Cardiovascular disease remains a leading cause of global mortality, underscoring the critical need for accurate and efficient diagnosis of cardiac conditions. Although the electrocardiogram (ECG) is a widely used diagnostic tool, its utility is often limited by signal artifacts and phenotypic similarities among distinct pathologies. To address these limitations, we propose a novel Hybrid Dynamic Graph Convolutional Network (HDGCN) to detect multi-lead ECG arrhythmia. In our HDGCN, ECG signals are represented as graph structures, where nodes correspond to sampling points and edges encode spatiotemporal relationships. Local spatiotemporal features are extracted using a pre-trained ResNet module to capture subtle morphological variations, while an adaptive graph convolutional module with a learnable adjacency matrix dynamically models deeper inter-lead dependencies. Additionally, wavelet-based denoising is applied during preprocessing to preserve clinically relevant features, and depthwise separable convolutions are incorporated to substantially reduce computational complexity. Experimental results demonstrate that HDGCN achieves an average accuracy of 99.57%. The model exhibits robust performance in detecting complex arrhythmias, notably attaining an improvement exceeding 6% for categories such as atrial premature beats.

Electronic Speckle Pattern Interferometry (ESPI) is widely used in fields such as displacement measurement, deformation analysis, and defect detection. However, ESPI fringe patterns are very difficult to denoise due to the coexistence of high-density fringe structures and strong speckle noise, where conventional methods often struggle to preserve fringe edges while suppressing noise. To fundamentally address this edge-preserving denoising challenge, we propose PMNet, a novel deep network architecture that is mathematically supported by the Perona-Malik (PM) anisotropic diffusion equation. Unlike generic denoising networks, our PMNet is not merely a “black box” approximator. Instead, it is designed to explicitly simulate the numerical scheme of the PM equation, which inherently possesses an edge-sensing mechanism through its gradient-dependent diffusion coefficient. To further adapt to the multi-scale nature of ESPI fringes and noise, we integrate a wavelet transform frontend that decomposes the image into frequency bands, allowing the subsequent PM-driven residual blocks to perform targeted, scale-aware denoising. The result is a highly interpretable and effective network that learns to dynamically balance noise removal and fringe preservation. Experiments on both simulated and real ESPI patterns demonstrate that PMNet significantly outperforms state-of-the-art methods in terms of denoising quality, fringe continuity, and generalization capability, providing a physically-principled and practically superior solution for ESPI image processing.

Age-related macular degeneration (AMD) is a leading cause of visual impairment in middle-aged and elderly populations. Existing diagnostic methods are often inefficient and subjective. To achieve accurate and efficient diagnosis of AMD lesions, this paper designs BELD-Net, an efficient segmentation network based on a large visual model. The model employs DINOv2 as its backbone to capture long-range and cross-scale features of AMD lesions through large-scale self-supervised pre-training. Combined with the query-based segmentation mechanism of Mask2Former, it significantly improves generalization in few-shot scenarios. BELD-Net incorporates LoRA modules into the Transformer layers of DINOv2 for low-rank adaptation, reducing computational cost and enhancing the model’s adaptability to fundus images while avoiding overfitting. Additionally, a Multi-scale Boundary Enhancement Module (MBEM) is designed using heterogeneous convolutions and an adaptive squeeze-and-excitation (SE) mechanism to address blurred and variable-sized lesion boundaries, thereby improving sensitivity to low-contrast edges. Experimental results demonstrate that BELD-Net achieves outstanding performance in automatic multi-lesion segmentation of AMD, with an mDice value of 59.61

With the continuous improvement of processes by textile manufacturers, the requirements for pilling of fabric materials are also increasing. At present, in the textile industry, the rating method for fabric pilling is to compare the sample with the standard sample in a specific environment to determine the grade of fabric pilling. However, this method is greatly influenced by subjective factors, and when facing a large number of samples, it consumes a large amount of manpower and material resources. Therefore, an objective, stable, and highly accurate rating method is needed. This paper proposes an end-to-end objective rating method based on a convolutional neural network model called DENet. The network employs a two-branch architecture that integrates an attention mechanism with multiscale convolution and connects large convolutional kernels across layers. To align with the subjective grading criteria for fabric pilling, this study simulates the pilling process of eight types of fabrics using two methods: the pilling box method and the circular trajectory pilling method. Consequently, eight fabric pilling image datasets were generated, covering knitted fabrics, woven fabrics, and nonwoven fabrics, including two patterned fabric types. The experiment results show that the accuracy of the rating method proposed in this paper is 97.95% for combined training on eight datasets, and the average accuracy is 98.07% for separate training on eight datasets, and the accuracy and generalization of the designed network model is high. The proposed system provides an efficient, objective, and stable solution for fabric pilling evaluation.

Electronic speckle pattern interferometry (ESPI) is an important non-destructive testing technique. Denoising the interference fringe pattern is the key link of this technique as well as a hot spot of current research. In this work, we introduce the Restormer Partial Differential Network (Res-PDNet) for ESPI fringe denoising. The method uses PDNet as the base network, which combines the partial differential equation denoising idea and deep learning model, and optimizes the network according to the characteristics of the fringe patterns. In order to balance fringe denoising and structure preservation, our approach incorporates two key enhancements. First, Multi-Dconv Head Transposed Attention and Gated-Dconv Feed-Forward Network modules of Restormer are added after the residual blocks of PDNet, which allows the network to capture more information about the fringe structure and texture. Second, orientation constraints of the fringe pattern are introduced in the loss function to further protect the fringe shape. The Res-PDNet network can accurately recognize the fringe structure and maintain the fringe shape while filtering out noise. It has a good denoising effect on the electron scattering interference fringe pattern.

Cancer is generally thought to be caused by the accumulation of mutations in driver genes. The identification of cancer driver genes is crucial for cancer research, diagnosis and treatment. Despite existing methods, challenges remain in comprehensively learning of the attributes and intricate interactions of genetic data. We propose a novel Multi-information Fusion Graph Convolutional Network (MF-GCN) for cancer driver gene identification, based on multi-omics pan-cancer data and Gene Regulatory Network (GRN) data. Directed topological and attribute graph networks learn gene interactions and self-attribute information, while a common graph network captures consistency between topology and attributes. An attention mechanism adaptively fuses these information with importance weights to identify cancer driver genes. Experimental results showed that MF-GCN can effectively identify cancer driver genes across three GRN datasets, with AUROC and AUPRC improvements of 2.66% and 2.69%, respectively, compared with the state-of-the-art approaches.

Video scene graph generation (VidSGG) aims to predict visual relation triplets in videos, which is a key step toward a deeper comprehension of video scenes. To enhance applicability in real-world scenarios, open-vocabulary settings have been explored in recent VidSGG frameworks. However, the heavy reliance on aligned visual and textual representations from pre-trained vision-language models (VLMs) limits the in-depth understanding of visual relationships, as they fail to comprehend compositional scene relationships. To address this, we propose a novel open-vocabulary VidSGG framework named semantic-unified cross-modal learning (SUCML), which leverages the exceptional capabilities of large language models (LLMs) in visual understanding and semantic reasoning to achieve robust visual relation prediction. Specifically, we incorporate rich knowledge and open-vocabulary capabilities into our framework and design a cross-modal adapter to facilitate the semantic-unified cross-modal representation learning process. We then utilize the impressive semantic understanding and reasoning capabilities of LLMs, providing predefined textual instructions along with the learned semantic-unified visual and text tokens as inputs to the pre-trained LLM for robust relation prediction. Extensive experimental results on two public datasets demonstrate that SUCML significantly outperforms existing methods, showcasing the promising potential of LLMs for high-level semantic reasoning and scene understanding.

Diabetic retinopathy (DR), the leading cause of vision loss among diabetic adults worldwide, underscores the importance of early detection and timely treatment using fundus images to prevent vision loss. However, existing deep learning methods struggle to capture the correlation and contextual information of subtle lesion features with the current scale of dataset. To this end, we propose a novel Multi-scale Spatial-aware Transformer Network (MSTNet) for DR classification. MSTNet encodes information from image patches at varying scales as input features, constructing a dual-pathway backbone network comprised of two Transformer encoders of different sizes to extract both local details and global context from images. To fully leverage structural prior knowledge, we introduce a Spatial-aware Module (SAM) to capture spatial local information within the images. Furthermore, considering the differences between medical and natural images, specifically that regions of interest in medical images often lack distinct subjectivity and continuity, we employ a Multiple Instance Learning (MIL) strategy to aggregate features from diverse regions, thereby enhancing correlation to subtle lesion areas. Ultimately, a cross-fusion classifier integrates dual-pathway features to produce the final classification result. We evaluate MSTNet on four public DR datasets, including APTOS2019, RFMiD2020, Messidor, and IDRiD. Extensive experiments demonstrate that MSTNet exhibits superior diagnostic and grading accuracy, achieving improvements of up to 2.0% in terms of ACC and 1.2% in terms of F1 score, highlighting its effectiveness in accurately assessing fundus images.

Skeleton-based temporal action segmentation aims to segment and classify human actions in untrimmed skeletal sequences. Existing methods struggle with distinguishing transition poses between adjacent frames and fail to adequately capture semantic dependencies between joints and actions. To address these challenges, we propose a large language model-based spatio-temporal semantic enhancement (LLM-STSE) method, a novel framework that combines adaptive spatio-temporal axial attention (ASTA-Attention) and dynamic semantic-guided multimodal action segmentation (DSG-MAS). ASTA-Attention models spatial and temporal dependencies using axial attention, whereas DSG-MAS dynamically generates semantic prompts based on joint motion and fuses them with skeleton features for more accurate segmentation. Experiments on MCFS and PKU-MMD datasets show that LLM-STSE achieves state-of-the-art performance, significantly improving action segmentation, especially in complex transitions, with substantial F1 score gains across multiple public datasets.

Electronic speckle pattern interferometry (ESPI) is a non-contact, full field, real-time measurement technology, which judges the position and size of the internal defects of the object through the external deformation caused by the internal defects under certain loading conditions. We present the effect of loading mode and loading parameters to the defect detection. Firstly, the finite element analysis method is used to establish models to simulate the defect detection of aluminum plates under different loading conditions. Mechanical models are established to simulate different loading mode, loading sizes, defect depth and defect sizes. Secondly, the interpolation method based on partial differential equation is applied to obtain the whole field out-of-plane displacement after finite element analysis. Thirdly, by analyzing the interference fringe patterns obtained from the out-of-plane displacement caused by different defects, the deformation rules in the detection of internal defects of aluminum plates are obtained under different loading conditions. Finally, the loading mode and loading range suitable for the internal defect detection of aluminum materials are summarized. This method can provide a basis for the selection of loading mode and parameters in the ESPI experimental system.

Fundus disease is a complex and universal disease involving a variety of pathologies. Its early diagnosis using fundus images can effectively prevent further diseases and provide targeted treatment plans for patients. Recent deep learning models for classification of this disease are gradually emerging as a critical research field, which is attracting widespread attention. However, in practice, most of the existing methods only focus on local visual cues of a single image, and ignore the underlying explicit interaction similarity between subjects and correlation information among pathologies in fundus diseases. In this paper, we propose a novel label-aware dual graph neural networks for multi-label fundus image classification that consists of population-based graph representation learning and pathology-based graph representation learning modules. Specifically, we first construct a population-based graph by integrating image features and non-image information to learn patient's representations by incorporating associations between subjects. Then, we represent pathologies as a sparse graph where its nodes are associated with pathology-based feature vectors and the edges correspond to probability of the co-occurrence of labels to generate a set of classifier scores by the propagation of multi-layer graph information. Finally, our model can adaptively recalibrate multi-label outputs. Detailed experiments and analysis of our results show the effectiveness of our method compared with state-of-the-art multi-label fundus image classification methods.

Fabric fuzzing and pilling have long been a problem in the textile industry, and consumer complaints about fabric fuzzing and pilling account for about 29.5% of the total textile product quality problems every year. At present, more and more subjective rating methods are used at home and abroad to rate fabric pilling, but there are still certain limitations, such as slow speed and subjectivity, which prompted the exploration of an accurate and objective automatic rating method to become a very urgent need. In response to these problems, this work is based on the multi-feature fusion convolutional neural network (CNN) Light-Smart-Network (LSNet) that has the capability to objectively grade fabric pilling. The fuzzing and pilling data is first set for six kinds of knitted fabrics, woven fabrics, and non-woven fabrics; next, the LSNet network determines the best parameters for objective rating. The network model is composed of a SqueezeNet branch (S1 Branch) and a ShuffleNet branch (S2 Branch), where the S1 Branch uses deep separable convolution (DSC) to improve the Fire module of the SqueezeNet network and adds an attention mechanism that makes the network favor contextual information. An S2 Branch adds short connections to the ShuffleNet network to combine feature maps of different sizes, perform fusion, enhance the network's ability to extract feature map information, and optimize some details. Experiments show that the accuracy of LSNet's objective rating for fabric fuzzing and pilling is 97.386% with a model size of only 5.25 M. Finally, the LSNet model is verified via a heat and feature map, which proves that LSNet has an objective rating for fabric fuzzing and pilling reliability.

Glaucoma is a common ocular condition that can cause significant physiological distress and may lead to vision loss in advanced stages. In clinical practice, segmenting the optic cup and disc regions in fundus images provides key physiological indicators, such as the cup-to-disc ratio and optic nerve fiber layer thickness. These indicators are crucial for clinicians to make accurate diagnoses and informed treatment decisions. Therefore, this research is of significant clinical and societal relevance for the early detection and grading of glaucoma. This thesis introduces a novel network architecture, SMS-U-Net (SE Multi-resolution module Self-attention U-Net), based on the U-Net framework, for joint segmentation of the optic cup and disc in fundus images. The proposed approach scales features from input images and uses the SE (Squeeze-and-Excitation) module to adaptively extract features from each feature map, minimizing information loss. To preserve image details and capture features from multiple receptive fields before downsampling, the conventional convolutional sequence in the encoding path is replaced by a multi-resolution module employing 3 × 3, 5 × 5, and 7 × 7 convolutional kernels. Furthermore, a self-attention mechanism is integrated into the decoder to highlight critical features for segmentation. Experimental results show that the proposed SMS-U-Net method achieves an optic disc Dice coefficient of 0.985 and an IoU of 0.968, as well as an optic cup Dice coefficient of 0.919 and an IoU of 0.851, demonstrating its effectiveness.

The application of deep learning technology in the field of materials science provides a new method for predicting the adsorption energy of high-performance alloy catalysts in hydrogen evolution reactions and material discovery. The activity and selectivity of catalytic materials are mainly influenced by the properties and positions of active sites and adsorption sites. However, current deep learning models have not sufficiently focused on the importance of active atoms and adsorbates, instead placing more emphasis on the overall structure of the catalytic materials. In this paper, the overall molecular graph and a masked graph, which ignores fixed atoms, are separately input into the Masked Graph Transformer (MGT) network to enhance the model's ability to recognize key sites in catalytic reactions. Second, we introduce a nonlinear message-passing mechanism to improve the dot-product attention in the Transformer and capture the directional information on the relative positions of nodes by integrating molecular geometric information through deep tensor products. Subsequently, we constructed the NLMP-TransNet framework, which combines MPNN and Transformer and optimizes the model's learning and prediction capabilities through weight sharing and residual connections. The MGT achieves an error rate of 0.5447 eV on the small data set OC20-Ni, surpassing existing technologies. Ablation studies confirm the necessity of focusing on site features for accurate adsorption energy prediction. Code is available at https://github.com/KristinSun/OCP-MGT.git.

With the application of freeform surface, beam shaping can achieve greater flexibility and precision. Ray mapping is an efficient geometric method for freeform surface design. However, the mapping may not always be integrable, which will result in actual performance falling short of expectations. Additionally, this method often relies on fitting the surface through the calculated coordinates of discrete points, and the fitting accuracy can also affect actual performance. In this study, adaptive weighted particle swarm optimization (AWPSO) is employed to optimize freeform polynomial coefficients to improve beam shaping performance. The initial freeform structure is obtained by non-integrable ray mapping method and fitted into a surface polynomial. Then, the process of optical design is transformed into a multi-dimensional optimization problem within mathematical models. Therefore, performance improvements do not depend on time-consuming ray tracing. A challenging beam shaping example of circle to regular hexagon is given to prove the effectiveness of the method. The results show a significant improvement in the uniformity of the ray spot on the target surface, with the contour becoming sharper and smoother after optimization. From the numerical results, the potential of this method is shown to be used in more irregular beam shaping conditions.

Gastrointestinal stromal tumor (GIST) is a common mesenchymal tumor in the gastrointestinal system. Its treatment outcomes and prognosis are closely related to mutations in the KIT11 gene locus. Therefore, accurately predicting the mutation status of the KIT11 gene locus is crucial for developing personalized treatment plans and improving patient prognosis. We propose a multimodal prediction model with co-attention fusion. This method accurately extracts features from clinical and CT image data using a two-stream interaction module to facilitate information interaction and fusion between modalities, which can effectively predict the KIT11 gene mutation status. Experimental results show that the proposed multimodal model with co-attention fusion achieves an accuracy of 0.840 and an AUC value of 0.862 in predicting KIT11 gene mutations in gastrointestinal stromal tumors.

The mode selection ability of whispering gallery mode (WGM) microcavities is crucial in applications such as sensors, lasers, and nonlinear optics. Though various shapes of microcavities have been studied for mode suppression, single-mode operation is still difficult to realize. Here, we demonstrate a Reuleaux-triangle resonator (RTR) with corner-cuts, which can reconstruct phase space to realize single-mode control. According to classical ray dynamics, the boundary of the RTR is optimized to obtain the stable 9-period islands with the consideration of suppression of standing waves and strong light scattering. The single-mode characteristic of the RTR is experimentally verified under the optimal coupling position, with a Q factor of 1.1 x 104. Our investigation reveals a new thread for mode suppression with potential in the fields of single-mode lasers and nonlinear optics.

Color formulation prediction aims to get different target colors based on a specified color range. It was difficult for existing color matching methods to achieve accurate nonlinear mapping from target colors to formulations due to the overlook of the global and local features of color information. Global featuress represent the long-range variation trends of the input information, while local features represent the numerical characteristics of the input data. To address this issue, a novel feature extraction block based on Transformer and parallel convolutional neural network was proposed for the first time. By integrating this block with a color feature extractor after the fusion of color features, a multimodal color feature fusion network framework was constructed to comprehensively learn the nonlinear mapping between multi-modal inputs and formulations, formulation reflectance curves. This network framework achieved a total loss of 0.0421 and a formulation loss of 0.0391 on test set, indicating more accurate color formula predictions.

The vertical coupling performance induced by 45° micro-mirrors in double-layer polymer optical waveguides is enhanced through CO2 laser smoothing. The optimal double- layer waveguides were designed via ray-tracing simulation. The results show that the minimum vertical coupling loss is 0.28 dB for the upper-layer waveguide and 0.40 dB for the lower-layer waveguide with a core size of 42 µm × 42 µm. A vertical coupling model considering the roughness of 45° micro-mirrors indicates that the surface roughness should be less than 100 nm to ensure a vertical coupling loss of less than 1 dB. The 45° micro-mirrors in double-layer polymer optical waveguides were fabricated using a combination of ultraviolet lithography and stepped laser-ablation method. CO2 laser smoothing at 0.6 W for 40 s reduces the average roughness of the micro-mirrors by 16.67 nm. The vertical coupling loss induced by the micro- mirrors treated with CO2 laser smoothing was measured as 2.89 dB at 850 nm and 1.29 dB at 1310 nm using a 50-µm core multi-mode fiber receiver. This is significantly lower than the average vertical coupling loss of 5.47 dB at 850 nm and 2.50 dB at 1310 nm induced by untreated 45° micro-mirrors. This straightforward and effective method for fabricating and optimizing 45° micro-mirrors holds significant potential for application in vertical coupling within optical printed circuit board (OPCB) interconnection technologies.