Neural Network-based Algorithm Research Articles

An overview of hand gesture recognition based on computer vision

Hand gesture recognition emerges as one of the foremost sectors which has gone through several developments within pattern recognition. Numerous studies and research endeavors have explored methodologies grounded in computer vision within this domain. Despite extensive research endeavors, there is still a need for a more thorough evaluation of the efficiency of various methods in different environments along with the challenges encountered during the application of these methods. The focal point of this paper is the comparison of different research in the domain of vision-based hand gesture recognition. The objective is to find out the most prominent methods by reviewing efficiency. Concurrently, the paper delves into presenting potential solutions for challenges faced in different research. A comparative analysis particularly centered around traditional methods and convolutional neural networks like random forest, long short-term memory (LSTM), heatmap, and you only look once (YOLO). considering their efficacy. Where convolutional neural network-based algorithms performed best for recognizing the gestures and gave effective solutions for the challenges faced by the researchers. In essence, the findings of this review paper aim to contribute to future implementations and the discovery of more efficient approaches in the gesture recognition sector.

Practical Application and Case Analysis of Computer Image Vision Technology in Music Education and Teaching

Exploring how to utilize images to enrich music teaching content and provide a more visually impactful learning experience is an important topic. Therefore, this paper introduces a convolutional neural network-based algorithm for extracting audio features to construct a music visualization model. By identifying features such as note pitches, it enhances pitch recognition and integrates with CNN algorithms for audio information visualization. Experimental results demonstrate an accuracy rate exceeding 97%, showcasing the significant advantage of this method in visualizing audio information in music multimedia classrooms. It provides technical support for bringing a new visual experience to music education.

Neural network-based algorithm for door handle recognition using RGBD cameras

The ability to recognize and interact with a variety of doorknob designs is an important component on the path to true robot adaptability, allowing robotic systems to effectively interact with a variety of environments and objects The problem addressed in this paper is to develop and implement a method for recognizing the position of a door handle by a robot using data from an RGBD camera. To achieve this goal, we propose a revolutionary approach designed for autonomous robots that allows them to identify and manipulate door handles in different environments using data obtained from RGBD cameras. This was achieved by creating and annotating a complete dataset consisting of 5000 images of door handles from different angles, with the coordinates of the vertices of the bounding rectangles labeled. The architectural basis of the proposed approach is based on MobileNetV2, combined with a special decoder that optimally increases the resolution to 448 pixels. A new activation function specially designed for this neural network is implemented to ensure increased accuracy and efficiency of raw data processing. The most important achievement of this study is the model's ability to work in real-time, processing up to 16 images per second. This research paves the way for new advancements in the fields of robotics and computer vision, making a substantial contribution to the practical deployment of autonomous robots in a myriad of life's spheres.

MOTOR IMAGERY SIGNAL CLASSIFICATION FOR BRAIN–COMPUTER INTERFACE USING RideNN WITH HOLO-ENTROPY FEATURES

The brain–computer interface (BCI) database’s motor assessment depends heavily on the motor imagery (MI) signal classification. By examining the multiple patterns of different creative tasks in the electroencephalogram (EEG) signals, the intention of humans is translated into computer-based commands in the MI-based signals of BCI data. Nevertheless, low accuracy and efficiency are issues with MI–EEG signals’ classification because of the low signal-to-noise ratio, huge individual differences, overall volatility, and complexity in the signal. To overcome these problems, this research proposes a rider optimization algorithm-based neural network (ROA-based NN) to classify the MI signals effectively. Pre-processing is done after collecting the dataset of raw EEG signals. The suitable electrodes, such as C3, C4, and Cz, are subsequently chosen from the MI signals. Using the holo-entropy-based WPD feature extractor, tunable Q-factor wavelet transform (T-QWT), and common spatial patterns in the model, the pertinent features are extracted from the chosen electrodes. The developed holo-entropy-based WPD feature examines the electrode structure’s association. As a result, the most diverse signals are removed from the chosen electrodes before being input into the proposed RideNN classifier, where the proposed ride optimization algorithm optimizes classification performance and correctly predicts and classes the output from the MI signals that have been analyzed. The developed RideNN classifier recognizes the patterns more accurately processes more data and tackles the noise and incomplete data effectively. Utilizing the parameters of accuracy, sensitivity, and specificity, the results are evaluated. The PROA-based RideNN and the combined features obtain the maximum accuracy of 92.24%, sensitivity of 92.26%, and specificity of 92.14% for the BCI competition-IV 2a database. The qPROA-based RideNN and the combined features obtain the maximum accuracy of 92.11%, sensitivity of 91.98%, and specificity of 92.35% for the BCI competition-IV 2b database.

Improving Earth-like planet detection in radial velocity using deep learning

Many novel methods have been proposed to mitigate stellar activity for exoplanet detection as the presence of stellar activity in radial velocity (RV) measurements is the current major limitation. Unlike traditional methods that model stellar activity in the RV domain, more methods are moving in the direction of disentangling stellar activity at the spectral level. As deep neural networks have already been proven to be one of the most effective tools in data mining, in this work, we explore their potential in the context of Earth-like planet detection in RV measurements. The goal of this paper is to present a novel convolutional neural network-based algorithm that efficiently models stellar activity signals at the spectral level, enhancing the detection of Earth-like planets. Based on the idea that the presence of planets can only produce a Doppler shift at the spectral level while the presence of stellar activity can introduce a variation in the profile of spectral lines (asymmetry and depth change), we trained a convolutional neural network to build the correlation between the change in the spectral line profile and the corresponding RV, full width at half maximum (FWHM) and bisector span (BIS) values derived from the classical cross-correlation function. This algorithm has been tested on three intensively observed stars: Alpha Centauri B (HD\,128621), Tau ceti (HD\,10700), and the Sun. By injecting simulated planetary signals at the spectral level, we demonstrate that our machine learning algorithm can achieve, for HD\,128621 and HD\,10700, a detection threshold of 0.5 m/s in semi-amplitude for planets with periods ranging from 10 to 300 days. This threshold would correspond to the detection of a sim 4\,$ M oplus $ in the habitable zone of those stars. On the HARPS-N solar dataset, thanks to significantly more data, our algorithm is even more efficient at mitigating stellar activity signals and can reach a threshold of 0.2 m/s, which would correspond to a 2.2\,$ M oplus $ planet on the orbit of the Earth. To the best of our knowledge, it is the first time that such low detection thresholds are reported for the Sun, but also for other stars, and therefore this highlights the efficiency of our convolutional neural network-based algorithm at mitigating stellar activity in RV measurements.

Implementation of a Long Short-Term Memory Neural Network-Based Algorithm for Dynamic Obstacle Avoidance.

Autonomous mobile robots are essential to the industry, and human-robot interactions are becoming more common nowadays. These interactions require that the robots navigate scenarios with static and dynamic obstacles in a safely manner, avoiding collisions. This paper presents a physical implementation of a method for dynamic obstacle avoidance using a long short-term memory (LSTM) neural network that obtains information from the mobile robot's LiDAR for it to be capable of navigating through scenarios with static and dynamic obstacles while avoiding collisions and reaching its goal. The model is implemented using a TurtleBot3 mobile robot within an OptiTrack motion capture (MoCap) system for obtaining its position at any given time. The user operates the robot through these scenarios, recording its LiDAR readings, target point, position inside the MoCap system, and its linear and angular velocities, all of which serve as the input for the LSTM network. The model is trained on data from multiple user-operated trajectories across five different scenarios, outputting the linear and angular velocities for the mobile robot. Physical experiments prove that the model is successful in allowing the mobile robot to reach the target point in each scenario while avoiding the dynamic obstacle, with a validation accuracy of 98.02%.

Development and validation of machine learning models to predict unplanned hospitalizations of patients with diabetes within the next 12 months

BACKGROUND: The incidence of diabetes mellitus (DM) both in the Russian Federation and in the world has been steadily increasing for several decades. Stable population growth and current epidemiological characteristics of DM lead to enormous economic costs and significant social losses throughout the world. The disease often progresses with the development of specific complications, while significantly increasing the likelihood of hospitalization. The creation and inference of a machine learning model for predicting hospitalizations of patients with DM to an inpatient medical facility will make it possible to personalize the provision of medical care and optimize the load on the entire healthcare system.AIM: Development and validation of models for predicting unplanned hospitalizations of patients with diabetes due to the disease itself and its complications using machine learning algorithms and data from real clinical practice.MATERIALS AND METHODS: 170,141 depersonalized electronic health records of 23,742 diabetic patients were included in the study. Anamnestic, constitutional, clinical, instrumental and laboratory data, widely used in routine medical practice, were considered as potential predictors, a total of 33 signs. Logistic regression (LR), gradient boosting methods (LightGBM, XGBoost, CatBoost), decision tree-based methods (RandomForest and ExtraTrees), and a neural network-based algorithm (Multi-layer Perceptron) were compared. External validation was performed on the data of the separate region of Russian Federation.RESULTS: The best results and stability to external validation data were shown by the LightGBM model with an AUC of 0.818 (95% CI 0.802–0.834) in internal testing and 0.802 (95% CI 0.773–0.832) in external validation.CONCLUSION: The metrics of the best model were superior to previously published studies. The results of external validation showed the relative stability of the model to new data from another region, that reflects the possibility of the model’s application in real clinical practice.

A novel hybrid optimized incremental relevance vector machine and filtering technique for state of charge estimation of lithium-ion batteries

Despite not requiring a comprehensive understanding of the intricate electrochemical reactions within the battery, the data-driven approach faces challenges regarding state of charge (SOC) fluctuations during abrupt changes in battery current. Currently, research on data-driven filtering methods predominantly focuses on neural network-based algorithms, with limited exploration of the relevance vector machine (RVM) algorithm that possesses advantages in handling small sample sizes, high sparsity, and probabilistic distributions. Drawing inspiration from this premise, the present manuscript introduces an innovative fusion algorithm, founded upon the framework of the adaptive Kalman filtering (AKF) algorithm, which intricately melds the optimized incremental RVM (OIRVM) with the Coulomb counting methodology, called Coulomb_AKF_OIRVM (CFR) algorithm. Experimental findings suggest that the CFR algorithm displays exceptional generalization capacity across different driving cycles and temperatures, exhibiting high estimation accuracy and robustness. The average root mean square error, determination coefficient, mean absolute error and maximum absolute error of the public datasets and experimental datasets are 0.25 %, 0.02 %, 0.20 %, and 0.85 %, respectively.

Computational 3D topographic microscopy from terabytes of data per sample

We present a large-scale computational 3D topographic microscope that enables 6-gigapixel profilometric 3D imaging at micron-scale resolution across >110 cm2 areas over multi-millimeter axial ranges. Our computational microscope, termed STARCAM (Scanning Topographic All-in-focus Reconstruction with a Computational Array Microscope), features a parallelized, 54-camera architecture with 3-axis translation to capture, for each sample of interest, a multi-dimensional, 2.1-terabyte (TB) dataset, consisting of a total of 224,640 9.4-megapixel images. We developed a self-supervised neural network-based algorithm for 3D reconstruction and stitching that jointly estimates an all-in-focus photometric composite and 3D height map across the entire field of view, using multi-view stereo information and image sharpness as a focal metric. The memory-efficient, compressed differentiable representation offered by the neural network effectively enables joint participation of the entire multi-TB dataset during the reconstruction process. Validation experiments on gauge blocks demonstrate a profilometric precision and accuracy of 10 µm or better. To demonstrate the broad utility of our new computational microscope, we applied STARCAM to a variety of decimeter-scale objects, with applications ranging from cultural heritage to industrial inspection.

Ovarian Cancer surgical consideration is markedly improved by the neural network powered-MIA3G multivariate index assay.

Surgery remains the main treatment option for an adnexal mass suspicious of ovarian cancer. The malignancy rate is, however, only 10-15% in women undergoing surgery. This results in a high number of unnecessary surgeries. A surveillance-based approach is recommended to form the basis for surgical referrals. We have previously reported the clinical performance of MIA3G, a deep neural network-based algorithm, for assessing ovarian cancer risk. In this study, we show that MIA3G markedly improves the surgical selection for women presenting with adnexal masses. MIA3G employs seven serum biomarkers, patient age, and menopausal status. Serum samples were collected from 785 women (IQR: 39-55 years) across 12 centers that presented with adnexal masses. MIA3G risk scores were calculated for all subjects in this cohort. Physicians had no access to the MIA3G risk score when deciding upon a surgical referral. The performance of MIA3G for surgery referral was compared to clinical and surgical outcomes. MIA3G was also tested in an independent cohort comprising 29 women across 14 study sites, in which the physicians had access to and utilized MIA3G prior to surgical consideration. When compared to the actual number of surgeries (n = 207), referrals based on the MIA3G score would have reduced surgeries by 62% (n = 79). The reduction was higher in premenopausal patients (77%) and in patients ≤55 years old (70%). In addition, a 431% improvement in malignancy prediction would have been observed if physicians had utilized MIA3G scores for surgery selection. The accuracy of MIA3G referral was 90.00% (CI 87.89-92.11), while only 9.18% accuracy was observed when the MIA3G score was not used. These results were corroborated in an independent multi-site study of 29 patients in which the physicians utilized MIA3G in surgical consideration. The surgery reduction was 87% in this cohort. Moreover, the accuracy and concordance of MIA3G in this independent cohort were each 96.55%. These findings demonstrate that MIA3G markedly augments the physician's decisions for surgical intervention and improves malignancy prediction in women presenting with adnexal masses. MIA3G utilization as a clinical diagnostic tool might help reduce unnecessary surgeries.

A neural network-based algorithm for the reconstruction and filtering of single particle trajectory in magnetic particle tracking.

Magnetic particle tracking (MPT) is a recently developed non-invasive measurement technique that has gained popularity for studying dense particulate or granular flows. This method involves tracking the trajectory of a magnetically labeled particle, the field of which is modeled as a dipole. The nature of this method allows it to be used in opaque environments, which can be highly beneficial for the measurement of dense particle dynamics. However, since the magnetic field of the particle used is weak, the signal-to-noise ratio is usually low. The noise from the measuring devices contaminates the reconstruction of the magnetic tracer's trajectory. A filter is then needed to reduce the noise in the final trajectory results. In this work, we present a neural network-based framework for MPT trajectory reconstruction and filtering, which yields accurate results and operates at very high speed. The reconstruction derived from this framework is compared to the state-of-the-art extended Kalman filter-based reconstruction.

Evaluation of multiple machine learning models for ASR expansion of concrete

Machine learning algorithms (MLA) show promise in predicting alkali-silica reaction (ASR) expansion. Nevertheless, the predictive performance of different machine learning algorithms (MLAs) varies across various engineering scenarios and datasets. Given that, this study was devoted to constructing an appropriate predictive model for ASR expansion of concrete. Multiple MLAs including evolutionary-based, support vector-based, assembly-based, and neural network-based algorithms, were employed to establish predictive models. An optimization technique, the particle swarm optimization algorithm, was also utilized and integrated with the aforementioned MLA models to effectively search for optimal hyperparameters and topological structures. The predictive performance was evaluated using statistical methods, which showed that the extreme gradient boosting (XGBoost) model had the best fitting goodness among the standard algorithms, with an R2 value of 0.9695. Moreover, the Particle Swarm Optimization (PSO) model significantly improved the predictive performance of all Machine Learning Algorithm (MLA) models. The error distribution analysis was then performed. There are fewer large errors in the predicted values of the PSO-XGBoost model. Monte Carlo simulation and probability distribution functions were used to thoroughly analyze the stability of each model after multiple independent runs. The results show that all models have excellent stability. The PSO-XGBoost model performed the best and was subsequently employed to conduct feature analysis to determine the significance of each input parameter and capture the underlying evidence of ASR expansion.

Spectrum-efficient user grouping and resource allocation based on deep reinforcement learning for mmWave massive MIMO-NOMA systems

Millimeter-wave (mmWave) massive multiple-input multiple-output non-orthogonal multiple access (MIMO-NOMA) is proven to be a primary technique for sixth-generation (6G) wireless communication networks. However, the great increase in users and antennas brings challenges for interference suppression and resource allocation for mmWave massive MIMO-NOMA systems. This study proposes a spectrum-efficient and fast convergence deep reinforcement learning (DRL)-based resource allocation framework to optimize user grouping and allocation of subchannel and power. First, an enhanced K-means grouping algorithm is proposed to reduce the multi-user interference and accelerate the convergence. Then, a dueling deep Q-network (DQN) structure is proposed to perform subchannel allocation, which further improves the convergence speed. Moreover, a deep deterministic policy gradient (DDPG)-based power resource allocation algorithm is designed to avoid the performance loss caused by power quantization and improve the system’s achievable sum-rate. The simulation results demonstrate that our proposed scheme outperforms other neural network-based algorithms in terms of convergence performance, and can achieve higher system capacity compared with the greedy algorithm, the random algorithm, the RNN algorithm, and the DoubleDQN algorithm.

State of Health Prediction of Electric Vehicles’ Retired Batteries Based on First-Life Historical Degradation Data Using Predictive Time-Series Algorithms

The exponential growth of electric and hybrid vehicles, now numbering close to 6 million on the roads, has highlighted the urgent need to address the environmental impact of their lithium-ion batteries as they approach their end-of-life stages. Repurposing these batteries as second-life batteries (SLBs) for less demanding non-automotive applications is a promising avenue for extending their usefulness and reducing environmental harm. However, the shorter lifespan of SLBs brings them perilously close to their ageing knee, a critical point where further use risks thermal runaway and safety hazards. To mitigate these risks, effective battery management systems must accurately predict the state of health of these batteries. In response to this challenge, this study employs time-series artificial intelligence (AI) models to forecast battery degradation parameters using historical data from their first life cycle. Through rigorous analysis of a lithium-ion NMC cylindrical cell, the study tracks the trends in capacity and internal resistance fade across both the initial and second life stages. Leveraging the insights gained from first-life data, predictive models such as the Holt–Winters method and the nonlinear autoregressive (NAR) neural network are trained to anticipate capacity and internal resistance values during the second life period. These models demonstrate high levels of accuracy, with a maximum error rate of only 2%. Notably, the NAR neural network-based algorithm stands out for its exceptional ability to predict local noise within internal resistance values. These findings hold significant implications for the development of specifically designed battery management systems tailored for second-life batteries.

Novel Approach to Phase-Sensitive Optical Time-Domain Reflectometry Response Analysis with Machine Learning Methods.

This paper is dedicated to the investigation of the metrological properties of phase-sensitive reflectometric measurement systems, with a particular focus on addressing the non-uniformity of responses along optical fibers. The authors highlight challenges associated with the stochastic distribution of Rayleigh reflectors in fiber optic systems and propose a methodology for assessing response non-uniformity using both cross-correlation algorithms and machine learning approaches, using chirped-reflectometry as an example. The experimental process involves simulating deformation impact by altering the light source's wavelength and utilizing a chirped-reflectometer to estimate response non-uniformity. This paper also includes a comparison of results obtained from cross-correlation and neural network-based algorithms, revealing that the latter offers more than 34% improvement in accuracy when measuring phase differences. In conclusion, the study demonstrates how this methodology effectively evaluates response non-uniformity along different sections of optical fibers.

Neural network establishes co-occurrence links between transformation products of the contaminant and the soil microbiome

It remains challenging to establish reliable links between transformation products (TPs) of contaminants and corresponding microbes. This challenge arises due to the sophisticated experimental regime required for TP discovery and the compositional nature of 16S rRNA gene amplicon sequencing and mass spectrometry datasets, which can potentially confound statistical inference. In this study, we present a new strategy by combining the use of 2H-labeled Stable Isotope-Assisted Metabolomics (2H-SIAM) with a neural network-based algorithm (i.e., MMvec) to explore links between TPs of pyrene and the soil microbiome. The links established by this novel strategy were further validated using different approaches. Briefly, a metagenomic study provided indirect evidence for the established links, while the identification of pyrene degraders from soils, and a DNA-based stable isotope probing (DNA-SIP) study offered direct evidence. The comparison among different approaches, including Pearson's and Spearman's correlations, further confirmed the superior performance of our strategy. In conclusion, we summarize the unique features of the combined use of 2H-SIAM and MMvec. This study not only addresses the challenges in linking TPs to microbes but also introduces an innovative and effective approach for such investigations.Environmental Implication: Taxonomically diverse bacteria performing successive metabolic steps of the contaminant were firstly depicted in the environmental matrix.

Power quality enhancement of dual stage three phase grid integrated SPV system using sequential neural network-based algorithm

In this paper, a single-layer sequential neural network (SNN) control algorithm has been proposed for managing the power quality (PQ) in the double-stage three-phase grid-integrated solar photovoltaic (SPV) system. The proposed algorithm efficiently extracts the active and reactive fundamental components of the load current, with a higher convergence rate. The extracted reference currents are matched with the sensed source currents to generate gating pulses for the voltage source converter (VSC). To handle power quality issues in distribution networks such as harmonic imbalances, voltage fluctuations (sags/swells), and the need for active and reactive power compensation, the VSC is controlled very precisely. The control algorithm consists of two-stage processes. In the initial stage, the incremental conductance (INC) based DC-DC converter has been incorporated to extract the peak power from the SPV system. In the later stage SNN-based algorithm is used to control the VSC. A 100 KW solar PV system connected to the utility grid has been designed by using a three-leg VSC in the MATLAB/Simulink environment. The system performance has been evaluated by comparing it to existing techniques. The simulation results comfortably meet the requirements of the IEEE-519–2014 and IEEE 1547–2003 standards for utility grid connection.

SpatialDDLS: an R package to deconvolute spatial transcriptomics data using neural networks.

Spatial transcriptomics has changed our way to study tissue structure and cellular organization. However, there are still limitations in its resolution, and most available platforms do not reach a single cell resolution. To address this issue, we introduce SpatialDDLS, a fast neural network-based algorithm for cell type deconvolution of spatial transcriptomics data. SpatialDDLS leverages single-cell RNA sequencing data to simulate mixed transcriptional profiles with predefined cellular composition, which are subsequently used to train a fully connected neural network to uncover cell type diversity within each spot. By comparing it with two state-of-the-art spatial deconvolution methods, we demonstrate that SpatialDDLS is an accurate and fast alternative to the available state-of-the art tools. The R package SpatialDDLS is available via CRAN-The Comprehensive R Archive Network: https://CRAN.R-project.org/package=SpatialDDLS. A detailed manual of the main functionalities implemented in the package can be found at https://diegommcc.github.io/SpatialDDLS.

A Deep Learning Approach for Improving Two-Photon Vascular Imaging Speeds.

A potential method for tracking neurovascular disease progression over time in preclinical models is multiphoton fluorescence microscopy (MPM), which can image cerebral vasculature with capillary-level resolution. However, obtaining high-quality, three-dimensional images with traditional point scanning MPM is time-consuming and limits sample sizes for chronic studies. Here, we present a convolutional neural network-based (PSSR Res-U-Net architecture) algorithm for fast upscaling of low-resolution or sparsely sampled images and combine it with a segmentation-less vectorization process for 3D reconstruction and statistical analysis of vascular network structure. In doing so, we also demonstrate that the use of semi-synthetic training data can replace the expensive and arduous process of acquiring low- and high-resolution training pairs without compromising vectorization outcomes, and thus open the possibility of utilizing such approaches for other MPM tasks where collecting training data is challenging. We applied our approach to images with large fields of view from a mouse model and show that our method generalizes across imaging depths, disease states and other differences in neurovasculature. Our pretrained models and lightweight architecture can be used to reduce MPM imaging time by up to fourfold without any changes in underlying hardware, thereby enabling deployability across a range of settings.

VGN: Value Decomposition With Graph Attention Networks for Multiagent Reinforcement Learning.

Although value decomposition networks and the follow on value-based studies factorizes the joint reward function to individual reward functions for a kind of cooperative multiagent reinforcement problem, in which each agent has its local observation and shares a joint reward signal, most of the previous efforts, however, ignored the graphical information between agents. In this article, a new value decomposition with graph attention network (VGN) method is developed to solve the value functions by introducing the dynamical relationships between agents. It is pointed out that the decomposition factor of an agent in our approach can be influenced by the reward signals of all the related agents and two graphical neural network-based algorithms (VGN-Linear and VGN-Nonlinear) are designed to solve the value functions of each agent. It can be proved theoretically that the present methods satisfy the factorizable condition in the centralized training process. The performance of the present methods is evaluated on the StarCraft Multiagent Challenge (SMAC) benchmark. Experiment results show that our method outperforms the state-of-the-art value-based multiagent reinforcement algorithms, especially when the tasks are with very hard level and challenging for existing methods.