Neural Network Classifier Research Articles

Breath Analysis Using Quartz Tuning Forks for Predicting Blood Glucose Levels Using Artificial Neural Networks.

Diabetes Mellitus (DM), a widespread metabolic disorder, poses lifelong health implications, demanding timely diagnosis and cautious monitoring for effective disease management. Traditional blood glucose tests are invasive and require medical expertise for intermittent checking, motivating the investigation of alternative, noninvasive methods. This study introduces an approach employing breath analysis through a set of 12 quartz tuning fork-based sensors enhanced using nanomaterials and dedicated artificial neural network (ANN) algorithms for data interpretation. The breath analysis methodology involves capturing unique breath signatures using the frequency-based sensor array. The accompanying neural network classification algorithm, customized for the sensor data, enables precise classification of data from 245 individuals as diabetic, prediabetic, or healthy. A neural network regression algorithm predicted blood glucose values and was compared with the actual values obtained from medical blood glucose measurement. The clinical relevance of the predicted blood glucose has been examined using error grids. The sensor array coupled with the ANN algorithm can identify diabetic, prediabetic, and control samples with 97% test accuracy. Blood glucose was predicted using neural network regression with a correlation coefficient of 0.89 and a mean square error of 0.13.

A modified lightweight quantum convolutional neural network for malicious code detection

Abstract Quantum neural network fully utilize the respective advantages of quantum computing and classical neural network, providing a new path for the development of artificial intelligence. In this paper, we propose a modified lightweight quantum convolutional neural network (QCNN), which contains a high-scalability and parameterized quantum convolutional layer and a quantum pooling circuit with quantum bit multiplexing, effectively utilizing the computational advantages of quantum systems to accelerate classical machine learning tasks. The experimental results show that the classification accuracy (precision, F1-score) of this QCNN on DataCon2020, Ember and BODMAS have been improved to 96.65% (94.3%, 96.74%), 92.4% (91.01%, 92.53%) and 95.6% (91.99%, 95.78%), indicating that this QCNN has strong robustness as well as good generalization performance for malicious code detection, which is of great significance to cyberspace security.

Neural network classification of lithological units based on integrated radar and multispectral data

Neural network classification of lithological units based on integrated radar and multispectral data

LMCD-OR: a large-scale, multilevel categorized diagnostic dataset for oral radiography

In recent years, digital dentistry has increasingly utilized advanced image analysis techniques, such as image classification and disease diagnosis, to improve clinical outcomes. Despite these advances, the lack of comprehensive benchmark datasets is a significant barrier. To address this gap, our research team develop LMCD-OR, a substantial collection of oral radiograph images designed to support extensive artificial intelligence (AI)-driven diagnostics. LMCD-OR comprises 3,818 digital imaging and communications in medicine (DICOM) oral X-ray images from local medical institutions that are meticulously annotated to provide broad category information for both primary dental outpatient services and detailed secondary disease diagnoses. This dataset is engineered to train and validate multiclassification models to improve the precision and scope of oral disease diagnostics. To ensure robust dataset validation, we employ four cutting-edge visual neural network classification models as benchmarks. These models are tested against rigorous performance metrics, demonstrating the ability of the dataset to support advanced image classification and disease diagnosis tasks. LMCD-OR is publicly available at http://dentaldataset.zeroacademy.net.

RS-Net: Hyperspectral Image Land Cover Classification Based on Spectral Imager Combined with Random Forest Algorithm

Recursive neural networks and transformers have recently become dominant in hyperspectral (HS) image classification due to their ability to capture long-range dependencies in spectral sequences. Despite the success of these sequential architectures, mainstream deep learning methods primarily handle two-dimensional structured data. However, challenges such as the curse of dimensionality, spectral variability, and confounding factors in hyperspectral remote sensing images limit their effectiveness, especially in remote sensing applications. To address this issue, this paper proposes a novel land cover classification algorithm that integrates random forests with a spectral transformer network structure (RS-Net). Firstly, this paper presents a combination of the Gramian Angular Field (GASF) and Gramian Angular Difference Field (GADF) algorithms, which effectively maps the multidimensional time series constructed for each pixel onto two-dimensional image features, enabling precise extraction and recognition in the backend network algorithms and improving the classification accuracy of land cover types. Secondly, to capture the relationships between features at different scales, this paper proposes a SpectralFormer network architecture using the Context and Structure Encoding (CASE) module to effectively learn dependencies between channels. This architecture enhances important features and suppresses unimportant ones, thereby addressing the semantic gap and improving the recognition capability of land cover features. Finally, the final prediction results are determined by a voting mechanism from the Random Forest algorithm, which synthesizes predictions from multiple decision trees to enhance classification stability and accuracy. To better compare the performance of RS-Net, this paper conducted extensive experiments on three benchmark HS datasets obtained from satellite and airborne imagers, comparing various classic neural network models. Surprisingly, the RS-Net algorithm achieves high performance and efficiency, offering a new and effective tool for land cover classification.

A Machine Learning and Deep Learning-Based Account Code Classification Model for Sustainable Accounting Practices

Accounting account codes are created within a specific logic framework to systematically and accurately record a company’s financial transactions. Currently, accounting reports are processed manually, which increases the likelihood of errors and slows down the process. This study aims to use image processing techniques to predict cash codes in accounting reports, automate accounting processes, improve accuracy, and save time. Deep learning embeddings from Inception V3, SqueezeNet, VGG-19, VGG-16, Painters, and DeepLoc networks were utilized in the feature extraction phase. A total of six learning algorithms, namely Logistic Regression, Gradient Boosting, Neural Network, kNN, Naive Bayes, and Stochastic Gradient Descent were employed to classify the images. The highest accuracy rate of 99.2% was achieved with the combination of the Inception V3 feature extractor and the Neural Network classifier. The results demonstrate that image processing methods significantly reduce error rates in accounting records, accelerate processes, and support sustainable accounting practices. This indicates that image processing techniques have substantial potential to contribute to digital transformation in accounting, helping businesses achieve their sustainability goals.

On the learning of high order polynomial reconstructions for essentially non-oscillatory schemes

Approximation accuracy and convergence behavior are essential required properties for the computed numerical solution of differential equations. These requirements restrict the application of deep learning networks in the domain of scientific computing. Moreover, the recipe to create suitable synthetic data which can be used to train a good model is also not very clear. This study focuses on learning of third order essentially non-oscillatory (ENO) and weighted essentially non-oscillatory (WENO) reconstructions using classification neural networks with small data sets. In particular, this work (i) proposes a novel way to obtain a third order WENO reconstruction which can be posed as classification problem, (ii) gives simple and novel approach to sample data sets which are small but rich enough to inherit the latent feature of inter-spatial regularity information in the constructed data, (iii) it is established that sampling of train data sets impacts quantitatively as well as qualitatively the required accuracy and non-oscillatory properties of resulting ENO3 and WENO3 schemes, (iv) proposes to use a limiter based multi model to retain desired accuracy as well as non-oscillatory properties of the resulting numerical schemes. Computational results are given which established that learned networks perform well and retain the features of the reconstruction methods.

Optical Image Processing using Eulerian Amplification

Recent advancements in digital photography, particularly through mobile phone cameras, microscopes, and satellite imaging, have made high-resolution image capture increasingly accessible. However, these captured images often contain subtle changes in color resolution and density that are challenging to observe with the naked eye. This study explores the application of Eulerian amplification as a method to detect and enhance such small-amplitude variations, making them visible for improved analysis. Unlike traditional approaches, Eulerian methods are particularly effective for processing smooth structures and high-quality printed materials where minor amplifications in color can reveal significant grading details. Using a pixel-by-pixel Eulerian path analysis, this method allows for a detailed comparison of color intensity values on a fine 8-bit scale, which ranges from 0 to 255. Through this analysis, color distribution patterns are visualized across the spatial structure of the image, which is particularly beneficial for identifying specific variations in quality in printed materials, such as highly graded cards. To manage large image structures, spatial decomposition using a multi-level Gaussian pyramid technique is employed. By applying temporal filtering within select frequency bands (0.4–4 Hz) at a coarse pyramid level, the algorithm effectively pools spatial data, enabling the detection of color density shifts that may indicate image defects. The amplified image data is further processed using a classifier neural network, which provides high sensitivity and specificity in detecting edge defects, image centering issues, scuffs, and breakages. The system demonstrates promising results in identifying true positives while maintaining low false-negative rates, thus confirming the reliability of deep machine learning algorithms in high-sensitivity defect detection. This paper presents the significance of Eulerian amplification in optical image processing, offering a robust tool for precise, automated defect detection across a variety of high-resolution image types.

Oxygen Content Control in the Electroslag Remelting Process: An Incremental Learning Strategy Based on Optimized Wasserstein Generative Adversarial Network with Gradient Penalty Data Augmentation

Electroslag remelting (ESR) is essential for producing high‐end special steel, but its complex process and numerous influencing factors make quality control challenging. This study addresses oxygen content control during ESR using a big data machine learning approach. An incremental learning strategy is proposed based on an optimized Wasserstein generative adversarial network with gradient penalty (WGAN‐GP) for data enhancement, focusing on G20Cr2Ni4A bearing steel. The WGAN‐GP model enhances time‐series data and metadata, utilizing long short‐term memory networks, fully connected networks, and attention mechanisms. The effectiveness of data enhancement is verified using a deep neural network classifier and statistical methods. Data is divided into historical and data streams, with an incremental learning strategy based on histogram gradient boosting regression trees to prevent catastrophic forgetting and improve efficiency through knowledge distillation and real‐time hyperparameter adjustment. Results show that the data augmentation method significantly improves model generalization and accuracy in small sample metallurgy. The incremental learning strategy enhances prediction accuracy for oxygen content, contributing to better cleanliness quality of electroslag steel. This study offers a novel approach for addressing small sample challenges in metallurgical processes.

Transforming Motor Imagery Analysis: A Novel EEG Classification Framework Using AtSiftNet Method

This paper presents an innovative approach for the Feature Extraction method using Self-Attention, incorporating various Feature Selection techniques known as the AtSiftNet method to enhance the classification performance of motor imaginary activities using electrophotography (EEG) signals. Initially, the EEG signals were sorted and then denoised using multiscale principal component analysis to obtain clean EEG signals. However, we also conducted a non-denoised experiment. Subsequently, the clean EEG signals underwent the Self-Attention feature extraction method to compute the features of each trial (i.e., 350×18). The best 1 or 15 features were then extracted through eight different feature selection techniques. Finally, five different machine learning and neural network classification models were employed to calculate the accuracy, sensitivity, and specificity of this approach. The BCI competition III dataset IV-a was utilized for all experiments, encompassing the datasets of five volunteers who participated in the competition. The experiment findings reveal that the average accuracy of classification is highest among ReliefF (i.e., 99.946%), Mutual Information (i.e., 98.902%), Independent Component Analysis (i.e., 99.62%), and Principal Component Analysis (i.e., 98.884%) for both 1 and 15 best-selected features from each trial. These accuracies were obtained for motor imagery using a Support Vector Machine (SVM) as a classifier. In addition, five-fold validation was performed in this paper to assess the fair performance estimation and robustness of the model. The average accuracy obtained through five-fold validation is 99.89%. The experiments’ findings indicate that the suggested framework provides a resilient biomarker with minimal computational complexity, making it a suitable choice for advancing Motor Imagery Brain–Computer Interfaces (BCI).

Stance Detection in the Context of Fake News—A New Approach

Online social networks (OSNs) are inundated with an enormous daily influx of news shared by users worldwide. Information can originate from any OSN user and quickly spread, making the task of fact-checking news both time-consuming and resource-intensive. To address this challenge, researchers are exploring machine learning techniques to automate fake news detection. This paper specifically focuses on detecting the stance of content producers—whether they support or oppose the subject of the content. Our study aims to develop and evaluate advanced text-mining models that leverage pre-trained language models enhanced with meta features derived from headlines and article bodies. We sought to determine whether incorporating the cosine distance feature could improve model prediction accuracy. After analyzing and assessing several previous competition entries, we identified three key tasks for achieving high accuracy: (1) a multi-stage approach that integrates classical and neural network classifiers, (2) the extraction of additional text-based meta features from headline and article body columns, and (3) the utilization of recent pre-trained embeddings and transformer models.

PointCloud-At: Point Cloud Convolutional Neural Networks with Attention for 3D Data Processing

The rapid growth in technologies for 3D sensors has made point cloud data increasingly available in different applications such as autonomous driving, robotics, and virtual and augmented reality. This raises a growing need for deep learning methods to process the data. Point clouds are difficult to be used directly as inputs in several deep learning techniques. The difficulty is raised by the unstructured and unordered nature of the point cloud data. So, machine learning models built for images or videos cannot be used directly on point cloud data. Although the research in the field of point clouds has gained high attention and different methods have been developed over the decade, very few research works directly with point cloud data, and most of them convert the point cloud data into 2D images or voxels by performing some pre-processing that causes information loss. Methods that directly work on point clouds are in the early stage and this affects the performance and accuracy of the models. Advanced techniques in classical convolutional neural networks, such as the attention mechanism, need to be transferred to the methods directly working with point clouds. In this research, an attention mechanism is proposed to be added to deep convolutional neural networks that process point clouds directly. The attention module was proposed based on specific pooling operations which are designed to be applied directly to point clouds to extract vital information from the point clouds. Segmentation of the ShapeNet dataset was performed to evaluate the method. The mean intersection over union (mIoU) score of the proposed framework was increased after applying the attention method compared to a base state-of-the-art framework that does not have the attention mechanism.

Texture-Based Classification of Fetal Growth Restriction From Intrauterine Neurosonographic Image.

Fetal growth restriction (FGR) is a condition where fetuses fail to reach their genetic potential for growth, posing a significant health challenge for newborns. The aim of this research was to explore the efficacy of texture-based analysis of neurosonographic images in identifying FGR in fetuses, which may provide a promising tool for early assessment of FGR. A retrospective analysis collected 100 intrauterine neurosonographic images from 50 FGR and 50 gestational age-appropriate fetuses. Using MaZda software, approximately 300 texture features were extracted from occipital white matter (OWM) and cerebellum of intrauterine neurosonographic images, respectively. Then 10 optimal features were separately selected by 3 algorithms, including the Fisher coefficient method, the method of minimizing classification error probability and average correlation coefficients, and the mutual information coefficient method. Further, the 10 statistically most significant features were selected from these sets to form the mixed feature set. After nonlinear discriminant analysis was performed to reduce feature dimensionality, the artificial neural network (ANN) classifier was conducted, respectively. For OWM and cerebellum, a total of 11 and 14 statistically significant features were selected. When the mixed feature sets of OWM and cerebellum were applied to ANN classifier, classification accuracy were 90.00% (κ = 0.800; P < .001) and 93.00% (κ = 0.860; P < .001), and the receiver operating characteristic curve for identifying FGR showed an area under the curve of 0.82 and 0.87. Texture analysis of fetal intrauterine neurosonographic images is a feasible and noninvasive strategy for evaluating FGR fetuses.

QSIM: A Quantum-inspired hierarchical semantic interaction model for text classification

Semantic interaction modeling is a fundamental technology in natural language understanding that guides models to extract deep semantic information from text. Currently, the attention mechanism is one of the most effective techniques in semantic interaction modeling, which learns word-level attention representation by measuring the relevance between different words. However, the attention mechanism is limited to word-level semantic interaction, it cannot meet the needs of fine-grained interactive information for some text classification tasks. In recent years, quantum-inspired language modeling methods have successfully constructed quantized representations of language systems in Hilbert spaces, which use density matrices to achieve fine-grained semantic interaction modeling.This paper presents a Quantum-inspired hierarchical Semantic Interaction Model (QSIM), which follows the sememe-word-sentence language construction principle and utilizes quantum entanglement theory to capture hierarchical semantic interaction information in Hilbert space. Our work builds on the idea of the attention mechanism and extends it. Specifically, we explore the original semantic space from a quantum theory perspective and derive the core semantic space using the Schmidt decomposition technique, where: (1) Sememe is represented as the unit vector in the two-dimensional minimum semantic space; (2) Word is represented as reduced density matrices in the core semantic space, where Schmidt coefficients quantify sememe-level semantic interaction. Compared to density matrices, reduced density matrices capture fine-grained semantic interaction information with lower computational cost; (3) Sentence is represented as quantum superposition states of words, and the degree of word-level semantic interaction is measured using entanglement entropy.To evaluate the model’s performance, we conducted experiments on 15 text classification datasets. The experimental results demonstrate that our model is superior to classical neural network models and traditional quantum-inspired language models. Furthermore, the experiment also confirms two distinct advantages of QISM: (1) flexibility, as it can be integrated into various mainstream neural network text classification architectures; and (2) practicability, as it alleviates the problem of parameter growth inherent in density matrix calculation in quantum language model.

Toward mechanistic models “augmented” by machine learning. Example of a drying simulation data set exploited by a Physics Informed Neural Network

Mechanistic modeling of drying is well-established since several decades. Based on fundamental balance equations and driven by relevant material parameters, it can predict the entire process, including configurations that were not observed before. Besides, thanks to their ability to tackle non-linear and dynamics problems, approaches based on Machine Learning (ML) based are capable of coping with complex situations even better than mechanistic modeling. The main drawback of mechanistic models is their complexity as operational tools, namely in providing the whole set of product characteristics, while the main drawback of ML tools is its restriction to the domain paved by the data set. This paper summarizes the physics of the mechanistic formulation and then presents the different possibilities of coupling the mechanistic and ML approaches to obtain an “augmented” mechanistic model. The idea is to merge the advantages of both worlds. Different strategies can be imagined: A full coupled method (hybrid model) A fully decoupled method A cascade coupling The second part of the paper gives examples of a fully decoupled approach the mechanistic model is used to populate a data set, which is further exploited by a neural network. The originality of the work is to compare a classical neural network with a Physics Informed Neural Network.

An AI-Based Approach for Developing a Recommendation System for Underground Mining Methods Pre-Selection

Selecting the most appropriate mining method to recover mineral resources is a critical decision-making task in mining project development. This study introduces an artificial intelligence-based mining methods recommendation system (AI-MMRS) for the pre-selection of underground mining methods. The study integrates and evaluates the capability of two approaches for mining methods selection (MMS): the memory-based collaborative filtering (CF) approach aided by the UBC-MMS system to predict the top-3 relevant mining methods and supervised machine learning (ML) classification algorithms to enhance the effectiveness and novelty of the AI-MMRS, addressing the limitations of the CF approach. The results reveal that the memory-based CF approach achieves an accuracy ranging from 81.8% to 87.9%. Among the classification algorithms, artificial neural network (ANN) and k-nearest neighbors (KNN) classifiers perform the best, with accuracy levels of 66.7% and 63.6%, respectively. These findings demonstrate the effectiveness and viability of both approaches in MMS, acknowledging their limitations and the need for continuous training and optimization. The proposed AI-MMRS for the pre-selection stage supplemented by the direct involvement of mining professionals in later stages of MMS, has the potential to significantly aid in the MMS decision-making, providing data-driven and experience-based recommendations following the ongoing evolution of mining practices.

Multimodal brain tumor segmentation and classification from MRI scans based on optimized DeepLabV3+ and interpreted networks information fusion empowered with explainable AI

Explainable artificial intelligence (XAI) aims to offer machine learning (ML) methods that enable people to comprehend, properly trust, and create more explainable models. In medical imaging, XAI has been adopted to interpret deep learning black box models to demonstrate the trustworthiness of machine decisions and predictions. In this work, we proposed a deep learning and explainable AI-based framework for segmenting and classifying brain tumors. The proposed framework consists of two parts. The first part, encoder-decoder-based DeepLabv3+ architecture, is implemented with Bayesian Optimization (BO) based hyperparameter initialization. The different scales are performed, and features are extracted through the Atrous Spatial Pyramid Pooling (ASPP) technique. The extracted features are passed to the output layer for tumor segmentation. In the second part of the proposed framework, two customized models have been proposed named Inverted Residual Bottleneck 96 layers (IRB-96) and Inverted Residual Bottleneck Self-Attention (IRB-Self). Both models are trained on the selected brain tumor datasets and extracted features from the global average pooling and self-attention layers. Features are fused using a serial approach, and classification is performed. The BO-based hyperparameters optimization of the neural network classifiers is performed and the classification results have been optimized. An XAI method named LIME is implemented to check the interpretability of the proposed models. The experimental process of the proposed framework was performed on the Figshare dataset, and an average segmentation accuracy of 92.68 % and classification accuracy of 95.42 % were obtained, respectively. Compared with state-of-the-art techniques, the proposed framework shows improved accuracy.

Unsupervised detection and mapping of sparks in the Electrochemical Discharge Machining (ECDM) process

Material removal in electrochemical discharge machining is caused by sparks generated in a tool immersed in an electrolytic solution. Being the primary machining agent in this non-contact machining process, mapping the locations of microscopic sparks is of great interest. The distribution of sparks around the tool surface could give insights into the machined hole properties like the size, surface finish, and depth as compared to the machining parameters such as applied voltage, tool size, rotation speed, and feed rate. This paper is focused on detecting sparks in photographs of the ECDM process captured using a high-speed camera. A novel approach of using a tri-planar reflective surface for capturing the location of sparks in 3D space using a 2D camera output is attempted. Traditional spark detection methods use neural network classifiers that need labeled data for training. This labeled data often comes from human intervention and contains inherent biases that could lead to misclassification. In this paper, an unsupervised spark detection methodology is demonstrated, which eliminates the need for human intervention and relies on the number of neighboring pixels detected in regions of interest (ROIs). The feasibility of using adaptive background modeling to classify thousands of images and identify the ones with sparks is demonstrated in this work. The masking technique combining effects of erosion followed by dilation is used to determine the exact boundaries of the spark contours in every image. Centroids for each of these contours are then transformed from the skewed coordinate system as observed in camera images, to a three-dimensional orthogonal coordinates system centered around the tool. The same procedure is repeated for various voltages to benchmark the distribution of sparks around a tool tip in an ECDM process.

Advanced Dual-Input Artificial Optical Synapse for Recognition and Generative Neural Network

Perovskite materials have emerged as leading candidates for optical synaptic devices due to their superior photosensitivity and tunable optoelectronic properties. However, the practical application of perovskite-based optoelectronic synaptic transistors has been hindered by issues of poor stability and high toxicity. This study developed an artificial synaptic thin film transistor (TFT) based on a Cs2AgBiBr6/InOx bilayer. The device demonstrated synaptic behavior under optoelectronic hybrid stimulation (largest wavelength ~520nm), such as excitatory postsynaptic current (EPSC), inhibitory postsynaptic current (IPSC), paired-pulse facilitation (PPF), spike-timing-dependent plasticity (STDP), short-term memory (STM) and long-term memory (LTM). Notably, the “light writing and voltage erasing” characteristics of these devices could be utilized to construct a convolutional neural network (CNN) classifier for the CIFAR-10 dataset, demonstrating noise tolerance close to the human eye. The loss in recognition accuracy was within 1% when Gaussian white noise and salt pepper noise were added. Furthermore, these devices exhibited great potential in Generative Adversarial Networks (CycleGAN), with the generated image quality achieving levels of 0.637 and 0.715 in Improved Perceptual Image Processing System (IPIPS) and Structural Similarity Index (SSIM) evaluation metrics for the zebra dataset, indicating good image quality. This study indicates that our Cs2AgBiBr6-based artificial optical synaptic thin-film transistors (TFTs) are promising for sensing and in-memory computing applications.

ANN uncertainty estimates in assessing fatty liver content from ultrasound data

Background and objectiveThis article uses three different probabilistic convolutional architectures applied to ultrasound image analysis for grading Fatty Liver Content (FLC) in Metabolic Dysfunction Associated Steatotic Liver Disease (MASLD) patients. Steatosis is a new silent epidemic and its accurate measurement is an impelling clinical need, not only for hepatologists, but also for experts in metabolic and cardiovascular diseases. This paper aims to provide a robust comparison between different uncertainty quantification strategies to identify advantages and drawbacks in a real clinical setting. MethodsWe used a classical Convolutional Neural Network, a Monte Carlo Dropout, and a Bayesian Convolutional Neural Network with the goal of not only comparing the goodness of the predictions, but also to have access to an evaluation of the uncertainty associated with the outputs. ResultsWe found that even if the prediction based on a single ultrasound view is reliable (relative RMSE [5.93%-12.04%]), networks based on two ultrasound views outperform them (relative RMSE [5.35%-5.87%]). In addition, the results show that the introduction of a “not confident” category contributes to increase the percentage of correctly predicted cases and to decrease the percentage of mispredicted cases, especially for semi-intrusive methods. ConclusionsThe possibility of having access to information about the confidence with which the network produces its outputs is a great advantage, both from the point of view of physicians who want to use neural networks as computer-aided diagnosis, and for developers who want to limit overfitting and obtain information about dataset problems in terms of out-of-distribution detection.