Feature Recognition Research Articles

Enhancing depression recognition through a mixed expert model by integrating speaker-related and emotion-related features

The World Health Organization predicts that by 2030, depression will be the most common mental disorder, significantly affecting individuals, families, and society. Speech, as a sensitive indicator, reveals noticeable acoustic changes linked to physiological and cognitive variations, making it a crucial behavioral marker for detecting depression. However, existing studies often overlook the separation of speaker-related and emotion-related features in speech when recognizing depression. To tackle this challenge, we propose a Mixture-of-Experts (MoE) method that integrates speaker-related and emotion-related features for depression recognition. Our approach begins with a Time Delay Neural Network to pre-train a speaker-related feature extractor using a large-scale speaker recognition dataset while simultaneously pre-training a speaker’s emotion-related feature extractor with a speech emotion dataset. We then apply transfer learning to extract both features from a depression dataset, followed by fusion. A multi-domain adaptation algorithm trains the MoE model for depression recognition. Experimental results demonstrate that our method achieves 74.3% accuracy on a self-built Chinese localized depression dataset and an MAE of 6.32 on the AVEC2014 dataset. Thus, it outperforms state-of-the-art deep learning methods that use speech features. Additionally, our approach shows strong performance across Chinese and English speech datasets, highlighting its effectiveness in addressing cultural variations.

Online monitoring of a high-power CW laser optical mirror based on infrared and visible image features visualization

In high-power CW (continuous wave) laser systems, laser damage to optical mirrors often occurs, and it is urgent to develop timely and effective monitoring technologies to avoid laser damage or emergency stop at the early damage stage. In this paper, an online monitoring method based on the combination of infrared and visible images is proposed. The proposed method is based on the feature recognition of infrared and visible images. After setting an appropriate threshold, a decision model can be employed to enable real-time monitoring and health status evaluation of optical mirrors. Here, the thresholds for abnormal and dangerous temperatures were established at 100°C and 200°C, respectively. The results show that the temperature of the optical mirror would significantly increase after damage, with the surface temperature of the element reaching up to 250°C. At the same time, the damaged area could be captured by a visible camera and the changes in gray values were displayed in the visible image. Online monitoring of the health status of optical elements can be achieved by assessing changes in infrared image temperature, visible image spot position, and area. This monitoring method serves as an early warning method for potential optical elements damage.

The ENaC taste receptor’s perceived mechanism of mushroom salty peptides revealed by molecular interaction analysis

The ENaC receptor acts as a taste receptor to recognize and perceive salty substances. This study explored the mechanisms by which the ENaC taste receptor recognizes and binds mushroom-derived salty peptides using molecular interaction and molecular simulation. The three subunits α, β, and γ of the ENaC taste receptor (SCNN1α, SCNN1β, and SCNN1γ) showed different recognition characteristics for the salty peptide. The salty peptide binding to the SCNN1α receptor was an entropy-driven reaction, while to SCNN1β and SCNN1γ was an enthalpy-driven reaction. With the salty peptide spatial resistance increasing, salty peptides bind to the ENaC taste receptor shifted from receptor pockets binding to receptor surface binding, with salty octapeptide ESPERPFL preferentially binding to amino acid residues in the receptor pockets 2, 3, and 4, salty nonapeptide KSWDDFFTR and decapeptide RIEDNLVIIR binding to amino acid residues in the pockets 2, 4 and on the surface of the receptor, and salty undecapeptide GQEDYDRLRPL preferentially binding to the atoms on the surface of the receptor. Receptor extracellular arginine, glutamate, aspartate, and lysine residues were the critical amino acid residues recognized to bind salty peptides. The salty peptide-ENaC receptor binding complex was stable around 0.3 nm, and the tight and multisite binding was the main reason the ENaC receptor sensed the salty peptide, enabling it to exert its taste effect. This research can provide a theoretical basis for understanding the taste properties of salty peptides recognized and perceived by the ENaC taste receptor.

Manufacturing Feature Recognition With a Sparse Voxel-Based Convolutional Neural Network

Abstract Automated manufacturing feature recognition is a crucial link between computer-aided design and manufacturing, facilitating process selection and other downstream tasks in computer-aided process planning. While various methods such as graph-based, rule-based, and neural networks have been proposed for automatic feature recognition, they suffer from poor scalability or computational inefficiency. Recently, voxel-based convolutional neural networks have shown promise in solving these challenges but incur a tradeoff between computational cost and feature resolution. This paper investigates a computationally efficient sparse voxel-based convolutional neural network for manufacturing feature recognition, specifically, an octree-based sparse voxel convolutional neural network. This model is trained on a large-scale manufacturing feature dataset, and its performance is compared to a voxel-based feature recognition model (FeatureNet). The results indicate that the octree-based model yields higher feature recognition accuracy (99.5% on the test dataset) with 44% lower graphics processing unit (GPU) memory consumption than a voxel-based model of comparable resolution. In addition, increasing the resolution of the octree-based model enables recognition of finer manufacturing features. These results indicate that a sparse voxel-based convolutional neural network is a computationally efficient deep learning model for manufacturing feature recognition to enable process planning automation. Moreover, the sparse voxel-based neural network demonstrated comparable performance to a boundary representation-based feature recognition neural network, achieving similar accuracy in single-feature recognition without having access to the exact 3D shape descriptors.

Comparison of brain activity metrics in Chinese and Russian students while perceiving information referencing self or others

Neurocomputing technology is a field of interdisciplinary research and development widely applied in modern digital medicine. One of the problems of neuroimaging technology is the creation of methods for studying human brain activity in socially oriented conditions by using modern information approaches. The aim of this study is to develop a methodology for collecting and processing psychophysiological data, which makes it possible to estimate the functional states of the human brain associated with the attribution of external information to oneself or other people. Self-reference is a person’s subjective assessment of information coming from the external environment as related to himself/herself. Assigning information to other people or inanimate objects is evaluating information as a message about someone else or about things. In modern neurophysiology, two approaches to the study of self-referential processing have been developed: (1) recording brain activity at rest, then questioning the participant for self-reported thoughts; (2) recording brain activity induced by self-assigned stimuli. In the presented paper, a technology was tested that combines registration and analysis of EEG with viewing facial video recordings. The novelty of our approach is the use of video recordings obtained in the first stage of the survey to induce resting states associated with recognition of information about different subjects in later stages of the survey. We have developed a software and hardware module, i. e. a set of related programs and procedures for their application consisting of blocks that allow for a full cycle of registration and processing of psychological and neurophysiological data. Using this module, brain electrical activity (EEG) indicators reflecting individual characteristics of recognition of information related to oneself and other people were compared between groups of 30 Chinese (14 men and 16 women, average age 23.2 ± 0.4 years) and 32 Russian (15 men, 17 women, average age 22.1 ± 0.4 years) participants. We tested the hypothesis that differences in brain activity in functional rest intervals between Chinese and Russian participants depend on their psychological differences in collectivism scores. It was revealed that brain functional activity depends on the subject relevance of the facial video that the participants viewed between resting-state intervals. Interethnic differences were observed in the activity of the anterior and parietal hubs of the default-mode network and depended on the subject attribution of information. In Chinese, but not Russian, participants significant positive correlations were revealed between the level of collectivism and spectral density in the anterior hub of the default-mode network in all experimental conditions for a wide range of frequencies. The developed software and hardware module is included in an integrated digital platform for conducting research in the field of systems biology and digital medicine.

Swimming posture recognition using inertial sensors and CNN-SVM: Unveiling the cellular molecular biomechanics nexus

The biomechanical mechanisms of swimming involve a number of aspects. The forces exerted by muscles during different swimming postures are crucial. These muscle contractions and relaxations follow specific biomechanical principles. This work aims to develop a swimming posture recognition system based on inertial sensors and a Convolutional Neural Network-Support Vector Machine (CNN-SVM) to improve the accuracy and real-time performance of posture recognition. First, an inertial sensor system to be worn on swimwear is designed to collect three-axis motion data, including acceleration, angular velocity, and magnetometer readings. The collected data are then preprocessed through denoising, normalization, and feature extraction steps to ensure high-quality input data. Next, a Convolutional Neural Network (CNN) is constructed to automatically extract high-level features from the preprocessed sensor data. The CNN model, through multi-layer convolution and pooling operations, effectively captures the spatiotemporal patterns in the motion data, extracting highly distinguishable features for posture recognition. To further improve the model’s classification performance, a Support Vector Machine (SVM) classifier is applied based on the CNN model. Specifically, CNN is responsible for feature extraction, while the SVM handles the final posture classification. Cross-validation is used to train and validate the model, assessing its performance. Experimental results show that the model achieves a 95% accuracy rate on the training dataset and maintains an accuracy rate above 93% on the test dataset. The system can accurately and in real-time recognize various swimming postures, including freestyle, breaststroke, backstroke, and butterfly. The recognition accuracy for all four swimming styles exceeds 91%. Understanding these biomechanical mechanisms helps in improving the accuracy of the recognition system. In summary, the proposed method for swimming posture recognition based on inertial sensors and CNN-SVM has significant advantages in accuracy and real-time performance. It allows for better interpretation of the sensor data and more precise identification of different postures. The high accuracy and generalization ability of the proposed system suggest that it can effectively capture and analyze the biomechanical nuances of swimming, providing valuable insights for swimming training and performance evaluation, and opening up new avenues for intelligent sports monitoring. evaluation.

P0382 Identifying atypical disease phenotypes in paediatric Ulcerative Colitis: Insights from the real world PIBD-SETQuality cohort

Abstract Background The heterogeneity of paediatric ulcerative colitis (P-UC) can complicate diagnosis and prompt start of optimal induction treatment, which is crucial for adequate disease control. Recognizing and understanding atypical disease patterns in P-UC will decrease misclassification and contribute to the development of standardized classification tools. This study aims to provide insight in the real world prevalence of atypical phenotypes in newly diagnosed P-UC patients. Methods Data were collected from the Paediatric Inflammatory Bowel Diseases Network for Safety, Efficacy, Treatment and Quality improvement of care (PIBD-SETQuality) cohort, a multicenter, international, prospective, observational study including therapy naive children <19 years old with IBD. Patients diagnosed with UC between January 1st 2017 and June 14th 2024 were enrolled. Diagnostic criteria and macroscopical atypical disease were based on the validated PIBD-classes algorithm by revised Porto Criteria1. The Mann-Whitney U test and Chi-square tests were used to compare, respectively, numerical and categorical data between patient groups. Results Baseline data of 197 P-UC patients were included. At diagnosis, 68% of the patients presented with pancolitis. Atypical disease occurred in 35.0% (69/197) of the patients of which relative rectal sparing was most common (Table 1). Aspecific upper gastrointestinal inflammation occurred in 10.6% (17/161) of the patients, of whom 6 also exhibited atypical disease phenotypes. Approximately 1 out of 10 patients with atypical disease presented with multiple atypical phenotypes. Patients with atypical disease did not show significant differences compared to patients without atypical disease in time to diagnosis (107 days vs. 104 days, P=0.19), absolute PUCAI score (40 [IQR 30-55] vs. 35 [IQR 25-58.75], p=0.10), initial induction therapy and need for treatment intensification during induction therapy (8 vs. 4 patients, p=0.22). Conclusion Atypical disease presentations are common in P-UC in this real world cohort, as is the presence of combined atypical disease features. Recognition of atypical disease features in P-UC, and understanding their impact on disease course, are crucial for prompt start of optimal tailored therapy. In this cohort, the presence of atypical disease did not result in a diagnostic delay or treatment intensification during induction period.

Rolling Bearing Fault Diagnosis Based on CBAM-WDCNN

Rolling bearing fault diagnosis methods based on convolutional neural networks (CNNs) have demonstrated excellent capabilities in processing complex nonlinear signals. Among them, the one-dimensional convolutional neural network greatly accelerates the signal processing process by directly processing the one-dimensional time-series signal and simplifying the input of the model. However, the traditional 1D convolutional neural network has problems such as poor feature recognition. In this paper, we propose a model based on the fusion of the attention mechanism CBAM and the deep convolutional neural network with a wide kernel in the first layer (WDCNN), which emphasizes the key features and suppresses the irrelevant noises, and significantly improves the model's recognition ability for fault features. The improved method is compared with other diagnostic algorithms to verify its higher accuracy and more stable training process.

CFNet:CSI compression feedback network based on WiFi sensing

Abstract To solve the problem of huge channel state information (CSI)data caused by a large number of antennas in large-scale MIMO systems,a CSI compressed feedback network model (CFNet)with encoder-decoder structure is proposed in this paper. Specifically, the encoder efficiently compresses the CSI data to reduce the burden of data transmission and storage. A key component, the Asymmetric Expansion Convolution (AEC) module, combines asymmetric convolution and dilated convolution to extract feature information from different directions, thereby increasing the model's receptive field and minimizing information loss. The compressed CSI is fed into a Gated Recurrent Unit (GRU), which captures long-term dependencies in the CSI sequence using the encoder's output and the hidden state of the upper window. Additionally, on the decoder side, the proposed Asymmetric Transpose Convolution reconstructs the compressed CSI data, enhancing the network's ability to capture and express CSI features through two complementary convolution operations, ultimately reducing communication costs. Thanks to the proposed encoder and decoder, CFNet can accurately identify fine-grained features of human activities recognition(HAR) while leveraging compressed sensing techniques, further improving data processing efficiency and recognition accuracy. The experiments conducted on two public datasets demonstrate that the evaluation metrics of the proposed network outperform those of advanced methods in the relevant field.

Oral biomechanical feature recognition method and system for vocal pronunciation resonance

This paper discusses a vocal pronunciation resonance recognition method based on oral biomechanical features, aiming to collect the movement data inside the oral cavity through high-precision sensors, and combine support vector machines and convolutional neural networks to achieve precise recognition and optimization of pronunciation resonance features. First, by using high-precision sensors such as tongue position sensors and soft palate movement sensors, key biomechanical data such as tongue position, oral morphology and soft palate movement of singers during pronunciation are obtained in real-time. Then, feature extraction and analysis are performed to extract biomechanical features such as movement amplitude, frequency, and speed of oral organs, and a resonance feature recognition model is constructed using support vector machines and convolutional neural networks to recognize and optimize the resonance area during pronunciation. Finally, this paper designs and develops an automated recognition system that can provide real-time feedback on the singer’s pronunciation data, and provide personalized training suggestions based on the recognition results to optimize the pronunciation resonance effect. The experimental results show that the resonance recognition method based on biomechanical data can significantly improve the accuracy and personalization of pronunciation, and help improve the efficiency and scientificity of vocal training. The system has good stability, with a response time of less than 500 ms and a CPU (Central Processing Unit) usage rate of no more than 70%. This method provides effective technical support for the digitalization and personalized optimization of vocal pronunciation, and has high practical value and promotion potential.

Application of Convolutional Neural Networks and Recurrent Neural Networks in Food Safety.

This review explores the application of convolutional neural networks (CNNs) and recurrent neural networks (RNNs) in food safety detection and risk prediction. This paper highlights the advantages of CNNs in image processing and feature recognition, as well as the powerful capabilities of RNNs (especially their variant LSTM) in time series data modeling. This paper also makes a comparative analysis in many aspects: Firstly, the advantages and disadvantages of traditional food safety detection and risk prediction methods are compared with deep learning technologies such as CNNs and RNNs. Secondly, the similarities and differences between CNNs and fully connected neural networks in processing image data are analyzed. Furthermore, the advantages and disadvantages of RNNs and traditional statistical modeling methods in processing time series data are discussed. Finally, the application directions of CNNs in food safety detection and RNNs in food safety risk prediction are compared. This paper also discusses combining these deep learning models with technologies such as the Internet of Things (IoT), blockchain, and federated learning to improve the accuracy and efficiency of food safety detection and risk warning. Finally, this paper mentions the limitations of RNNs and CNNs in the field of food safety, as well as the challenges in the interpretability of the model, and suggests the use of interpretable artificial intelligence (XAI) technology to improve the transparency of the model.

Feature recognition: A unified framework for fault diagnosis and prognosis of discrete event systems

Feature recognition: A unified framework for fault diagnosis and prognosis of discrete event systems

Reliable Raman spectroscopic authentication of ST25 rice through representative sampling of bulk grains and effective recognition of dissimilar spectral features by two-trace two-dimensional correlation analysis

Reliable Raman spectroscopic authentication of ST25 rice through representative sampling of bulk grains and effective recognition of dissimilar spectral features by two-trace two-dimensional correlation analysis

Management and investigation of high myopia in infants and young children

A growing array of behavioral, pharmacological, and optical interventions are now available. These interventions have typically been tested on children aged 6 to 16 who exhibit a typical pattern of myopia onset and progression. In such instances, myopia begins to develop after the age of 5 or 6 and progresses variably until stabilizing during the teenage years or early adulthood. However, there is a much rarer pattern of myopia where children exhibit very high levels of myopia in infancy or early childhood, often exceeding the ‐6D threshold of high myopia. In these cases, numerous diagnostic, health, and visual challenges must be addressed. Managing these issues may initially take higher clinical priority than treating observed progression.High myopia in infants also poses challenges for optical correction to ensure normal visual development and prevent amblyopia. High myopia combined with significant behavioral or developmental issues can be particularly difficult, as uncorrected myopia or contact lens intolerance can exacerbate the behavioral issues of these children. The visual impairment from uncorrected refractive errors can negatively affect both their behavior and development. For instance, the inability to see faces and read emotions can be profoundly isolating, sometimes described as "visual autism" due to the impact of visual isolation.The aim of this presentation is to provide an overview of this complex and multifaceted topic, offering recommendations for the identification, investigation, and management of high myopia in infants and young children. These patients may initially present to primary eye care services where specialist multidisciplinary investigation and management are not available. In such settings, a thorough clinical history, comprehensive examination, and recognition of the unusual features of these children are critical to facilitate prompt and appropriate referral.This patient group is commonly referred to hospital‐based eye clinics. Here, the initial challenge is to correctly identify cases that require detailed investigation and to ensure that the appropriate clinical and diagnostic services are utilized. Management can be wide‐ranging, including advanced ocular imaging, electrophysiology, genetic investigations, involvement of pediatricians and clinical geneticists, genetic counseling, optical correction strategies, and appropriate follow‐up protocols for high‐risk children.

NOTIFICATION: The Role of Pathologists in Recognition of Morphologic and Biologic Features of Genetically Mutated Breast Cancer

NOTIFICATION: The Role of Pathologists in Recognition of Morphologic and Biologic Features of Genetically Mutated Breast Cancer

Incorporating image representation and texture feature for sensor-based gymnastics activity recognition

Incorporating image representation and texture feature for sensor-based gymnastics activity recognition

Feature selection through adaptive sparse learning for scene recognition

Scene recognition is an important and challenging task in the field of computer vision. Current research typically focuses on local features in scene images by utilizing pretrained convolutional neural networks (CNN) to obtain a feature dictionary. However, these local features often contain similar features to other scene categories, leading to feature confusion and subsequently low accuracy in scene recognition. Therefore, focusing on local features for scene recognition is controversial. In contrast to existing works, we consider extracting common features within the same category from scene images and using them as discriminative features between different categories. We propose a scene recognition method based on adaptive sparse learning feature selection. This method leverages sparse learning to distinguish the contribution of each feature in the deep features to the classification task, aiming to construct a salient feature dictionary that describes the importance of features in scene images. Subsequently, an adaptive weight optimization method is employed through alternating updates to automatically adjust feature weights, enabling the selection of common features within the same scene category from the compact dictionary. Experimental results on multiple benchmark scene recognition datasets demonstrate that our proposed method is superior to state-of-the-art methods.

Application of thermal radiation imaging detection based on spatial optical characteristics in motion attitude recognition

Application of thermal radiation imaging detection based on spatial optical characteristics in motion attitude recognition

Research on wood defects feature imbalance optimization and recognition

Research on wood defects feature imbalance optimization and recognition

Facial Feature Recognition with Multi-task Learning and Attention-based Enhancements

Facial Feature Recognition with Multi-task Learning and Attention-based Enhancements