the equilibrium between landslide and non-landslide samples in similar environments, resulting in poor reliability of landslide susceptibility assessment. This study proposed a non-landslide samples optimization method with a constraint of disaster-pregnant environment similarity to construct the balanced landslide and non-landslide samples. We employed the heterogeneous stacking and blending ensemble learning models to generate the landslide susceptibility assessment. This study focused on the Bailong River Basin with complex environment and frequent landslides as the study area. First, we extracted 12 landslide influencing factors based on multiple sources and analyzed the spatial distribution patterns of landslides. Second, we constructed similar environments based on the assessment units obtained from the curvature watershed method and selected an equal amount of both landslide and non-landslide samples in every different environment. Finally, three classic neural network models, namely multilayer perceptron, convolutional neural network, and gated recurrent unit were used as base models for stacking and blending ensemble learning models to assess landslide susceptibility. The findings suggested that the landslide susceptibility assessment results with optimized non-landslide samples were more reliable, especially the proposed method improved the prediction results in sample-sparse regions. The stacking ensemble constructed in this study demonstrated the highest area under the curve of 0.88 for the testing dataset, outperforming the blending ensemble and three base models. The issue of unreliable landslide susceptibility assessment results in sample-sparse regions within complex environments can be effectively addressed by considering the balanced sampling of non-landslide samples under the constraint of similar environments.

Proton spot dose estimation based on positron activity distributions with neural network.

Positron emission tomography (PET) has been investigated for its ability to reconstruct proton-induced positron activity distributions in proton therapy. This technique holds potential for range verification in clinical practice. Recently, deep learning-based dose estimation from positron activity distributions shows promise for in vivo proton dose monitoring and guided proton therapy. This study evaluates the effectiveness of three classical neural network models, recurrent neural network (RNN), U-Net, and Transformer, for proton dose estimating. It also investigates the characteristics of these models, providing valuable insights for selecting the appropriate model in clinical practice. Proton dose calculations for spot beams were simulated using Geant4. Computed tomography (CT) images from four head cases were utilized, with three for training neural networks and the remaining one for testing. The neural networks were trained with one-dimensional (1D) positron activity distributions as inputs and generated 1D dose distributions as outputs. The impact of the number of training samples on the networks was examined, and their dose prediction performance in both homogeneous brain and heterogeneous nasopharynx sites was evaluated. Additionally, the effect of positron activity distribution uncertainty on dose prediction performance was investigated. To quantitatively evaluate the models, mean relative error (MRE) and absolute range error (ARE) were used as evaluation metrics. The U-Net exhibited a notable advantage in range verification with a smaller number of training samples, achieving approximately 75% of AREs below 0.5mm using only 500 training samples. The networks performed better in the homogeneous brain site compared to the heterogeneous nasopharyngeal site. In the homogeneous brain site, all networks exhibited small AREs, with approximately 90% of the AREs below 0.5mm. The Transformer exhibited the best overall dose distribution prediction, with approximately 92% of MREs below 3%. In the heterogeneous nasopharyngeal site, all networks demonstrated acceptable AREs, with approximately 88% of AREs below 3mm. The Transformer maintained the best overall dose distribution prediction, with approximately 85% of MREs below 5%. The performance of all three networks in dose prediction declined as the uncertainty of positron activity distribution increased, and the Transformer consistently outperformed the other networks in all cases. Both the U-Net and the Transformer have certain advantages in the proton dose estimation task. The U-Net proves well suited for range verification with a small training sample size, while the Transformer outperforms others at dose-guided proton therapy.

PreDBP-PLMs: Prediction of DNA-binding proteins based on pre-trained protein language models and convolutional neural networks

The recognition of DNA-binding proteins (DBPs) is the crucial step to understanding their roles in various biological processes such as genetic regulation, gene expression, cell cycle control, DNA repair, and replication within cells. However, conventional experimental methods for identifying DBPs are usually time-consuming and expensive. Therefore, there is an urgent need to develop rapid and efficient computational methods for the prediction of DBPs. In this study, we proposed a novel predictor named PreDBP-PLMs to further improve the identification accuracy of DBPs by fusing the pre-trained protein language model (PLM) ProtT5 embedding with evolutionary features as input to the classic convolutional neural network (CNN) model. Firstly, the ProtT5 embedding was combined with different evolutionary features derived from the position-specific scoring matrix (PSSM) to represent protein sequences. Then, the optimal feature combination was selected and input to the CNN classifier for the prediction of DBPs. Finally, the 5-fold cross-validation (CV), the leave-one-out CV (LOOCV), and the independent set test were adopted to examine the performance of PreDBP-PLMs on the benchmark datasets. Compared to the existing state-of-the-art predictors, PreDBP-PLMs exhibits an accuracy improvement of 0.5 % and 5.2 % on the PDB186 and PDB2272 datasets, respectively. It demonstrated that the proposed method could serve as a useful tool for the recognition of DBPs.

Deep learning pneumoconiosis staging and diagnosis system based on multi-stage joint approach

BackgroundPneumoconiosis has a significant impact on the quality of patient survival due to its difficult staging diagnosis and poor prognosis. This study aimed to develop a computer-aided diagnostic system for the screening and staging of pneumoconiosis based on a multi-stage joint deep learning approach using X-ray chest radiographs of pneumoconiosis patients.MethodsIn this study, a total of 498 medical chest radiographs were obtained from the Department of Radiology of West China Fourth Hospital. The dataset was randomly divided into a training set and a test set at a ratio of 4:1. Following histogram equalization for image enhancement, the images were segmented using the U-Net model, and staging was predicted using a convolutional neural network classification model. We first used Efficient-Net for multi-classification staging diagnosis, but the results showed that stage I/II of pneumoconiosis was difficult to diagnose. Therefore, based on clinical practice we continued to improve the model by using the Res-Net 34 Multi-stage joint method.ResultsOf the 498 cases collected, the classification model using the Efficient-Net achieved an accuracy of 83% with a Quadratic Weighted Kappa (QWK) score of 0.889. The classification model using the multi-stage joint approach of Res-Net 34 achieved an accuracy of 89% with an area under the curve (AUC) of 0.98 and a high QWK score of 0.94.ConclusionsIn this study, the diagnostic accuracy of pneumoconiosis staging was significantly improved by an innovative combined multi-stage approach, which provided a reference for clinical application and pneumoconiosis screening.

Rapid diagnosis of celiac disease based on plasma Raman spectroscopy combined with deep learning

Celiac Disease (CD) is a primary malabsorption syndrome resulting from the interplay of genetic, immune, and dietary factors. CD negatively impacts daily activities and may lead to conditions such as osteoporosis, malignancies in the small intestine, ulcerative jejunitis, and enteritis, ultimately causing severe malnutrition. Therefore, an effective and rapid differentiation between healthy individuals and those with celiac disease is crucial for early diagnosis and treatment. This study utilizes Raman spectroscopy combined with deep learning models to achieve a non-invasive, rapid, and accurate diagnostic method for celiac disease and healthy controls. A total of 59 plasma samples, comprising 29 celiac disease cases and 30 healthy controls, were collected for experimental purposes. Convolutional Neural Network (CNN), Multi-Scale Convolutional Neural Network (MCNN), Residual Network (ResNet), and Deep Residual Shrinkage Network (DRSN) classification models were employed. The accuracy rates for these models were found to be 86.67%, 90.76%, 86.67% and 95.00%, respectively. Comparative validation results revealed that the DRSN model exhibited the best performance, with an AUC value and accuracy of 97.60% and 95%, respectively. This confirms the superiority of Raman spectroscopy combined with deep learning in the diagnosis of celiac disease.

CVS-FLN: a novel IoT-IDS model based on metaheuristic feature selection and neural network classification model

CVS-FLN: a novel IoT-IDS model based on metaheuristic feature selection and neural network classification model

CoreViT: A new vision transformer model for lithofacies identification in cores

Lithofacies identification is crucial in fields such as geology, geotechnical engineering, rock mechanics, and petroleum engineering, as it reveals the physical, chemical, and mineralogical characteristics of rocks that aid in identifying oil and gas reservoir types, volumes, and productivity. Traditional methods of identifying rock types, such as manual sample identification, micrograph identification, and experimental identification, are costly and time-consuming. Several studies have endeavored to enhance the degree of lithofacies identification automation using machine learning (ML) and deep learning (DL) techniques. However, their capacity to recognize intricate data remains restricted. To overcome this challenge, we introduce CoreViT, a modified architecture of the advanced vision transformer (ViT) that includes the parallel transformer encoder (PTE) for exchanging information among image patches, and the class encoder (CE) for improved automatic lithofacies identification. Our study, focusing on core images including algal limestone, mudstone, and non-algal limestone from the Fengxi Well 1 in the Qaidam Basin, demonstrates CoreViT's average recognition accuracy at approximately 97.5% with an average loss of 0.071. Compared with other classical convolutional neural network (CNN) models, CoreViT achieves relatively high accuracy and low loss. This study highlights the superiority of the ViT model over traditional deep convolutional neural networks (DCNNs) and suggests the great potentiality of applying the ViT model in lithofacies identification in core samples.

On demand design of multi-freedom metamaterial absorbers based on recurrent prediction networks

Metamaterial absorbers (MMAs) have received a lot of attention due to their wide range of promising applications. In general, metamaterials involve a large number of geometric parameters, so the traditional simulation requires researchers to have rich experience and sufficient computational resources. In this paper, we demonstrate the forward and on demand design of multi-freedom MMAs based on the recurrent prediction network (RPN). We used the variable contribution analysis (VCA) algorithm to eliminate 55 absorption points whose contribution score was less than 1.0, the maximum fitting coefficient (R2) was increased by 7.0%, and the average calculation time was reduced by 20.326s. multi-freedom design is achieved by quantifying the design material into binary features as geometric parameter inputs into the RPN model. We solve the prediction peak error problem by many-to-one RPN, compared with the many-to-many prediction of classical deep neural network (DNN) model, the RPN error range is 0-0.12 compared to the 0-0.25 error range of the DNN. The ultra-wideband MMA design with the absorption of 0.2-4.2µm broadband is more than 90%, and the highest absorption rate is 99.2%. This method can be used in zoom imaging, metamaterials filters and other fields.

Two-stage fusion framework driven by domain knowledge for penetration prediction of laser welding

To solve the problem of industrial application difficulties caused by insufficient depth exploration and poor interpretability of the pure data drive neural network. A two-stage fusion framework driven by domain knowledge was proposed and focused on industrial applications: laser welding penetration monitoring. First, a multi-sensor acquisition system based on plasma and molten pool signals is developed, extracting multiple features from plasma and molten pool images. Two Back Propagation (BP) neural network information fusion models are established in the first fusion stage, leveraging the characteristics of the plasma plume and molten pool front. In the second fusion stage, the Dempster-Shafer (DS) evidence theory is employed to fuse the information from the two BP neural network models. The proposed method, achieving a high accuracy of 98.7 %, is validated by comparing its average accuracy and anti-interference capability with classical Convolutional Neural Network (CNN) and BP neural network models. The physical relationship between input and output is elucidated through an established physical model, enhancing the interpretability and anti-interference capability of the model, and thereby improving the stability and accuracy of the system.

Research of 2D-COS with metabolomics modifications through deep learning for traceability of wine

To tackle the difficulty of extracting features from one-dimensional spectral signals using traditional spectral analysis, a metabolomics analysis method is proposed to locate two-dimensional correlated spectral feature bands and combine it with deep learning classification for wine origin traceability. Metabolomics analysis was performed on 180 wine samples from 6 different wine regions using UPLC-Q-TOF-MS. Indole, Sulfacetamide, and caffeine were selected as the main differential components. By analyzing the molecular structure of these components and referring to the main functional groups on the infrared spectrum, characteristic band regions with wavelengths in the range of 1000–1400 nm and 1500–1800 nm were selected. Draw two-dimensional correlation spectra (2D-COS) separately, generate synchronous correlation spectra and asynchronous correlation spectra, establish convolutional neural network (CNN) classification models, and achieve the purpose of wine origin traceability. The experimental results demonstrate that combining two segments of two-dimensional characteristic spectra determined by metabolomics screening with convolutional neural networks yields optimal classification results. This validates the effectiveness of using metabolomics screening to determine spectral feature regions in tracing wine origin. This approach effectively removes irrelevant variables while retaining crucial chemical information, enhancing spectral resolution. This integrated approach strengthens the classification model's understanding of samples, significantly increasing accuracy.

A Study on Handwritten Text Recognition Classification using Diverse Deep Learning Techniques and Computation of CTC Loss

Handwriting recognition encompasses the conversion of hand-written text images into digital text, where input images yield predicted textual output. Optical Character Recognition (OCR) technology has conventionally fulfilled this role. With surging mobile phone usage, leveraging text detection via mobile cameras gains significance in fields like medical script processing and exam evaluation. To enhance image quality, noise reduction techniques like binarization and thresholding are applied. Image processing entails letter segmentation and extraction. In this paper, we propose a neural network classifier model amalgamating Convolution Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks. Leveraging an existing feature dataset, the model is trained. Input images are evaluated through the neural network, yielding recognized words in a text document format. Ultimately, the predicted output is derived via Connectionist Temporal Classification (CTC)- based loss computation. Keywords: Handwritten Recognition, Deep Learning Techniques, Optical Character Recognition.

ВИКОРИСТАННЯ ШТУЧНОГО ІНТЕЛЕКТУ ДЛЯ РОЗПІЗНАВАННЯ ТЕРОРИСТИЧНИХ ТА ВОРОЖИХ ВІЙСЬКОВИХ ОБ’ЄКТІВ

The development of artificial intelligence technologies and their application in solving military tasks is the main concept of scientific advancement in Ukraine. Despite the availability of existing solutions for classifying limited types of civilian objects, there is a need to develop new technologies to gain an advantage over the enemy. Existing artificial intelligence models available in the public domain are oriented towards diagnosing restricted types of civilian objects. This limitation prevents the use of these models for analyzing military enemy equipment. Automating the process of recognizing enemy military objects by constructing a deep learning classification model. The process of building a classification model for enemy military equipment using deep neural networks is considered. The Sequential model type is chosen as the basis, which includes batch normalization layers added after each convolutional layer before further processing. Accelerated hyperparameter search is provided by the Random Search tool. The quality of classification models was measured using classical accuracy metrics. The implementation of the proposed solutions is carried out using the Python programming language and the Python Imaging Library, TensorFlow, Keras Tuner, and Matplotlib libraries. A convolutional neural network classification model has been constructed capable of classifying enemy military tanks and vehicles. The model demonstrates the best accuracy metrics on training/testing datasets of 0.6/0.55 with errors of 0.658/0.705, respectively. A comparative analysis of classification models with different types of architectures showed the advantage of the model in terms of accuracy and error rate when using an architecture that includes batch normalization layers. The created classification model is recommended for further training on new data or for solving computer vision tasks.

Accelerating the Screening of Modified MA2Z4 Catalysts for Hydrogen Evolution Reaction by Deep Learning‐Based Local Geometric Analysis

Machine learning (ML) integrated with density functional theory (DFT) calculations have recently been used to accelerate the design and discovery of single‐atom catalysts (SACs) by establishing deep structure–activity relationships. The traditional ML models are always difficult to identify the structural differences among the single‐atom systems with different modification methods, leading to the limitation of the potential application range. Aiming to the structural properties of several typical two‐dimensional MA2Z4‐based single‐atom systems (bare MA2Z4 and metal single‐atom doped/supported MA2Z4), an improved crystal graph convolutional neural network (CGCNN) classification model was employed, instead of the traditional machine learning regression model, to address the challenge of incompatibility in the studied systems. The CGCNN model was optimized using crystal graph representation in which the geometric configuration was divided into active layer, surface layer, and bulk layer (ASB‐GCNN). Through ML and DFT calculations, five potential single‐atom hydrogen evolution reaction (HER) catalysts were screened from chemical space of 600 MA2Z4‐based materials, especially V1/HfSn2N4(S) with high stability and activity (ΔGH* is 0.06 eV). Further projected density of states (pDOS) analysis in combination with the wave function analysis of the SAC‐H bond revealed that the SAC‐dz2 orbital coincided with the H‐s orbital around the energy level of −2.50 eV, and orbital analysis confirmed the formation of σ bonds. This study provides an efficient multistep screening design framework of metal single‐atom catalyst for HER systems with similar two‐dimensional supports but different geometric configurations.

Improved Artificial Neural Network Classification Model based Metaheuristic Optimization for Handwritten Character Recognition

This study addresses the concerns regarding the performance of Handwritten Character Recognition (HCR) systems, focusing on the classification stage. It is widely acknowledged that the development of the classification model significantly impacts the overall performance of HCR. The problems identified specifically pertain to the classification model, particularly in the context of the Artificial Neural Network (ANN) learning problem, leading to low accuracy in recognizing handwritten characters. The objective of this study is to improve and refine the ANN classification model to achieve better HCR. To achieve this goal, this study proposed a hybrid Flower Pollination Algorithm with Artificial Neural Network (FPA-ANN) classification model for HCR. The FPA is one of the metaheuristic approaches is utilized as an optimization technique to enhance the performance of ANN, particularly by optimizing the network training process of ANN. The experimentation phase involves using the National Institute of Standards and Technology (NIST) handwritten character database. Finally, the proposed FPA-ANN classification model is analysed based on generated confusion matrix and evaluated performance of the classification model in terms of precision, sensitivity, specificity, F-score and accuracy.

Visualisation of moisture content distribution maps and classification of freshness level of loquats

Moisture content (MC) and freshness undergo evident changes during the storage of loquat fruits and thus are important indices of loquat fruit quality. Conventional methods for assessing MC and freshness are typically destructive and time-consuming, and cannot provide distribution maps of MC in the whole loquats. In this study, the use of hyperspectral imaging (HSI) was investigated for rapid and non-destructive prediction of MC in loquats, and for the visualisation of their MC distribution maps. Furthermore, freshness level discrimination models for loquats at different storage times were established using partial least squares discriminant analysis (PLS-DA), multinomial logistic regression (MLogR), and back-propagation neural network (BPNN) classification models. MC was well-predicted, with the best residual predictive deviation (RPD) of 2.40. Using the best prediction model, the distribution maps of MC in loquats were visualised. Additionally, the competitive adaptive reweighted sampling (CARS)-BPNN model obtained superior discrimination performance, with 95.56 % and 93.33 % accuracy for the calibration and prediction sets. Our results show that HSI is a promising fast and non-destructive method for determining MC and freshness of loquats.

Detection and metrology of micro-nano particles through the integration of image and laser modes

Micro-nano particles are an indispensable component in various industrial and scientific applications. Accurate granular information is critical for manufacturing products involving these particles. This study presents a novel detection method that merges image and laser modes to achieve precise size and shape determination of micro-nano particles. The proposed method integrates adaptive image analysis with laser particle size fitting to ensure accurate feature detection. Experimental results using ternary precursor samples demonstrate that the fusion mode outperforms both the laser and image modes in terms of size and shape accuracy. The errors of D10, D50, and D90 of the fusion mode are 4.89%, 1.88%, and 2.94%, respectively, which are significantly lower than those obtained using the laser or image mode alone. Additionally, the fusion mode retains the particle shape information from the image mode. The fusion model offers a promising approach for dual-mode detection of micro-nano particles, demonstrating enhanced robustness compared to single-mode methods. Furthermore, the study introduces an image processing method based on a BP neural network classification model that accurately classifies and detects agglomerated and non-agglomerated particles. The classification accuracy reaches 94.98%, and the particle size information remains within the acceptable range.

Diabetic retinopathy detection using Bilayered Neural Network classification model with resubstitution validation

In recent years, eye diseases in diabetic patients are one of the most common has been diabetic retinopathy (DR). which leads to complete blindness in advanced stages. Diabetes affects the blood vessels in the retina and causes vision loss. One of the ways to decrease the risk of this issue is to detect diabetic retinopathy in its early stages. This study describes a computer-aided screening system (DREAM) that uses a neural network classification model in machine learning to assess fundus images with different illumination and fields of vision and provide a severity grade for diabetic retinopathy. Moreover, the methodology of this study based on:•Enhancement techniques have been used on dataset images, histogram equalization, noise reduction and image scaling,•vSLAM has been selected as feature extraction,•Bilayered Neural Network under resubstitution validation used as a classification model.Finally, after testing on the DR severity grading system is tested on 6332 images of detection of diabetic retinopathy images, the result of the ROC curve is 0.985 on image dataset and obtains accuracy reached 98.5%. The classification result has been founded under MATLAB platform, beside it that be work with real time analysis and detection when patient eyes analysis.

A resource-efficient quantum convolutional neural network

Quantum Convolutional Neural Network (QCNN) has achieved significant success in solving various complex problems, such as quantum many-body physics and image recognition. In comparison to the classical Convolutional Neural Network (CNN) model, the QCNN model requires excellent numerical performance or efficient computational resources to showcase its potential quantum advantages, particularly in classical data processing tasks. In this paper, we propose a computationally resource-efficient QCNN model referred to as RE-QCNN. Specifically, through a comprehensive analysis of the complexity associated with the forward and backward propagation processes in the quantum convolutional layer, our results demonstrate a significant reduction in computational resources required for this layer compared to the classical CNN model. Furthermore, our model is numerically benchmarked on recognizing images from the MNIST and Fashion-MNIST datasets, achieving high accuracy in these multi-class classification tasks.

Early detection of infestation by mustard aphid, vegetable thrips and two-spotted spider mite in bok choy with deep neural network (DNN) classification model using hyperspectral imaging data

Although exclusion measures (e.g., air filters, biosecurity practices) can be employed to prevent occurrence of pest outbreaks, indoors vegetable farms in Singapore are still susceptible to various arthropod pests. Due to strong interest from the industry to pursue pesticide-free production, indoors pest management is often focused on early detection for timely containment and eradication, implying the importance of robust and vigorous pest monitoring programs. In recent years, application of machine vision technologies, especially hyperspectral imaging (HSI), has been studied for their capacity to early detect pest infestation. However, there is a lack of studies conducted in actual indoor environments and on multiple arthropod pests. Thus, this study aimed to non-destructively collect hyperspectral data of bok choy Brassica rapa subspecies chinensis which were healthy or infested with either mustard aphids Lipaphis erisymi, vegetable thrips Echinothrips americanus or two-spotted spider mites Tetranichus urticae in indoor environment to build deep neural network (DNN) classification model for early detection. Based on HSI data of control and infested plants collected daily over a period of two weeks, we found that point percentage change (PPC) values associated with leaf reflectance in 420–440 nm, 500–520 nm, 620–637 nm, 720–800 nm, and 850 nm were sensitive to infestation by the mentioned arthropod pests. Deep Neural Network (DNN) classification models trained on collected HSI data were found to outperform Linear Discriminant Analysis (LDA) and Support Vector Machine (SVM) classification models. DNN models achieved 92.8 ± 0.4 % overall classification accuracy across all days. As early as two days after infestation, DNN models could achieved classification precision values of 96.4 %, 96.9 %, 93.9 % and 100 % for control plants and plants infested with either aphids, spider mites or thrips respectively. These results highlight the feasibility of multiclass early detection of different arthropod pests and the potential of HSI system coupled with DNN classification as an autonomous plant health monitoring tool in indoor crop production.