High Recognition Accuracy Research Articles

Explainable deep learning for sEMG-based similar gesture recognition: A Shapley-value-based solution

Surface electromyography (sEMG) based gesture recognition shows promise in enhancing human-robot interaction. However, accurately recognizing similar gestures is a challenging task, and the underlying mechanisms of gesture recognition are not well understood. To address these issues, we developed a new solution called the Shapley-value-based similar gesture recognition (SV-SGR) method. Our solution combines deep learning and game theory to achieve high recognition accuracy and interpretability. First, we devised a data preprocessing method that converts sEMG signals into sEMG color images, which can be more effectively utilized by deep learning techniques. Next, we established a deep-neural-network-based model for gesture recognition using the processed sEMG color images. Then, we designed a global explanation approach based on Shapley values to quantify the contribution of each channel to recognizing similar gestures. Finally, we carried out an explanation analysis, which provides feedback on the recognition model to enhance the precision of gesture recognition. Extensive comparisons and interpretable analyses have been conducted on real-world datasets, and the results demonstrate that the SV-SGR method outperforms other baselines under various experimental conditions. The interpretable analysis method based on Shapley values effectively enhances the performance of recognizing similar gestures and provides valuable insights into the decision-making process of recognition models.

Rapeseed Seed Coat Color Classification Based on the Visibility Graph Algorithm and Hyperspectral Technique

Information technology and statistical modeling have made significant contributions to smart agriculture. Machine vision and hyperspectral technologies, with their non-destructive and real-time capabilities, have been extensively utilized in the non-destructive diagnosis and quality monitoring of crops and seeds, becoming essential tools in traditional agriculture. This work applies these techniques to address the color classification of rapeseed, which is of great significance in the field of rapeseed growth diagnosis research. To bridge the gap between machine vision and hyperspectral technology, a framework is developed that includes seed color calibration, spectral feature extraction and fusion, and the recognition modeling of three seed colors using four machine learning methods. Three categories of rapeseed coat colors are calibrated based on visual perception and vector-square distance methods. A fast-weighted visibility graph method is employed to map the spectral reflectance sequences to complex networks, and five global network attributes are extracted to fuse the full-band reflectance as model input. The experimental results demonstrate that the classification recognition rate of the fused feature reaches 0.943 under the XGBoost model, confirming the effectiveness of the network features as a complement to the spectral reflectance. The high recognition accuracy and simple operation process of the framework support the further application of hyperspectral technology to analyze the quality of rapeseed.

Research on the Recognition and Application of Montreal Forced Aligner for Singing Audio

This paper discusses the feasibility of obtaining phoneme-aligned time segments for singing audio using the Montreal Forced Aligner (MFA) tool. Initially, the recognition effectiveness of singing audio data is tested using the open-source MFA model. Subsequently, samples with high recognition accuracy are manually annotated and used to train the MFA model. Finally, the recognition effectiveness of the open-source MFA model and the MFA model trained on singing audio data is observed. It is found that the performance of the trained MFA model is significantly improved. This also confirms the feasibility of MFA in recognizing singing audio.

Generating bimodal privacy-preserving data for face recognition

The performance of state-of-the-art face recognition systems depends crucially on the availability of large-scale training datasets. However, increasing privacy concerns nowadays accompany the collection and distribution of biometric data, which has already resulted in the retraction of valuable face recognition datasets. The use of synthetic data represents a potential solution, however, the generation of privacy-preserving facial images useful for training recognition models is still an open problem. Generative methods also remain bound to the visible spectrum, despite the benefits that multispectral data can provide. To address these issues, we present a novel identity-conditioned generative framework capable of producing large-scale recognition datasets of visible and near-infrared privacy-preserving face images. The framework relies on a novel identity-conditioned dual-branch style-based generative adversarial network to enable the synthesis of aligned high-quality samples of identities determined by features of a pretrained recognition model. In addition, the framework incorporates a novel filter to prevent samples of privacy-breaching identities from reaching the generated datasets and improve both identity separability and intra-identity diversity. Extensive experiments on six publicly available datasets reveal that our framework achieves competitive synthesis capabilities while preserving the privacy of real-world subjects. The synthesized datasets also facilitate training more powerful recognition models than datasets generated by competing methods or even small-scale real-world datasets. Employing both visible and near-infrared data for training also results in higher recognition accuracy on real-world visible spectrum benchmarks. Therefore, training with multispectral data could potentially improve existing recognition systems that utilize only the visible spectrum, without the need for additional sensors.

Identification and classification of surface defects for digital twin models of the workpiece.

Workpiece surface defect detection is an indispensable part of intelligent production. The surface information obtained by traditional 2D image detection has some limitations due to the influence of environmental light factors and part complexity. However, the digital twin model has the characteristics of high fidelity and scalability, and the digital twin surface can be obtained by a device with a scanning accuracy of 0.02mm to achieve the representation of the real surface of the workpiece. The surface defect detection system for digital twin models is proposed based on the improved YOLOv5 model in this paper. Firstly, the digital twin model of the workpiece is reconstructed by the point cloud data obtained by the scanning device, and the surface features with defects are captured. Subsequently, the training dataset is calibrated based on the defect surface, where the defect types include Inclusion, Perforation, pitting surface and Rolled-in scale. Finally, the improved YOLOv5 model with CBAM mechanism and BiFPN module was used to identify the surface defects of the digital twin model and compare it with the original YOLOv5 model and other common models. The results show that the improved YOLOv5 model can realize the identification and classification of surface defects. Compared with the original YOLOv5 model, the mAP value of the improved YOLOv5 model has increased by 0.2%, and the model has high precision. On the basis of the same data set, the improved YOLOv5 model has higher recognition accuracy than other models, improving 11.7%, 3.4%, 6.2%, 33.5%, respectively. As a result, this study provides a practical and systematic detection method for digital twin model surface during the intelligent production process, and realizes the rapid screening of the workpiece with defects.

Memory-electroluminescence for multiple action-potentials combination in bio-inspired afferent nerves

The development of optoelectronics mimicking the functions of the biological nervous system is important to artificial intelligence. This work demonstrates an optoelectronic, artificial, afferent-nerve strategy based on memory-electroluminescence spikes, which can realize multiple action-potentials combination through a single optical channel. The memory-electroluminescence spikes have diverse morphologies due to their history-dependent characteristics and can be used to encode distributed sensor signals. As the key to successful functioning of the optoelectronic, artificial afferent nerve, a driving mode for light-emitting diodes, namely, the non-carrier injection mode, is proposed, allowing it to drive nanoscale light-emitting diodes to generate a memory-electroluminescence spikes that has multiple sub-peaks. Moreover, multiplexing of the spikes can be obtained by using optical signals with different wavelengths, allowing for a large signal bandwidth, and the multiple action-potentials transmission process in afferent nerves can be demonstrated. Finally, sensor-position recognition with the bio-inspired afferent nerve is developed and shown to have a high recognition accuracy of 98.88%. This work demonstrates a strategy for mimicking biological afferent nerves and offers insights into the construction of artificial perception systems.

Identification of Different Dairy Products Using Raman Spectroscopy Combined with Fused Lasso Distributionally Robust Logistic Regression

Raman spectroscopy has been more widely used recently in the quality detection of dairy products. Because Raman spectroscopy can conduct rapid analyses of small sample sizes at high dimensions, its use in the dairy industry is becoming a hot topic for researchers. To improve the robustness and accuracy of logistic regression identification method, a new Raman spectroscopy identification method was proposed that combines a distributionally robust optimization technique and fused lasso technique with logistic regression. Then, Raman spectroscopy was used to analyze two types of dairy products that were collected for anti-jamming identification testing to verify the effectiveness of the new method. The experimental results show that the proposed method is more robust and has a higher recognition accuracy than the traditional logistic regression.

Enhancing Robustness of Surface Electromyography Pattern Recognition at Different Arm Positions for Transhumeral Amputees Using Deep Adversarial Inception Domain Adaptation

Pattern recognition in myoelectric control that relies on the myoelectric activity associated with arm motions is an effective control method applied to myoelectric prostheses. Individuals with transhumeral amputation face significant challenges in effectively controlling their prosthetics, as muscle activation varies with changes in arm positions, leading to a notable decrease in the accuracy of motion pattern recognition and consequently resulting in a high rejection rate of prosthetic devices. Therefore, to achieve high accuracy and arm position stability in upper-arm motion recognition, we propose a Deep Adversarial Inception Domain Adaptation (DAIDA) based on the Inception feature module to enhance the generalization ability of the model. Surface electromyography (sEMG) signals were collected from 10 healthy subjects and two transhumeral amputees while performing hand, wrist, and elbow motions at three arm positions. The recognition performance of different feature modules was compared, and ultimately, accurate recognition of upper-arm motions was achieved using the Inception C module with a recognition accuracy of 90.70% ± 9.27%. Subsequently, validation was performed using data from different arm positions as source and target domains, and the results showed that compared to the direct use of a convolutional neural network (CNN), the recognition accuracy on untrained arm positions increased by 75.71% (p < 0.05), with a recognition accuracy of 91.25% ± 6.59%. Similarly, in testing scenarios involving multiple arm positions, there was a significant improvement in recognition accuracy, with recognition accuracy exceeding 90% for both healthy subjects and transhumeral amputees.

Pure ZrO2 Ferroelectric Thin Film for Nonvolatile Memory and Neural Network Computing.

The recent discovery of ferroelectricity in pure ZrO2 has drawn much attention, but the information storage and processing performances of ferroelectric ZrO2-based nonvolatile devices remain open for further exploration. Here, a ZrO2 (∼8 nm)-based ferroelectric capacitor using RuO2 oxide electrodes is fabricated, and the ferroelectric orthorhombic phase evolution under electric field cycling is studied. A ferroelectric remnant polarization (2Pr) of >30 μC/cm2, leakage current density of ∼2.79 × 10-8 A/cm2 at 1 MV/cm, and estimated polarization retention of >10 years are achieved. When the ferroelectric capacitor is connected with a transistor, a memory window of ∼0.8 V and eight distinct states can be obtained in such a ferroelectric field-effect transistor (FeFET). Through the conductance manipulation of the FeFET, a high object image recognition accuracy of ∼93.32% is achieved on the basis of the CIFAR-10 dataset in the convolutional neural network (CNN) simulation, which is close to the result of ∼94.20% obtained by floating-point-based CNN software. These results demonstrate the potential of ferroelectric ZrO2 devices for nonvolatile memory and artificial neural network computing.

Integrating Image Perception and Time‐to‐First‐Spike Coding in MoS2 Phototransistors for Spiking Neural Network

AbstractHuman vision system remains alert for dangers and adopts high‐speed and low‐power coding methods to convert the image information to spike signals. To meet the demand for danger alert in machine vision, it is important to design intelligent sensors to integrate the functions of image perception and high‐efficiency coding for high‐priority analysis. Inspired by the human visual system, a MoS2 phototransistor is introduced on SiNx substrate enabling simultaneous image perception and time‐to‐first‐spike (TTFS) coding. The device demonstrates exceptional performance in encoding 3‐bit grayscale images, achieving a low mean squared error of 0.008 and a high structural similarity index of 0.9784. Spiking neural networks (SNN) with TTFS coding achieve high recognition accuracy (98.86%) while reducing spike count by 75%. The device array also perceives motion direction and object states by converting data temporally. This work establishes a hardware foundation to promote the performance of SNNs in efficiently identifying crucial information.

Spectrometer-Less Remote Sensing Image Classification Based on Gate-Tunable van der Waals Heterostructures.

Remote sensing technology, which conventionally employs spectrometers to capture hyperspectral images, allowing for the classification and unmixing based on the reflectance spectrum, has been extensively applied in diverse fields, including environmental monitoring, land resource management, and agriculture. However, miniaturization of remote sensing systems remains a challenge due to the complicated and dispersive optical components of spectrometers. Here, m-phase GaTe0.5Se0.5 with wide-spectral photoresponses (250-1064nm) and stack it with WSe2 are utilizes to construct a two-dimensional van der Waals heterojunction (2D-vdWH), enabling the design of a gate-tunable wide-spectral photodetector. By utilizing the multi-photoresponses under varying gate voltages, high accuracy recognition can be achieved aided by deep learning algorithms without the original hyperspectral reflectance data. The proof-of-concept device, featuring dozens of tunable gate voltages, achieves an average classification accuracy of 87.00% on 6 prevalent hyperspectral datasets, which is competitive with the accuracy of 250-1000nm hyperspectral data (88.72%) and far superior to the accuracy of non-tunable photoresponse (71.17%). Artificially designed gate-tunable wide-spectral 2D-vdWHs GaTe0.5Se0.5/WSe2-based photodetector present a promising pathway for the development of miniaturized and cost-effective remote sensing classification technology.

Flexible retinomorphic vision sensors with scotopic and photopic adaptation for a fully flexible neuromorphic machine vision system

AbstractBioinspired neuromorphic machine vision system (NMVS) that integrates retinomorphic sensing and neuromorphic computing into one monolithic system is regarded as the most promising architecture for visual perception. However, the large intensity range of natural lights and complex illumination conditions in actual scenarios always require the NMVS to dynamically adjust its sensitivity according to the environmental conditions, just like the visual adaptation function of the human retina. Although some opto‐sensors with scotopic or photopic adaption have been developed, NMVSs, especially fully flexible NMVSs, with both scotopic and photopic adaptation functions are rarely reported. Here we propose an ion‐modulation strategy to dynamically adjust the photosensitivity and time‐varying activation/inhibition characteristics depending on the illumination conditions, and develop a flexible ion‐modulated phototransistor array based on MoS2/graphdiyne heterostructure, which can execute both retinomorphic sensing and neuromorphic computing. By controlling the intercalated Li+ ions in graphdiyne, both scotopic and photopic adaptation functions are demonstrated successfully. A fully flexible NMVS consisting of front‐end retinomorphic vision sensors and a back‐end convolutional neural network is constructed based on the as‐fabricated 28 × 28 device array, demonstrating quite high recognition accuracies for both dim and bright images and robust flexibility. This effort for fully flexible and monolithic NMVS paves the way for its applications in wearable scenarios.

Swing-phase detection of locomotive mode transitions for smooth multi-functional robotic lower-limb prosthesis control.

Robotic lower-limb prostheses, with their actively powered joints, may significantly improve amputee users' mobility and enable them to obtain healthy-like gait in various modes of locomotion in daily life. However, timely recognition of the amputee users' locomotive mode and mode transition still remains a major challenge in robotic lower-limb prosthesis control. In the paper, the authors present a new multi-dimensional dynamic time warping (mDTW)-based intent recognizer to provide high-accuracy recognition of the locomotion mode/mode transition sufficiently early in the swing phase, such that the prosthesis' joint-level motion controller can operate in the correct locomotive mode and assist the user to complete the desired (and often power-demanding) motion in the stance phase. To support the intent recognizer development, the authors conducted a multi-modal gait data collection study to obtain the related sensor signal data in various modes of locomotion. The collected data were then segmented into individual cycles, generating the templates used in the mDTW classifier. Considering the large number of sensor signals available, we conducted feature selection to identify the most useful sensor signals as the input to the mDTW classifier. We also augmented the standard mDTW algorithm with a voting mechanism to make full use of the data generated from the multiple subjects. To validate the proposed intent recognizer, we characterized its performance using the data cumulated at different percentages of progression into the gait cycle (starting from the beginning of the swing phase). It was shown that the mDTW classifier was able to recognize three locomotive mode/mode transitions (walking, walking to stair climbing, and walking to stair descending) with 99.08% accuracy at 30% progression into the gait cycle, well before the stance phase starts. With its high performance, low computational load, and easy personalization (through individual template generation), the proposed mDTW intent recognizer may become a highly useful building block of a prosthesis control system to facilitate the robotic prostheses' real-world use among lower-limb amputees.

A multi-omics study to monitor senescence-associated secretory phenotypes of Alzheimer's disease.

Alzheimer's disease (AD) is characterized by the progressive degeneration and damage of neurons in the brain. However, developing an accurate diagnostic assay using blood samples remains a challenge in clinic practice. The aim of this study was to explore senescence-associated secretory phenotypes (SASPs) in peripheral blood using mass spectrometry based multi-omics approach and to establish diagnostic assays for AD. This retrospective study included 88 participants, consisting of 29 AD patients and 59 cognitively normal (CN) individuals. Plasma and serum samples were examined using high-resolution mass spectrometry to identify proteomic and metabolomic profiles. Receiver operating characteristic (ROC) analysis was employed to screen biomarkers with diagnostic potential. K-nearest neighbors (KNN) algorithm was utilized to construct a multi-dimensional model for distinguishing AD from CN. Proteomics analysis revealed upregulation of five plasma proteins in AD, including RNA helicase aquarius (AQR), zinc finger protein 587B (ZNF587B), C-reactive protein (CRP), fibronectin (FN1), and serum amyloid A-1 protein (SAA1), indicating their potential for AD classification. Interestingly, KNN-based three-dimensional model, comprising AQR, ZNF587B, and CRP, demonstrated its high accuracy in AD recognition, with evaluation possibilities of 0.941, 1.000, and 1.000 for the training, testing, and validation datasets, respectively. Besides, metabolomics analysis suggested elevated levels of serum phenylacetylglutamine (PAGIn) in AD. The multi-omics outcomes highlighted the significance of the SASPs, specifically AQR, ZNF587B, CRP, and PAGIn, in terms of their potential for diagnosing AD and suggested neuronal aging-associated pathophysiology.

Human figure detection in Han portrait stone images via enhanced YOLO-v5

The unearthed Han Dynasty portrait stones are an important part of China’s ancient artistic heritage, and detecting human images in these stones is a critical prerequisite for studying their artistic value. However, high-precision target detection techniques often result in a large number of parameters, making them unsuitable for portable devices. In this work, we propose a new human image target detection model based on an enhanced YOLO-v5. We discovered that the complex backgrounds, dense group targets, and significant scale variations of targets within large scenes in portrait stones present significant challenges for human target image detection. Therefore, we first incorporated the SPD-Conv convolution and Coordinate Attention self-attention mechanism modules into the YOLO-v5 architecture, aiming to enhance the model’s recognition precision for small target images within Han portrait stones and strengthen its resistance to background disturbances. Moreover, we introduce DIoU NMS and Alpha-IoU Loss to improve the detector’s performance in dense target scenarios, reducing the omission of densely packed objects. Finally, the experimental results from our collected dataset of Han Dynasty stone figure images demonstrate that our method achieves fast convergence and high recognition accuracy. This approach can be better applied to the target detection tasks of special character images in complex backgrounds.

Flexible Multimodal Magnetoresistive Sensors Based on Alginate/Poly(vinyl alcohol) Foam with Stimulus Discriminability for Soft Electronics Using Machine Learning.

Flexible foam-based sensors have attracted substantial interest due to their high specific surface area, light weight, superior deformability, and ease of manufacture. However, it is still a challenge to integrate multimodal stimuli-responsiveness, high sensitivity, reliable stability, and good biocompatibility into a single foam sensor. To achieve this, a magnetoresistive foam sensor was fabricated by an in situ freezing-polymerization strategy based on the interpenetrating networks of sodium alginate, poly(vinyl alcohol) in conjunction with glycerol, and physical reinforcement of core-shell bidisperse magnetic particles. The assembled sensor exhibited preferable magnetic/strain-sensing capability (GF ≈ 0.41 T-1 for magnetic field, 4.305 for tension, -0.735 for bending, and -1.345 for pressing), quick response time, and reliable durability up to 6000 cycles under external stimuli. Importantly, a machine learning algorithm was developed to identify the encryption information, enabling high recognition accuracies of 99.22% and 99.34%. Moreover, they could be employed as health systems to detect human physiological motion and integrated as smart sensor arrays to perceive external pressure/magnetic field distributions. This work provides a simple and ecofriendly strategy to fabricate biocompatible foam-based multimodal sensors with potential applications in next-generation soft electronics.

Identification of the Surface Cracks of Concrete Based on ResNet-18 Depth Residual Network

To ensure the safety and durability of concrete structures, timely detection and classification of concrete cracks using a low-cost and high-efficiency method is necessary. In this study, a concrete surface crack damage detection method based on the ResNet-18 residual network was developed. This method was implemented by training a model with images to extract the cracks, where the image processing algorithms and deep learning were combined. The results show that the computational accuracy can meet the requirements by utilizing the established image dataset and appropriate model hyperparameters. The trained model had high recognition accuracy when the 256 × 256 resolution images were adopted, and the worst accuracy of crack recognition in the test set was over 90%. The average accuracy in the test set was 91.3% when considering environmental interference generated by processing the images with a brightness adjustment, salt-and-pepper noise, and localized interference. Then, it was demonstrated that the present model possesses good robustness for crack identification in different environments. The average recognition accuracy when dealing with images of a real bridge, which are outside the training dataset, was 99.7%. The residual network model developed in this study has the advantages of low cost, high efficiency, and practicality compared to traditional detection methods. Compared to the existing deep learning methods, the model created in this study requires less computational resources and storage space, and shows a faster training speed and higher accuracy.

Design and Optimization of Human-Computer Interaction System for Education Management Based on Artificial Intelligence

The continuous improvement and refinement of artificial intelligence (AI) technology has facilitated the broader application of human-computer interaction in the field of education management. The construction of an educational management human-computer interaction system based on AI technology can optimize and improve key parameters of educational management human-computer interaction scenarios, thereby creating a more comprehensive mobile learning (m-learning) application system. This paper is based on AI technology, analyzing gesture semantics and speech semantics, and combining fusion algorithms to construct an education management human interaction system. The performance changes of the system were compared with real experimental operations and the NOBOOK platform analysis. The results show that the education management human-computer interaction system constructed in this article can enhance the m-learning experience of participants. It ensures high recognition accuracy, leading to higher scores in all dimensions of indicator evaluation. Therefore, as one of the crucial forms of m-learning, the human-computer interaction system for education management based on AI can establish a foundation for the further enhancement and development of education management.

Ferroelectricity-Defects Synergistic Artificial Synapses for High Recognition Accuracy Neuromorphic Computing.

The ability of ferroelectric memristors to modulate conductance and offer multilevel storage has garnered significant attention in the realm of artificial synapses. On one hand, the resistance change of ferroelectric memristors mainly depends on the polarization reversal. On the other hand, the defects such as oxygen vacancies, which are inevitable presence during high-temperature processes, can undergo diffusion drift with the polarization reversal, thereby change the interface potential barrier. Thus, it is both desirable and necessary to investigate the synergistic effect of ferroelectricity and defects. Here, we prepare BaTiO3 ferroelectric memristor by pulse laser deposition and achieve resistance switching through the synergistic effect of ferroelectricity and oxygen vacancies. The memristor shows excellent switching characteristics with a large switching ratio (104) and good stability (103 s). It effectively emulates the features of artificial synapses and accomplishes decimal logical neural computing. In the neuromorphic system crafted with the memristor, the recognition accuracy of the 28 × 28 pixel image reaches 94.9%. These findings strongly support the research of ferroelectric memristors in neuromorphic devices.

One memristor–one electrolyte-gated transistor-based high energy-efficient dropout neuronal units

Abstract Artificial neural networks (ANN) have been extensively researched due to their significant energy-saving benefits. Hardware implementations of ANN with dropout function would be able to avoid the overfitting problem. This letter reports a dropout neuronal unit (1R1T-DNU) based on one memristor–one electrolyte-gated transistor with an ultralow energy consumption of 25 pJ/spike. A dropout neural network is constructed based on such a device and has been verified by MNIST dataset, demonstrating high recognition accuracies (> 90%) within a large range of dropout probabilities up to 40%. The running time can be reduced by increasing dropout probability without a significant loss in accuracy. Our results indicate the great potential of introducing such 1R1T-DNUs in full-hardware neural networks to enhance energy efficiency and to solve the overfitting problem.