High Recognition Rate Research Articles (Page 1)

This paper presents a dual-structured convolutional neural network (CNN) for image classification, which integrates two parallel branches: CNN-A with spatial attention and CNN-B with channel attention. The spatial attention module in CNN-A dynamically emphasizes discriminative regions by aggregating channel-wise information, while the channel attention mechanism in CNN-B adaptively recalibrates feature channel importance. The extracted features from both branches are fused through concatenation, enhancing the model’s representational capacity by capturing complementary spatial and channel-wise dependencies. Extensive experiments on a 12-class image dataset demonstrate the superiority of the proposed model over state-of-the-art methods, achieving 98.06% accuracy, 96.00% precision, and 98.01% F1-score. Despite a marginally longer training time, the model exhibits robust convergence and generalization, as evidenced by stable loss curves and high per-class recognition rates (>90%). The results validate the efficacy of dual attention mechanisms in improving feature discrimination for complex image classification tasks.

IntroductionEarly diagnosis of non-keratinizing nasopharyngeal carcinoma (NK-NPC) is a significant clinical challenge. This study assessed combined antibodies and built a nomogram for more accurate NK-NPC screening.MethodsClinical data of 1330 individuals at high risk of nasopharyngeal carcinoma (NPC) from June 2021 to December 2024 were collected retrospectively. They were randomly divided into a training set (n = 930) and a validation set (n = 400) at a ratio of 7:3. The training set was further divided into the NK-NPC group and the non-NK-NPC group. Univariate and multivariate analyses were used to screen for risk factors of cancer, based on which a risk prediction nomogram model was constructed. The predictive performance of the model was evaluated using indicators such as the area under the receiver operating characteristic curve (AUC), integrated discrimination improvement (IDI), decision curve analysis (DCA), and Youden index. Additionally, an external validation set (cases from January-May 2025 at the same hospital) further assessed the model.ResultsSex, EBNA1-IgA, VCA-IgA, and Rta-IgG were independent risk factors for NK-NPC in high-risk populations (P < 0.05). The validation results of the nomogram model constructed based on the above factors showed that the AUC values of the receiver operating characteristic (ROC) curves in the training set and validation set were 0.898 and 0.963. Decision curve analysis showed that the net benefit value of this model was higher than that of the traditional model within the threshold probability range of 10% to 60%. The external validation results showed that the sensitivity of the model was 100% and the specificity was 87.8%.ConclusionThe NK-NPC prediction nomogram model constructed in this study has a high recognition rate and good calibration. It can serve as an effective prediction tool for NK-NPC in high-risk populations of nasopharyngeal carcinoma.

Optoelectronic neuromorphic devices, which mimic the functionalities of the human eye and brain neural systems, have attracted significant interest for enabling highly energy-efficient computing systems for next-generation artificial intelligence applications. However, several key challenges persist, including a limited dynamic range for light-induced synaptic weights, low optical photogain, lack of spectral selectivity, and incompatibility with heterogeneous integration. Addressing these issues is essential for unlocking the full potential of optosynaptic devices in advanced AI systems. In this work, we develop artificial solar-blind optoelectronic synaptic devices exhibiting high pattern recognition rates (>92%) in neural network training using ultrawide-bandgap amorphous gallium oxide (a-GaOx) thin-film transistors (TFTs). The device functions through deep ultraviolet (DUV) optically induced potentiation and gate-terminal electrical depression processes, exhibiting excellent plasticity and a wide conductance weight update range. This performance is attributed to its superior TFT switching characteristics, strong DUV photoresponse with a dynamic gain exceeding 108, and UV-triggered persistent photoconductivity (PPC) lasting over 1000 s. Moreover, the device can be fabricated at a low temperature of 450 °C, ensuring compatibility with the complementary metal-oxide-semiconductor (CMOS) back-end-of-line (BEOL) process.

Investigation into the biochemical properties of loquat fruits and their optical properties of hyperspectral imaging spectral correlation during the progression of postharvest fungal infection.

This study investigated the competence of health professionals at all levels of expertise, medical students, residents, and cardiology consultants, in diagnosing electrocardiography (ECG) abnormalities, with the primary emphasis on professional experience, gender, age, and years in practice. ECG remains a foundational method to diagnose cardiac functions; however, there are disparities among healthcare providers’ interpretation skills. Typically, previous studies have examined narrow cohorts, which limits insights into ECG interpretative abilities across the healthcare spectrum. In this study, 72 participants completed a comprehensive ECG interpretation assessment involving common and complex abnormalities, while cognitive engagement was tracked via eye-tracking metrics to evaluate interpretation processes. Data were collected from the professionals using surveys and eye-tracking technology to evaluate ECG interpretation skills. Informed consent was obtained from the participants, and the Qatar Biomedical Research Institute’s IRB: QBRI-IRB-2020-01-009 approved the ethical standards followed. The results showed significant differences in diagnostic accuracy based upon expertise. Cardiology consultants showed the highest accuracy, while younger participants between 26 and 30 years outperformed the older groups, and males generally had higher recognition rates overall compared to females. The eye-tracking analysis highlighted prolonged engagement with complex ECG segments, which suggested interpretative challenges. Importantly, targeted educational interventions improved participants’ recognition accuracy notably. These findings underscore tailored training programs’ potential to enhance ECG interpretation skills, and ultimately improve diagnostic accuracy and patient outcomes. Further, the Random Forest model achieved 81% accuracy in classifying long versus short ECG readings. This study highlights critical gaps in ECG proficiency and advocates for continuous, demographics-based training enhancements. Future research should prioritize longitudinal studies to evaluate targeted training interventions’ effect on enhancing ECG interpretation skills across a diverse range of healthcare professionals.

Stroke remains one of the most significant health concerns in the world that not only results in deaths but also in disabilities and the earlier a patient is diagnosed and treated, the better are the outcomes. Machine learning (ML) and deep learning (DL) are the components of Artificial Intelligence (AI) that have not yet reached their full potential in enhancing the diagnosis of the stroke because of gradually emerging medical applications. In the review, the functioning of AI technologies in stroke care was investigated with the approach to medical imaging methods as well as clinical decision support systems/symptom recognition tools and predictive models as concerns electronic health records (EHR). AI-enhanced medical imaging instruments have a high rate of ischemic and hemorrhagic stroke recognition, as well as the large vessel occlusion (and the volume of infarct core and penumbra). The same is true of medical imaging tools that can match the capacity of expert radiologists. The mobile health applications along with wearable devices are associated with real-time symptom monitoring that ensures early health intervention especially to patients who reside in isolate or underprivileged settings. The advantages of fastness, accuracy, and distant accessibility are continuously undermined by issues of bias in algorithms, along with the data quality, and also clinical integration and regulatory clearance procedures. AI holds significant promise in changing how stroke is diagnosed and treated but there is still a long way to get there and that will entail an ethical application and a powerful validation and that includes working jointly with practitioners and researchers and policymakers on behalf of an evenhanded and successful outcome.

In this paper, the single-mode vortex beam is used to illuminate targets of different shapes, and the targets are recognized using machine learning algorithms based on the orbital angular momentum (OAM) spectral information of the echo signal. We innovatively utilize three neural networks—multilayer perceptron (MLP), convolutional neural network (CNN) and residual neural network (ResNet)—to train extensive echo OAM spectrum data. The trained models can rapidly and accurately classify the OAM spectrum data of different targets’ echo signals. The results show that the residual network (ResNet) performs best under all turbulence intensities and can achieve a high recognition rate when Cn2=1×10−13 m−2/3. In addition, even when the target size is η=0.3, the recognition rate of ResNet can reach 97%, while the robustness of MLP and CNN to the target size is lower; the recognition rates are 91.75% and 91%, respectively. However, although the recognition performance of CNN and MLP is slightly lower than that of ResNet, their training time is much lower than that of ResNet, which can achieve a good balance between recognition performance and training time cost. This research has a promising future in the fields of target recognition and intelligent navigation based on multi-dimensional information.

The potential to decode handwriting trajectories from brain signals has yet to be fully explored in clinical brain‐computer interfaces (BCIs). Here, intracortical neural signals are recorded from a paralyzed individual during attempted handwriting of complex characters. An innovative decoding framework is introduced to address both shape and temporal distortions between neural activity and movement, effectively resolving the misalignment issue commonly encountered in clinical BCIs due to the lack of accurate movement labels. The results demonstrated the reconstruction of highly accurate and human‐recognizable handwriting trajectories, significantly outperforming conventional methods. Furthermore, the new framework enabled effective multi‐day data fusion, leading to additional improvements in trajectory quality. By employing a dynamic time warping approach to translate trajectories into text, a recognition rate up to 91.1% is achieved within a 1000‐character database. Additionally, the framework is applied to reconstruct single‐trial trajectories of English letters using a previously published dataset, achieving similarly high recognition rates. Collectively, these findings present a novel BCI decoding scheme capable of accurately reconstructing handwriting trajectories, demonstrating its applicability to both alphabetic and logographic brain‐to‐text translation. This approach has the potential to revolutionize communication for individuals with motor impairments by enabling accurate brain‐to‐text translation across diverse languages.

Human’s facial expressions and emotions have direct impact on their action and decision-making abilities. Basic CNN models are complexity of speeding up the operation to minimize the complexity. In this paper, we have proposed a Deep Convolutional Neural Networks along with Bi-Long Short Term Memory, which is followed by a single and cross-fusion attention mechanism for gathering both spatial and channel information from feature vector maps. Piecewise Cubic Polynomial and linear activation function was used to speed up Interactive Learning Information (ILI). Global Average Pooling (GAP) computes weights for feature vector maps; softmax classifier is used to classify images into 7 classes based on the expression present on the input images. The proposed model’s performance was compared with benchmarking methods like NGO-BiLSTM, ICNN-BiLSTM and HCNN-LSTM. The proposed model resulted with better accuracy than other methods with 82.89%, 96.78%, 95.78%, and 95.87% on FER 2013, CK+, RAF-DB and JAFFE datasets and also resulted in lower False Recognition Rate (FAR) of 7.23%, 1.42%, 1.96% and 1.78% on all four datasets respectively. The proposed model has performed well than other benchmarking models with high Genuine Recognition Rate (GAR) of 88.57% on FER2013, 97.23% on CK+, 96.87% on RAF-DB and 96.32% on JAFFE datasets respectively.

The body serves as a vital medium for emotional expression. In this study, we aimed to extend the Bochum Emotional Stimulus Set (BESST), which comprises 565 body pictures depicting expressions of neutral emotions and six basic emotions (happiness, sadness, fear, anger, disgust, and surprise). This study included two tasks. One hundred college students (Mage = 18.82; SDage = 0.89) underwent the emotion recognition task of pictures (choose one out of seven emotions) and 103 college students (Mage = 19.83; SDage = 0.84) underwent the rating task of emotional dimensions (motivation, valence, arousal, and dominance; rated on a scale of 1-9). The results showed that: (1) On average, neutrality had the highest recognition rate (94%), followed by sadness (89.04%) and happiness (81.26%), anger (74.03%), disgust (69.61%), fear (67.97%), and surprise (64.45%). (2) The main effects of emotional type in ratings of four dimensions were all significant. Specifically, fear, disgust, and sadness have similar emotional characteristics (withdrawal-motivation, high arousal, low pleasure, and low dominance), and emotions with higher biological relevance (such as disgust) have a greater degree of emotional characteristics. Happiness and anger share similar emotional characteristics, including approach motivation, high arousal, and dominance; however, they represent opposite ends of the valence dimension. (3) There were gender differences in motivation dimension ratings of surprised pictures and arousal dimension ratings of neutral pictures. We provided the recognition rate and rating values for each of the four dimensions for each picture. This set serves as a valuable resource for cross-cultural research and behavioral or clinical studies related to emotional disorders.

The mechanism of chunk restructuring in the memory superiority effect of Insight: Dissociating the roles of decomposition and composition.

Self-powered optoelectronic synaptic devices have garnered significant attention due to their ultra-low energy consumption and self-rectification properties. However, the mechanism of mimicking their inhibitory behaviors remains unclear, presenting a challenge in attempting to realize optically inhibitory behaviors. This study fabricates formamidinium lead iodide perovskite-based synaptic devices that exhibit self-powered-optical potentiation and electrical inhibition behaviors. The mechanism underlying the inhibitory behaviors is argued to be the defect trap at room temperature and iodine ion migration at lower temperatures. Considering the optical potentiation behaviors and inhibitory mechanism clarified here, ethanediamine dihydroiodide is incorporated into the perovskite layer to regulate the synaptic behaviors. Impressively, this additive results in a shift of the self-powered-optical potentiation to its inhibition. First-principles calculations reveal that an increase of iodide vacancy formation energies facilitates this transformation by possibly modulating the carrier trap and ion migration behaviors. Additionally, the optically excitatory and optically inhibitory synaptic behaviors of the integrated systems with and without EDADI are exploited to implement MINIST and CIFAR-10 recognition tasks and achieve the high recognition rates of 97.95% and 77.36%, respectively. This work significantly advances the understanding of mimicking self-powered optically inhibitory synaptic behaviors and contributes to the development of all-optical bidirectional self-powered neuromorphic computing systems.

This study investigates the current state of plant knowledge and use among nomadic communities in Mongolia's Gobi Desert, where rapid socioeconomic and environmental changes threaten traditional practices. Through a comprehensive survey of 50 participants aged 12–65 across four aimags, we documented the recognition, utilization, and cultural significance of 17 key plant species. High recognition rates were found for Haloxylon ammodendron , Agriophyllum pungens , and Rheum nanum , with medicinal use being the most frequently reported category. Environmental changes, particularly reduced water availability and habitat disruption due to mining activities, have significantly affected plant abundance and use. Importantly, while most respondents demonstrated knowledge of specific land management practices, including seasonal migration and protecting water sources, some were unfamiliar with formal conservation techniques. Concerns about the decline of plant knowledge among younger generations were prevalent among participants. The expansion of mining activities, affecting a substantial portion of the study area, emerged as a primary concern. These findings underscore the cultural, economic, and medicinal importance of plant species in the region and highlight the urgent need to preserve traditional ecological knowledge while developing strategies for biocultural diversity conservation amidst increasing challenges to nomadic livelihoods.

High efficiency of charge carrier conduction is crucial for photoelectrical performance in ultraviolet C (UVC) photodetectors (PDs) based on heteroepitaxial beta-gallium oxide (β-Ga2O3) thin films. However, the presence of in-plane rotational domains due to anisotropic symmetry severely degraded the efficiency of charge carrier conduction by trapping and recombination of carriers in conventional lateral PD (LPD). Here, we demonstrate an approach that enables vertical conduction configuration while preserving the high crystallinity of epitaxial Si-doped β-Ga2O3 (Si:Ga2O3) through the epilayer transfer using a hole pattern sapphire nanomembrane (HPSN) growth template. Based on the characterization of domain orientation and photoresponsivity in transferred epitaxial Si:Ga2O3 membranes, we reveal the defect-related anisotropic conduction arising from the vertical interdomain and lateral intradomain conduction. Compared to the indirect intradomain pathway in LPD, the vertical PD (VPD) exhibited high efficiency of charge carrier conduction through the direct interdomain pathways. As a result, the self-powered VPD exhibits high rectifying characteristics with a high detectivity of 1.02 × 1013 Jones and a fast response time of 93 ms. Moreover, the multipixel UVC imaging PD arrays have been successfully demonstrated without any external applied bias, showing high recognition rates and practical utility for reliable UVC imaging applications. Our work not only demonstrates the feasibility of obtaining single-crystal epitaxial membranes for a wide range of material systems but also provides pathways for overcoming material limitations with defect-induced optoelectrical systems.

The realization of vision-based neuromorphic computing relies significantly on advancements in optoelectronic synaptic chip technology. Currently, the development of optoelectronic devices is mainly limited by high power consumption due to high bias voltage and low recognition rates caused by the background noise. A low-power deep-ultraviolet optoelectronic synapse device is developed by doping Mg into zinc oxide to modulate oxygen-vacancy defects. Specifically, the synaptic behavior still has excellent persistent photoconductivity response at 12 mV bias, and the low single synaptic energy consumption is 2.34 pJ. Meanwhile, the artificial neural network of the device is constructed according to the excitation and inhibition characteristics, and the recognition rate of handwritten digits is as high as 95.41%. In addition, on the basis of demonstrating ultraviolet image visual learning and memory, the device provides an encryption algorithm verification array integrating sensing and storage with energy consumption as low as 20 nJ. This low-power and strong anti-interference deep-ultraviolet optoelectronic synapse device supports the development of high-performance visual neural-state computing.

Summary Research into third-party damage events affecting the safe transportation of oil and gas pipelines has revealed challenges caused by the long pipeline distances and the large volume of vibration signal data collected. This makes it difficult to effectively identify disturbances. The available signal data are massive but contains few discernible events. A monitoring method based on phase-sensitive optical fiber backscatter Rayleigh scattering interferometry (Φ-OTDR) and an ensemble empirical mode decomposition (EEMD)-neural network fusion algorithm is proposed. By utilizing Φ-OTDR technology and laying single-core optical fibers alongside the pipeline in the same trench, disturbance signals along the pipeline are collected. Combined with the EEMD algorithm to handle the nonstationarity of the signals, key features are extracted from six types of events, including mechanical damage, knocking damage, vehicle passage, and manual excavation. After feature extraction, neural networks are used for classification and recognition. Experimental data show that this technology can achieve a high recognition rate of 99.0% and a low false alarm rate of 1.0%. Furthermore, the classification effect of the model is further verified through confusion matrix analysis. The practical application of this technology significantly improves the monitoring accuracy of pipeline disturbance events, effectively reduces the false alarm rate, and provides strong technical support for the safety protection of long-distance oil and gas pipelines.

In response to the problem that most current automatic modulation recognition models cannot achieve high recognition rates while maintaining a certain training speed, this study proposes a neural network model based on depthwise separable convolution, residual connection, and channel attention mechanism. By using residual structures and Swish activation to alleviate gradient problems to support deep network training, dynamically optimizing feature channels using SE modules, and significantly reducing computational costs through depthwise separable convolution and global pooling, the model not only achieves lightweighting and ensures a certain training speed, but also has certain feature extraction capabilities, which can achieve high recognition rates. This study conducted comparative experiments to train different models in five SNR environments on the DeepSig RadioML 2018.01A dataset. The CNN_ResNet model can achieve a recognition rate of 92.67% in a 10dB environment, which is 20.07% higher than the basic CNN model. The results of the experiments demonstrate that the improved model exhibits a significantly higher recognition rate compared to other models, while maintaining a certain training speed.

Traditional badminton serve violation penalties rely on the subjective judgment of the referee and have significant human errors. To improve the automatic identification efficiency and accuracy of serving violations, research proposes a badminton serve violation detection system based on an adaptive enhancement algorithm. The system adopts an improved adaptive lifting algorithm for object detection and extracts key frames based on it. The system analyzes and extracts key features of the serving moment to determine whether the badminton serving is legal or illegal. The experiment outcomes reveal that the system can correctly recognize the state of serving time, and performs well in the recognition rate of compliant serving and unauthorized serving, especially under natural and artificial light conditions, with a recognition rate of over 95%. It shows excellent performance in target detection and serving compliance discrimination. The proposed badminton serve violation detection system has a high recognition rate and real-time detection capability, which can effectively meet the needs of refereeing in sports events. The study provides a new referee assistance tool that can help referees make accurate and scientifically based decisions in complex situations.

Aiming at the problem that the recognition rate of existing automatic modulation recognition models needs to be improved under high signal-to-noise ratio conditions, a model consisting of phase transformation, residual CNN network and bidirectional LSTM network is proposed. First, the DeepSig RadioML 2018.01A dataset is normalized as the model input; the phase parameter is extracted through the phase recognition module, and then the phase is corrected according to the phase parameter; then the spatial features are extracted through the residual CNN network to avoid gradient vanishing and explosion; then the data is passed to the bidirectional LSTM network to extract the bidirectional time series features of the data; finally, the deep neural network is used for classification and recognition. Experimental results show that under high signal-to-noise ratio conditions, the model improves the recognition rate of modulation modes such as 16PSK, and the highest recognition rate and average recognition rate reach 96.79% and 62.13% respectively. Compared with other existing models, the overall optimization of recognition rate and model efficiency is achieved.

The hippocampus plays an important role in the formation of episodic memory. To identify patterns of hippocampal firing activity specific to episodic memory, we performed Multiple-Unit Firing Activity (MUA) recognition using deep learning methods. Briefly, adult male rats habituated to their home cage experienced one of four experimental episodic stimuli (restraint stress, contact with a female rat, contact with a male rat, or contact with a novel object) for 10 minutes. The patterns of recorded brain spike signals (300–10 kHz) in hippocampal CA1 were classified using machine learning methods such as Convolutional Neural Networks (CNNs), Support Vector Machines (SVMs), deep learning model VGG16, and combination models composed of VGG16 with SVM or VGG19 with SVM. As a result, the model of VGG19 with SVM detected MUA with ripple-like wave firings corresponding to specific episodes, achieving a validation accuracy of 96.79% which was the highest recognition rate in all of deep learning models. The results suggest that MUA of CA1 containing ripple firings corresponds to specific episodic memories. By capturing ripple firings, MUA analysis can assess and diagnose memory function, which may help detect various cognitive disorders.