Classification Recognition Research Articles

A Hybrid Approach for Sports Activity Recognition Using Key Body Descriptors and Hybrid Deep Learning Classifier

This paper presents an approach for event recognition in sequential images using human body part features and their surrounding context. Key body points were approximated to track and monitor their presence in complex scenarios. Various feature descriptors, including MSER (Maximally Stable Extremal Regions), SURF (Speeded-Up Robust Features), distance transform, and DOF (Degrees of Freedom), were applied to skeleton points, while BRIEF (Binary Robust Independent Elementary Features), HOG (Histogram of Oriented Gradients), FAST (Features from Accelerated Segment Test), and Optical Flow were used on silhouettes or full-body points to capture both geometric and motion-based features. Feature fusion was employed to enhance the discriminative power of the extracted data and the physical parameters calculated by different feature extraction techniques. The system utilized a hybrid CNN (Convolutional Neural Network) + RNN (Recurrent Neural Network) classifier for event recognition, with Grey Wolf Optimization (GWO) for feature selection. Experimental results showed significant accuracy, achieving 98.5% on the UCF-101 dataset and 99.2% on the YouTube dataset. Compared to state-of-the-art methods, our approach achieved better performance in event recognition.

Heterogeneity in category recognition across the visual field.

Visual information emerging from the extrafoveal locations is important for visual search, saccadic eye movement control, and spatial attention allocation. Our everyday sensory experience with visual object categories varies across different parts of the visual field which may result in location-contingent variations in visual object recognition. We used a body, animal body, and chair two-forced choice object category recognition task to investigate this possibility. Animal body and chair images with various levels of visual ambiguity were presented at the fovea and different extrafoveal locations across the vertical and horizontal meridians. We found heterogeneous body and chair category recognition across the visual field. Specifically, while the recognition performance of the body and chair presented at the fovea were similar, it varied across different extrafoveal locations. The largest difference was observed when the body and chair images were presented at the lower-left and upper-right visual fields, respectively. The lower/upper visual field bias of the body/chair recognition was particularly observed in low/high stimulus visual signals. Finally, when subjects' performances were adjusted for a potential location-contingent decision bias in category recognition by subtracting the category detection in full noise condition, location-dependent category recognition was observed only for the body category. These results suggest heterogeneous body recognition bias across the visual field potentially due to more frequent exposure of the lower visual field to body stimuli.Significance Statement Our study reveals that visual object recognition exhibits notable variations across different visual field regions, with a pronounced bias in recognizing body images in the lower visual field. This heterogeneity in recognition performance suggests that the frequent exposure of certain visual field areas to specific object categories, such as bodies, influences our visual processing abilities. These findings highlight the importance of considering spatial attention and saccadic eye movements in understanding visual object recognition and have potential implications for designing more effective visual information displays and interfaces.

Modelling of Neutrosophic Set-Based k-Nearest Neighbors Classifier for Virus Pneumonia and COVID-19 Recognition

COVID19 otherwise called Severe Acute Respiratory Syndrome Corona virus-2 is an infectious illness. Another transmittable infection called Pneumonia is mainly attributable to infection because of bacteria in the alveoli of the lungs. Once a diseased lung tissue has infection, it elevates excretion in it. Specialists conduct health examinations and identify the patient through ultrasound, biopsy, or Chest X-ray of lungs to identify whether the patient has these diseases. Incorrect treatment, misdiagnosis, and if the disease was disregarded will result in the fatality. The development of Deep Learning and neutrosophic set (NS) supports the decision-making procedure of professionals to identify patients with this disease. NS is a prolongation of the fuzzy set and classical theories. The NS determines three memberships such as T, I and F. T, I, and F display the degree of truth, the false, and the indeterminacy membership, correspondingly. This enables a more nuanced representation of contradiction, uncertainty, and ambiguity within the dataset, allowing superior handling of imprecise and complex data. This study develops a new Deep learning with Neutrosophic Set-Based k-Nearest Neighbors Classifier for disease detection (DLNSKNN-DD) technique. The major purpose of the DLNSKNN-DD method is to identify the existence of virus pneumonia and COVID-19. In the DLNSKNN-DD technique, the feature extraction from the medical images is carried out by residual network (ResNet50v2). Moreover, the parameter tuning of the ResNetv2 model is done using Adadelta optimizer. The DLNSKNN-DD technique exploits NSKNN model for classification purposes. The performance evaluation of the DLNSKNN-DD algorithm can be assessed on medicinal image dataset. The experimental outcomes underlined the effectual recognition results of the DLNSKNN-DD technique on the identification of diseases

A deep learning identification method of tight sandstone lithofacies integrating multilayer perceptron and multivariate time series

Lithofacies classification and identification are of great significance in the exploration and evaluation of tight sandstone reservoirs. Existing methods of lithofacies identification in tight sandstone reservoirs face issues such as lengthy manual classification, strong subjectivity of identification, and insufficient sample datasets, which make it challenging to analyze the lithofacies characteristics of these reservoirs during oil and gas exploration. In this paper, the Fuyu oil formation in the Songliao Basin is selected as the target area, and an intelligent method for recognizing the lithophysics reservoirs in tight sandstone based on hybrid multilayer perceptron (MLP) and multivariate time series (MTS-Mixers) is proposed. Firstly, appropriate logging curve parameters are selected based on the contribution rate as the basis of the lithofacies intelligent discrimination. Second, preprocessing operations are performed on the logging dataset to ensure the quality of the experimental data. Finally, the MLP-MTS hybrid intelligent model is constructed by combining the powerful information extraction and classification recognition capabilities of the MLP and MTS models to complete the intelligent recognition of the petrography of tight sandstone reservoirs. The experimental results demonstrate that the recognition efficiency of MLP-MTS model for all kinds of lithofacies phases is more than 90%, which verifies the good applicability of deep learning model in solving the process of lithofacies phase recognition in reservoirs.

Unraveling the Interplay Between Memristive and Magnetoresistive Behaviors in LaCoO3/SrTiO3 Superlattice‐Based Neural Synaptic Devices

AbstractMemristors and magnetic tunnel junctions are showing great potential in data storage and computing applications. A magnetoelectrically coupled memristor utilizing electron spin and electric field‐induced ion migration can facilitate their operation, uncover new phenomena, and expand applications. In this study, devices consisting of Pt/(LaCoO3/SrTiO3)n/LaCoO3/Nb:SrTiO3 (Pt/(LCO/STO)n/LCO/NSTO) are engineered using pulsed laser deposition to form the LCO/STO superlattice layer, with Pt and NSTO serving as the top and bottom electrodes, respectively. The results show that both memristive and magnetoresistive properties can coexist without any compromise in performance, and the values of ROFF/RON and tunnel magnetoresistance (TMR) ratio are both improved by ≈1000% compared to a single‐period heterostructure. Notably, the Pt/(LCO/STO)5/LCO/NSTO device demonstrates superior multilevel storage performance, characterized by extended endurance, reliable retention, high ROFF/RON ratio, significant TMR ratio, and fundamental synaptic behaviors. Furthermore, density functional theory (DFT) is employed to calculate the changes in oxygen vacancies, affecting the overall energy bands and magnetic moments in the monolayer and multi‐periodic structures. Simulations using the handwritten digit recognition classification achieve the highest accuracy of 94.38%. These attributes suggest that the devices hold considerable promise for application in data storage and neuromorphic computing, offering a platform for high‐density neural circuits in intelligent electronic devices.

Cross-dataset micro-expression identification based on facial ROIs contribution quantification

Abstract In this work, we propose a novel cross-dataset micro-expression identification method based on facial regions of interest contribution quantification, where the training samples are from the source dataset and the test samples are from the target dataset. Specifically, a micro-expression video clip (i.e., a sample) is first sampled at intervals to capture an image sequence consisting of multiple consecutive video frames. Then, locate and crop the facial area in the image frame, and perform facial landmark detection on the cropped area. Next calculate the optical-flow field of the image sequence, and extract the improved Main Directional Mean Optical-flow feature. Finally, the learned group sparse model is used as the classifier to predict the label information of the unmarked target samples, so as to complete the category recognition of micro-expressions. Extensive cross-dataset comparative experiments on four spontaneous micro-expression datasets, CASME, CASME II, SMIC-HS, and SAMM, show that the recognition strategy proposed in this work is effective. Compared with several other advanced recognition methods, our method has better cross-dataset recognition performance, not only the classification accuracy is higher, but also the classification stability is better when faced with different target datasets and different micro-expression categories. It can be applied in many fields such as clinical diagnosis, interrogation, negotiation, and national security.

Novel Multimode Assay Based on Asymmetrically Competitive CRISPR and Raman Barcode Spectra for Multiple Hepatocellular Carcinoma Biomarkers Detection.

Commercial pregnancy test strips (PTS) possess the advantages of lower price, higher stability, and better repeatability and have been popularized to integrate with novel sensing strategies to detect other disease biomarkers, which accelerates the commercialization process of those novel sensing strategies. However, the current integration of novel sensing strategies into commercial PTS still faced the problems of insufficient quantification, low sensitivity, and lack of multiple detection capabilities. Hence, we proposed the concept of "visual classification recognition, spectral signal subdivision" for multiple hepatocellular carcinoma biomarkers (miRNA122 and miRNA233) detection with dual signals based on asymmetric competitive CRISPR (acCRISPR) and surface-enhanced Raman spectroscopy coupling with PTS, named the acCRISPR-PTS-SERS assay. In this assay, acCRISPR was used as a nonamplified cascaded signal amplification method to improve the sensitivity of detection. Two AuNPs-based core-shell Raman tags, each corresponding to different miRNA biomarkers, were used to achieve both visual recognition and spectral segmentation to enhance the quantification of PTS detection and the capability for multiple detection. Under the optimal conditions, the LOD for miRNA122 and miRNA223 were 10.36 and 4.65 fM, respectively. The sensitivity was enhanced by nearly 2 orders of magnitude. In the future, simultaneous hand-held detection for fingerprint barcodes of different cancers can be achieved with the assistance of a microfluidic chip and smartphone.

Deep learning models can predict violence and threats against healthcare providers using clinical notes

Violence, verbal abuse, threats, and sexual harassment of healthcare providers by patients is a major challenge for healthcare organizations around the world, contributing to staff turnover, distress, absenteeism, reduced job satisfaction, and worsening mental and physical health. To enable interventions prior to possible violent episodes, we trained two deep learning models to predict violence against healthcare workers 3 days prior to violent events for case and control patients. The first model is a document classification model using clinical notes, and the second is a baseline regression model using largely structured data. Our document classification model achieved an F1 score of 0.75 while our model using structured data achieved an F1 of 0.72, both exceeding the predictive performance of a psychiatry team who reviewed the same documents (0.5 F1). To aid in the explainability and understanding of risk factors for violent events, we additionally trained a named entity recognition classifier on annotations of the same corpus, which achieved an overall F1 of 0.7. This study demonstrates the first deep learning model capable of predicting violent events within healthcare settings using clinical notes, surpassing the first published baseline of human experts. We anticipate our methods can be generalized and extended to enable intervention at other hospital systems.

Adaptive optimization of electromyographic channels for intelligent prosthetic hands based on individual differences

Intelligent prosthetic hands typically require an increase in the number of acquisition channels to improve gesture recognition accuracy, resulting in increased device complexity and cost. However, there are individual differences in muscle strength, body mass index, and exercise habits. Electromyographic prosthetic hands currently use standardized electromyographic channel configurations, which lack adaptability to individual differences. To address these issues, this paper proposes the electrode configuration adaptive optimization algorithm, which enhances and integrates traditional genetic algorithms and simulated annealing algorithms, and implements adaptive optimization solutions for different subjects. Experimental results show that the optimization outcomes differ among different subjects. Compared to a single optimization algorithm, the proposed algorithm can adaptively optimize the electrode configuration based on individual differences while ensuring recognition effectiveness, retaining electrode channel information that significantly contributes to gesture classification recognition, and meeting the stable recognition of their motion intentions by different subjects.

Indigenously modified endoluminal vacuum-assisted closure therapy for post-operative gastrointestinal transmural defects: Case series and review of literature.

A gastrointestinal (GI) transmural defect is defined as a total rupture of the GI wall and these defects can be divided into three main categories, including perforation, leaks and fistulae. Recognition of the specific classification of the defect is important for choosing the best therapeutic modality. We present a case series of patients with gastrointestinal transmural defects which were managed with indigenously modified endoluminal vacuum-assisted closure.

An enhanced ResNet deep learning method for multimodal signal-based locomotion intention recognition

Exoskeletons designed for rehabilitation and movement assistance are essential for addressing the physiological issues of aging in the elderly. Recognizing locomotion intention is a preliminary step towards developing balance recovery functions, as many current commercial exoskeletons have limited capabilities in this regard. The primary aim of this research is to study locomotion intention recognition algorithms to advance balance recovery technologies. This paper proposes a deep learning method, which selects ResNet as the foundational network and utilizes DSADS and Mex datasets to derive general insights from different types of locomotion recognition, focusing on daily activities and exercise respectively. To optimize ResNet, a Multi-Source Data Fusion Module and the Convolutional Block Attention Module were integrated to enhance performance when sensor data is insufficient. Experiments demonstrate that training accuracy improved, and the loss function converged more rapidly. Additionally, the Adaptive Pooling Module and Adaptive Gated Network were added to improve network stability and accuracy, resulting in a 5 % increase in accuracy. The model utilizes a self-collected dataset comprising accelerometer, gyroscope, and force data, incorporating extensive falling data to enhance its applicability. It achieves 98 % accuracy, 0.98 precision, 0.99 recall, 0.98 F1 Score, and 0.008 loss value for the recognition of six locomotion intentions in the self-collected dataset, meeting both the number and accuracy of activity category recognition. It outperforms existing methods on multiple evaluation metrics. Analysis of lower limb motion data validates the feasibility of the algorithm and highlights its potential for future applications in lower limb exoskeletons.

The factors affecting aerobics athletes’ performance using artificial intelligence neural networks with sports nutrition assistance

This work aims to comprehensively explore the influencing factors of aerobics athletes’ performance by integrating sports nutrition assistance and artificial intelligence neural networks. First, a personalized assessment and analysis of athletes’ nutritional needs are conducted, collecting various data including fitness tests, physiological monitoring, and surveys to establish a personalized nutritional needs model for athletes. In order to gain a more comprehensive understanding of the characteristics and requirements of aerobic athletes, exercise data are integrated with nutritional data, and deep learning analysis is performed using neural network algorithms. Moreover, in terms of artificial intelligence technology, optimization algorithms such as ShuffleNet V3 and Inception V3 are employed based on the complexity and characteristics of aerobic exercise. Besides, a channel attention mechanism is introduced to enhance the model’s recognition accuracy. Lastly, a ShuffleNet V3-based aerobic exercise classification and recognition model is proposed. It achieves accurate classification and recognition of aerobic exercise by integrating exercise nutrition, ShuffleNet V3, and attention mechanisms. The results reveal that this model outperforms the Convolutional Neural Network (CNN) baseline algorithm on accuracy and F1 score. On the MultiSports dataset, the proposed model achieves an accuracy of 95.11%, surpassing other models by 2.66%. On the self-built dataset, the accuracy reaches 96.73%, outperforming other algorithms by 2.56%. This indicates that the proposed model demonstrates significant accuracy in aerobics movement classification recognition with sports nutrition assistance, contributing to a more comprehensive intersection of deep learning and sports science research.

Producing plankton classifiers that are robust to dataset shift

AbstractModern plankton high‐throughput monitoring relies on deep learning classifiers for species recognition in water ecosystems. Despite satisfactory nominal performances, a significant challenge arises from dataset shift, which causes performances to drop during deployment. In our study, we integrate the ZooLake dataset, which consists of dark‐field images of lake plankton (Kyathanahally et al. 2021a), with manually annotated images from 10 independent days of deployment, serving as test cells to benchmark out‐of‐dataset (OOD) performances. Our analysis reveals instances where classifiers, initially performing well in in‐dataset conditions, encounter notable failures in practical scenarios. For example, a MobileNet with a 92% nominal test accuracy shows a 77% OOD accuracy. We systematically investigate conditions leading to OOD performance drops and propose a preemptive assessment method to identify potential pitfalls when classifying new data, and pinpoint features in OOD images that adversely impact classification. We present a three‐step pipeline: (i) identifying OOD degradation compared to nominal test performance, (ii) conducting a diagnostic analysis of degradation causes, and (iii) providing solutions. We find that ensembles of BEiT vision transformers, with targeted augmentations addressing OOD robustness, geometric ensembling, and rotation‐based test‐time augmentation, constitute the most robust model, which we call BEsT. It achieves an 83% OOD accuracy, with errors concentrated on container classes. Moreover, it exhibits lower sensitivity to dataset shift, and reproduces well the plankton abundances. Our proposed pipeline is applicable to generic plankton classifiers, contingent on the availability of suitable test cells. By identifying critical shortcomings and offering practical procedures to fortify models against dataset shift, our study contributes to the development of more reliable plankton classification technologies.

Combining Endpoint Detection with a Convolutional Neural Network Classifier for the Automatic Recognition of Cardiac Arrhythmias in Electrocardiogram Signals

Combining Endpoint Detection with a Convolutional Neural Network Classifier for the Automatic Recognition of Cardiac Arrhythmias in Electrocardiogram Signals

Transfer Learning for Turkish Cuisine Classification

Thanks to developments in data-oriented domains like deep learning and big data, the integration of artificial intelligence with food category recognition has been a topic of interest for decades. The capacity of image classification to produce more precise outcomes in less time has made it a popular topic in computer vision. For the purpose of food categorization, three well-known CNN-based models—EfficientNetV2M, ResNet101, and VGG16—were fine-tuned in this research. Moreover, the pre-trained Vision Transformer (ViT) was used for feature extraction, followed by classification using a Random Forest (RF) algorithm. All the models were assessed on the TurkishFoods-15 dataset. It was found that the ViT and RF models were most effective in accurately capturing food images, with precision, recall, and F1-score values of 0.91, 0.86, and 0.88 respectively.

Influence of the graphene oxide-coated steel fiber on the microstructure optimization of UHPC

Benefiting from its superior mechanical properties, large specific surface area and abundant oxygen-containing functional groups, graphene oxide (GO) can generate nucleation and pore-infilling effects to optimize the microstructure of the cementitious composites, thus reinforcing the mechanical performance and durability of cement-based materials. However, due to the dispersion issues and relatively high cost, the widespread use of GO in the cement industry has not yet been realized. In the present study, we applied three different methods to coat industrial GO nanosheets on the steel fiber’s surface and reinforce the UHPC microstructure. The proposed coating method effectively disperses GO within the ultra-high-performance concrete (UHPC) matrix and targets its enhancement effects on the interfacial transition zone (ITZ), which has the weakest pore structure in UHPC. The result presents that the three-step dip method of GO coating on the steel fiber demonstrates higher effectiveness in reinforcing the microstructure of UHPC, the porosity of the matrix is 1.88 %, with a relative decrease of 37.1 ∼ 60.3 %. In addition, the average distance of ITZ after GO enhanced UHPC is about 3.7 µm, about 37.3 %-68.1 % shorter, and the ITZ porosity is 3.8 %, which is 15.6 %-65.5 % lower. With the assistance of the metal intrusion technique, backscattered electron characterization and deep learning-based analysis, the pore structure features of UHPC were extracted with a classification recognition accuracy of 88.5 %, and the pore intrusion volume inside the characteristic region of the UHPC matrix increased by 1.89–2.14 times. The extraction characteristics illustrated that the GO strengthening on steel fibers was more effective in ITZ, with an optimization efficiency of about 8.1 % higher than in other regions. We expect that the findings of this study could provide deep insights into GO modification for cementitious composite reinforcement applications and open new pathways for affordable alternatives to nano-enhanced composites.

Impact of Machine Learning Integration in Qur’anic Studies

The advancement in the field of computer science, especially in machine learning (ML), represents a flourishing innovation that carries great importance in the domain of education. The beneficial impact of ML can also be observed in the realm of Qur’anic studies, particularly in Arabic text recognition and recitation analysis. This paper presents a comprehensive analysis of 34+ published scholarly articles devoted to Qur’anic studies. This work explores the convergence of machine learning methodologies and Qur’anic studies, examining the innovative applications and methodologies for Arabic text and voice classification. The fusion of ML algorithms makes the work easy and accurate to analyze, interpret, and extract valuable insights from the sacred text. Subsequently, we delve deeper into the emergent field of ML algorithms like k-NN, ANN, BLSTM, MFCC, SVM, NB and DL approaches have been adapted for Qur’anic texts classification, recitation and recitation analysis on accuracy, speed, class recognition, response rate and biasness benchmark. This work covers a diverse range of applications, including automated Qur’anic exegesis and analysis of usage of Ahkam Al-Tajweed. The main contribution of the work is to provide insight into how ML facilitates in Arabic and Kufic textual analysis, linguistic subtleties, and thematic structures of the Qur’anic text. Using the deep learning approaches, the reciters, recitation style and of the Quranic text has also explained in the work.

Dynamical mode recognition of coupled flame oscillators by supervised and unsupervised learning approaches

Combustion instability in gas turbines and rocket engines, as one of the most challenging problems in combustion research, arises from the complex interactions among flames influenced by chemical reactions, heat and mass transfer, and acoustics. Identifying and understanding combustion instability is essential for ensuring the safe and reliable operation of many combustion systems, where exploring and classifying the dynamical behaviors of complex flame systems is a core task. To facilitate fundamental studies, the present work concerned dynamical mode recognition of coupled flame oscillators made of flickering buoyant diffusion flames, which have gained increasing attention in recent years but are not sufficiently understood. The time series data of flame oscillators were generated through fully validated reacting flow simulations. Due to the limitations of expertise-based models, a data-driven approach was adopted. In this study, a nonlinear dimensional reduction model of variational autoencoder (VAE) was used to project the high dimensional data onto a 2-dimensional latent space. Based on phase trajectories in the latent space, both supervised and unsupervised classifiers were proposed for datasets with and without well-known labeling, respectively. For labeled datasets, we established the Wasserstein-distance-based classifier (WDC) for mode recognition; for unlabeled datasets, we developed a novel unsupervised classifier (GMM-DTW) combining dynamic time warping (DTW) and Gaussian mixture model (GMM). Through comparing with conventional approaches for dimensionality reduction and classification, the proposed supervised and unsupervised VAE-based approaches exhibit a prominent performance across seven assessment metrics for distinguishing dynamical modes, implying their potential extension to dynamical mode recognition in complex combustion problems.

A review on depth perception techniques in organoid images

Organoids are an in vitro model that can simulate the complex structure and function of tissues in vivo. Functions such as classification, screening and trajectory recognition have been realized through organoid image analysis, but there are still problems such as low accuracy in recognition classification and cell tracking. Deep learning algorithm and organoid image fusion analysis are the most advanced organoid image analysis methods. In this paper, the organoid image depth perception technology is investigated and sorted out, the organoid culture mechanism and its application concept in depth perception are introduced, and the key progress of four depth perception algorithms such as organoid image and classification recognition, pattern detection, image segmentation and dynamic tracking are reviewed respectively, and the performance advantages of different depth models are compared and analyzed. In addition, this paper also summarizes the depth perception technology of various organ images from the aspects of depth perception feature learning, model generalization and multiple evaluation parameters, and prospects the development trend of organoids based on deep learning methods in the future, so as to promote the application of depth perception technology in organoid images. It provides an important reference for the academic research and practical application in this field.

Airborne LiDAR Point Cloud Classification Using Ensemble Learning for DEM Generation.

Airborne laser scanning (ALS) point clouds have emerged as a predominant data source for the generation of digital elevation models (DEM) in recent years. Traditionally, the generation of DEM using ALS point clouds involves the steps of point cloud classification or ground point filtering to extract ground points and labor-intensive post-processing to correct the misclassified ground points. The current deep learning techniques leverage the ability of geometric recognition for ground point classification. However, the deep learning classifiers are generally trained using 3D point clouds with simple geometric terrains, which decrease the performance of model inferencing. In this study, a point-based deep learning model with boosting ensemble learning and a set of geometric features as the model inputs is proposed. With the ensemble learning strategy, this study integrates specialized ground point classifiers designed for different terrains to boost classification robustness and accuracy. In experiments, ALS point clouds containing various terrains were used to evaluate the feasibility of the proposed method. The results demonstrated that the proposed method can improve the point cloud classification and the quality of generated DEMs. The classification accuracy and F1 score are improved from 80.9% to 92.2%, and 82.2% to 94.2%, respectively, by using the proposed methods. In addition, the DEM generation error, in terms of mean squared error (RMSE), is reduced from 0.318-1.362 m to 0.273-1.032 m by using the proposed ensemble learning.