Xi'an Siyuan University

Graph neural networks (GNNs) have demonstrated strong performance in handling graph-structured data in recent years, particularly in capturing complex inter-node relationships among data samples, showcasing advantages over traditional neural networks. However, challenges persist, including difficulties in cross-modal information fusion, inadequate modeling of modal relationships, and high computational costs. To address these limitations, this paper proposes GGMEN, a novel model that integrates the local neighborhood aggregation capability of graph convolutional networks with the global structural expressiveness of graph isomorphic networks (GINs). Leveraging shallow feature extraction via time-frequency joint analysis, the paper extracts 14 representative physiological statistical features. Simultaneously, the Transformer model captures spatial features from individual facial expression video frames, enabling spatio-temporal modeling of facial expressions. The graph convolutional network (GCN) layer models temporal dependencies in physiological signals and spatial relationships of facial key points, while the GIN layer enhances modeling of complex higher-order relationships. Multi-modal emotion perception is achieved through attention-based modality fusion. Experiments on multiple datasets validate the model’s effectiveness across multiple emotion perception benchmarks, achieving an emotion recognition accuracy of 81.25% on dataset DEAP. Comparative analyses with existing models confirm the accuracy improvement of the proposed framework.

Abstract Estragole and anethole are the major flavoring and aroma components of the essential oils extracted from basil ( Ocimum basilicum L.), anise ( Pimpinella anisum ), star anise ( Illicium verum ), and fennel ( Foeniculum vulgare ). Besides, they also exhibit bioactivities, such as antioxidation, antimicrobial, immunomodulation, inti-inflammation, and anti-diabetes. These properties make them being faced with substantial applications in the industries of food, cosmetics, pharmaceutics, and agriculture. Commercial estragole and anethole are currently obtained from extraction from their original plants, which suffers from unstable quality and resources. Microbial synthesis is a promising alternative approach to sustainably obtain estragole and anethole, but has not been achieved. In this study, we modularly redesigned the synthetic pathways based on the Escherichia coli chassis. Firstly, a p -coumaryl alcohol over-accumulator (436.28 mg/L) was engineered to facilitate the validation of the undemonstrated downstream module. De novo biosynthesis of estragole and anethole was achieved by stepwise analysis and BioBricks assembly. The intracellular availability of S -adenosyl-L-methionine (SAM) was revealed to be the key limiting factor, and was enhanced by deregulating the synthesis. Finally, the productions of estragole and anethole were improved by 246.88% and 202.44% compared with the initial strains. This study represents the first synthesis of estragole and anethole in microbial chassis, and the demonstration of microbial synthesis of propenyl phenols. Besides, it provided an alternative sustainable approach to obtain flavoring and aroma chemicals.

Impedance Spectroscopy (IS) is a valuable technique for characterizing sensing materials in the frequency domain. However, interpreting IS data and linking them to equivalent circuit models with combined effects can be challenging for beginners in the technique. To address this issue, this work revisits the fundamentals of IS and presents a didactic Virtual Instrument (VI) designed to support students and researchers through intuitive simulation and analysis tools. While similar features exist in well-established research software, our VI takes an educational approach by allowing real-time manipulation of model parameters and immediate visualization of their impact on impedance spectra. The software also performs model fitting to experimental data, promoting the development of critical analysis skills. The tool was initially applied with undergraduate research students, demonstrating strong potential for educational use. As a proof of concept, humidity-sensitive MWCNT-based gas sensors were characterized, and fitting results highlighted the most responsive electrical parameters. This approach offers a practical and accessible way to teach IS principles and electrical analysis of materials.

ABSTRACTThe defect detection of prefabricated building components during production and use is the key to ensuring building quality. Traditional manual methods are inefficient and prone to missed detections, while existing deep learning models face challenges such as insufficient detection accuracy, high computational resource consumption, and poor real‐time performance in complex backgrounds and small target scenes. To address this issue, this study uses YOLOv5 as the basic object detection algorithm, introduces an attention mechanism, replaces the loss function, and improves the Adam optimizer to obtain YOLOv5s. To reduce the size of the YOLOv5s, GhostNet is introduced as a feature extraction network to construct the YOLOv5s‐GhostNet detection model for defect detection of prefabricated building components. In actual defect detection of building components, the YOLOv5s‐GhostNet researched and designed only had 12 defect classification errors, and the classification accuracy was improved to 99.6%. The proposed model had an AUC area of 0.90 for defect detection in building components and a detection speed of 42FPS. In visual analysis, the confidence coefficient of the proposed model for detecting “crack” defects was 99%. YOLOv5s‐GhostNet has extremely high accuracy and efficiency in detecting defects in prefabricated building components and has broad application prospects. In order to verify the performance of the YOLOv5s GhostNet model proposed in this article, traditional YOLOv5, VGG‐16, ResNet, and HRNet were selected as benchmark models for comparison. The experimental results show that the proposed model outperforms these benchmark models in terms of detection accuracy, detection speed, and model complexity, further verifying its effectiveness and superiority in practical applications. The application of this model not only promotes the further development of deep learning technology in architecture but also provides a reference for defect detection in other fields.

In order to improve the accuracy of identifying financial fraud and reduce the false alarm rate, a new adaptive identification and warning method for financial fraud based on deep Q-learning is proposed. Firstly, analyse the extraction of financial behaviour characteristics, including transaction funds, networks, cycles, and supervised features. Secondly, deep Q-learning combines deep neural networks and reinforcement learning to automatically extract key features from financial transaction data, and continuously optimise strategies through interaction with the environment to achieve accurate identification of financial fraud behaviour. Finally, based on the deep Q-learning model, financial fraud behaviour is assessed for risk and classified into warning levels, and corresponding warning response strategies are implemented according to different risk levels. The experimental results show that the adaptive identification accuracy of financial fraud behaviour using our method can reach up to 98.9%, with a maximum false alarm rate of around 1%.

International audience

To address the issues of speed and accuracy in traffic image target detection under foggy conditions, this paper proposes a dehazing algorithm using bilateral filtering and an improved YOLOX-S algorithm for foggy traffic image target detection. The bilateral filter is integrated into the Retinex algorithm to perform image dehazing and enhance the quality of foggy images. Based on the improved YOLOX-S architecture, deep separable convolutions are introduced to replace traditional convolutions in the FPN and YOLOHead residual structures, thereby reducing the model's parameter count. Additionally, the ECANet channel attention mechanism is incorporated to enhance the network's focus, reducing false positive and false negative rates in traffic image target detection under foggy conditions. Experimental results show that the bilateral filtering algorithm effectively improves the quality of foggy images, achieving a peak signal-to-noise ratio (PSNR) of 24.032 dB and a structural similarity index (SSIM) of 0.873. The improved YOLOX-S algorithm demonstrates precise target detection, with an average precision of 98% and a detection frame rate of 23 frames per second.

With the deepening of educational reform in the new era, innovation and entrepreneurship education has gradually become an important means of cultivating high-quality talents in colleges and universities. However, it is difficult for pure technical education to meet the social demand for comprehensive development, and the integration of curriculum ideology and politics and practice into innovation and entrepreneurship education has become a key way to enhance students' professional ethics and social responsibility. This paper discusses the practical strategy of combining innovation and entrepreneurship education with curriculum ideology and politics in colleges and universities, and puts forward four paths of integration and practice in combination with the application in the medical field. These include enhancing teachers' ideological and political literacy and innovation and entrepreneurship education ability, optimizing curriculum design and teaching methods, constructing and optimizing practice platforms, and improving evaluation mechanisms and incentives, etc., aiming to comprehensively cultivate medical students' professional ethics, sense of social responsibility, and innovation and practice ability. The integration of curriculum ideology not only improves students' professional skills, but also shapes their correct values and sense of social responsibility at the ideological level, and delivers comprehensive medical talents that meet the needs of the new era for the society.