Representation Features Research Articles

Relational descriptors for retrieving design features in a B-rep model using the similarity-based retrieval approach

Design features refer to local shapes or regions within a part that perform specific functions such as fastening and force transmission. These design features must be identified from product design results to conduct design verification, manufacturing evaluation, and process planning. Design features are formed by combining various form features, which poses a challenge when using existing methods to retrieve individual features. Therefore, this study introduced a relational descriptor that describes the relational characteristics between topological elements to retrieve design features in boundary representation (B-rep) models. In addition, a method to retrieve design features by combining the relational descriptor with shape descriptors was proposed. Experiments were performed to identify specific design features to validate the proposed method. The experimental results successfully retrieved all the design features included in the B-rep model.

Enhanced battery life prediction with reduced data demand via semi-supervised representation learning

Accurate prediction of the remaining useful life (RUL) is crucial for the design and management of lithium-ion batteries. Although various machine learning models offer promising predictions, one critical but often overlooked challenge is their demand for considerable run-to-failure data for training. Collection of such training data leads to prohibitive testing efforts as the run-to-failure tests can last for years. Here, we propose a semi-supervised representation learning method to enhance prediction accuracy by learning from data without RUL labels. Our approach builds on a sophisticated deep neural network that comprises an encoder and three decoder heads to extract time-dependent representation features from short-term battery operating data regardless of the existence of RUL labels. The approach is validated using three datasets collected from 34 batteries operating under various conditions, encompassing over 19,900 charge and discharge cycles. Our method achieves a root mean squared error (RMSE) within 25 cycles, even when only 1/50 of the training dataset is labelled, representing a reduction of 48% compared to the conventional approach. We also demonstrate the method’s robustness with varying numbers of labelled data and different weights assigned to the three decoder heads. The projection of extracted features in low space reveals that our method effectively learns degradation features from unlabelled data. Our approach highlights the promise of utilising semi-supervised learning to reduce the data demand for reliability monitoring of energy devices.

Combining visual intelligence and social-physical urban features facilitates fine-scale seasonality characterization of urban thermal environments

Traditional methods for assessing urban thermal environments (UTEs) often rely on GIS and remote sensing data, suffering from data limitations in coverage, accuracy, and availability. To address these challenges, we established a novel framework developing visual intelligence simulated indices (VISIs) for improved UTE characterization. This framework combines GIS spatial data and pre-trained large-scale and local-trained GeoAI models to generate visual intelligence, extracting GIS-like visual elements from remote sensing imagery. These GIS-like VISIs comprehensively represent land surface temperature (LST) and urban fabric for cost-effective UTE analysis. We applied our framework to Zhengzhou, China, a city with diverse landscapes. Comparative experiments among Random Forests (RF), Neural Networks (NN), and Convolutional Neural Networks (CNN) demonstrated that RF models exhibited superior performance in LST representation (0.7 test R2 and 0.9 training R2). Interestingly, CNNs performed better than RFs and NNs when less cloud cover (<6 %) was present, suggesting that data diversity is essential for CNN to learn generalizable features for LST representation. For urban fabric representation, feature importance analysis indicated that VISIs (37 %) derived from large-scale GeoAI models outperform spectral bands (20 %). Furthermore, both Pearson correlation and Shapley values showed that VISIs (e.g., building features) better characterize urban heat islands than traditional remote sensing ecological indices (RSEIs) such as NDBI, underscoring the cost-effectiveness of our visual intelligence approach. In summary, our visual intelligence method effectively consolidates UTE seasonality characterization into a predictable scenario even given the unpredictable availability of remote sensing and GIS data, revealing novel perspectives on sensing unseen details for UTEs.

Superpixel perception graph neural network for intelligent defect detection of aero-engine blade

Aero-engine is the core component of aircraft and other spacecraft. The high-speed rotating blades provide power by sucking in air and fully combusting, and various defects will inevitably occur, threatening the operation safety of aero-engine. Therefore, regular inspections are essential for such a complex system. However, existing traditional technology which is borescope inspection is labor-intensive, time-consuming, and experience-dependent. To endow this technology with intelligence, a novel superpixel perception graph neural network (SPGNN) is proposed by utilizing a multi-stage graph convolutional network (MSGCN) for feature extraction and superpixel perception region proposal network (SPRPN) for region proposal. First, to capture complex and irregular textures, the images are transformed into a series of patches, to obtain their graph representations. Then, MSGCN composed of several GCN blocks extracts graph structure features and performs graph information processing at graph level. Last but not least, the SPRPN is proposed to generate perceptual bounding boxes by fusing graph representation features and superpixel perception features. Therefore, the proposed SPGNN always implements feature extraction and information transmission at the graph level in the whole SPGNN pipeline, to alleviate the reduction of receptive field and information loss. To verify the effectiveness of SPGNN, we construct a simulated blade dataset with 3000 images. A public aluminum dataset is also used to validate the performances of different methods. The experimental results demonstrate that the proposed SPGNN has superior performance compared with the state-of-the-art methods.

P2P credit risk management with KG-GNN: a knowledge graph and graph neural network-based approach

Credit risk management is crucial for the credit loan decision-making process on P2P platforms, essential for mitigating credit defaults. The rapid decline of China’s P2P platform market highlights the urgent need for Internet financial institutions to enhance their credit risk management strategies. Previous studies have applied machine learning to assess credit risk; however, their effectiveness is often hampered by a lack of consideration for semantic information and individual correlation. In response, we propose an innovative approach, KG-GNN, which integrates knowledge graph (KG) and graph neural network (GNN). KG-GNN leverages KG to encapsulate semantic information within complex categorical features and explore potential relationships between borrowers. Utilizing GNN, our framework extracts representation features from the KG to build comprehensive and accurate credit risk models. Our findings indicate that KG-GNN not only can predict credit risk more accurately than conventional machine learning models but also improves the stream model performance by over 20% through KG and GNN-based data augmentation techniques. By integrating KG with GNN, our approach enriches the methodologies for credit risk management and can be adapted to other data mining challenges that require processing complex semantic and relational information, thereby enhancing model learning capabilities.

DAT-Net: Deep Aggregation Transformer Network for automatic nuclear segmentation

Detection and segmentation of cell nuclei in hematoxylin and eosin-stained tissue images pose critical clinical challenges, including intricate backgrounds, variable nuclear appearances, overlapping nuclei, and unclear boundaries. These complexities render automated instance segmentation a difficult research area. Thus, this paper proposes a deep aggregation transformer network for automatic nucleus instance segmentation. The method employs a robust Vision Transformer-based backbone to extract informative features from images. To fuse multi-level information, we propose the group aggregation multi-level feature fusion module, which merges low-level features for shape representation and high-level features for pixel edge segmentation, enhancing overall model performance. The depthwise integrated multilayer perceptron module enhances feature learning by combining the global context-dependent representation of MLP and the local representation of convolution through channel separation. This fosters feature learning across different modes, thereby improving network representation. The bidirectional path interaction feature processing module is introduced for multi-scale information extraction and fusion. This module enhances the entire feature pyramid hierarchy by transferring deep semantic information to shallow layers with robust spatial information via bidirectional paths. The Large kernel Attention module at each layer extensively improves feature extraction through a large receptive field. Our approach outperforms state-of-the-art models, including StarDIST, CPP-Net, TSFD-Net and TransNuSeg, by 7.8%, 6.3%, 1.5% and 3.5% in average panoptic quality for different cell nuclei segmentation categories across tissues of the Pannuke dataset. Furthermore, DAT-Net outperforms all other state-of-the-art models in terms of the F1 score for the five classes of cell nuclei in the PanNuke dataset.

An effective multi-modal adaptive contextual feature information fusion method for Chinese long text classification

Chinese long text classification plays a vital role in Natural Language Processing. Compared to Chinese short texts, Chinese long texts contain more complex semantic feature information. Furthermore, the distribution of these semantic features is uneven due to the varying lengths of the texts. Current research on Chinese long text classification models primarily focuses on enhancing text semantic features and representing Chinese long texts as graph-structured data. Nonetheless, these methods are still susceptible to noise information and tend to overlook the deep semantic information in long texts. To address the above challenges, this study proposes a novel and effective method called MACFM, which introduces a deep feature information mining method and an adaptive modal feature information fusion strategy to learn the semantic features of Chinese long texts thoroughly. First, we present the DCAM module to capture complex semantic features in Chinese long texts, allowing the model to learn detailed high-level representation features. Then, we explore the relationships between word vectors and text graphs, enabling the model to capture abundant semantic information and text positional information from the graph. Finally, we develop the AMFM module to effectively combine different modal feature representations and eliminate the unrelated noise information. The experimental results on five Chinese long text datasets show that our method significantly improves the accuracy of Chinese long text classification tasks. Furthermore, the generalization experiments on five English datasets and the visualized results demonstrate the effectiveness and interpretability of the MACFM model.

A Joint Network of Edge-Aware and Spectral-Spatial Feature Learning for Hyperspectral Image Classification.

Hyperspectral image (HSI) classification is a vital part of the HSI application field. Since HSIs contain rich spectral information, it is a major challenge to effectively extract deep representation features. In existing methods, although edge data augmentation is used to strengthen the edge representation, a large amount of high-frequency noise is also introduced at the edges. In addition, the importance of different spectra for classification decisions has not been emphasized. Responding to the above challenges, we propose an edge-aware and spectral-spatial feature learning network (ESSN). ESSN contains an edge feature augment block and a spectral-spatial feature extraction block. Firstly, in the edge feature augment block, the edges of the image are sensed, and the edge features of different spectral bands are adaptively strengthened. Then, in the spectral-spatial feature extraction block, the weights of different spectra are adaptively adjusted, and more comprehensive depth representation features are extracted on this basis. Extensive experiments on three publicly available hyperspectral datasets have been conducted, and the experimental results indicate that the proposed method has higher accuracy and immunity to interference compared to state-of-the-art (SOTA) method.

Graph Feature Refinement and Fusion in Transformer for Structural Damage Detection.

Structural damage detection is of significance for maintaining the structural health. Currently, data-driven deep learning approaches have emerged as a highly promising research field. However, little progress has been made in studying the relationship between the global and local information of structural response data. In this paper, we have presented an innovative Convolutional Enhancement and Graph Features Fusion in Transformer (CGsformer) network for structural damage detection. The proposed CGsformer network introduces an innovative approach for hierarchical learning from global to local information to extract acceleration response signal features for structural damage representation. The key advantage of this network is the integration of a graph convolutional network in the learning process, which enables the construction of a graph structure for global features. By incorporating node learning, the graph convolutional network filters out noise in the global features, thereby facilitating the extraction to more effective local features. In the verification based on the experimental data of four-story steel frame model experiment data and IASC-ASCE benchmark structure simulated data, the CGsformer network achieved damage identification accuracies of 92.44% and 96.71%, respectively. It surpassed the existing traditional damage detection methods based on deep learning. Notably, the model demonstrates good robustness under noisy conditions.

Application of an ensemble CatBoost model over complex dataset for vehicle classification.

The classification of vehicles presents notable challenges within the domain of image processing. Traditional models suffer from inefficiency, prolonged training times for datasets, intricate feature extraction, and variable assignment complexities for classification. Conventional methods applied to categorize vehicles from extensive datasets often lead to errors, misclassifications, and unproductive outcomes. Consequently, leveraging machine learning techniques emerges as a promising solution to tackle these challenges. This study adopts a machine learning approach to alleviate image misclassifications and manage large quantities of vehicle images effectively. Specifically, a contrast enhancement technique is employed in the pre-processing stage to highlight pixel values in vehicle images. In the feature segmentation stage, Mask-R-CNN is utilized to categorize pixels into predefined classes. VGG16 is then employed to extract features from vehicle images, while an autoencoder aids in selecting features by learning non-linear input features and compressing representation features. Finally, the CatBoost (CB) algorithm is implemented for vehicle classification (VC) in diverse critical environments, such as inclement weather, twilight, and instances of vehicle blockage. Extensive experiments are conducted using different large-scale datasets with various machine learning platforms. The findings indicate that CB (presumably a specific method or algorithm) attains the highest level of performance on the large-scale dataset named UFPR-ALPR, with an accuracy rate of 98.89%.

Handwritten Signature Verification Method Using Convolutional Neural Network

Automatic signature verification methods play a significant role in providing a secure and authenticated handwritten signature in many applications, to prevent forgery problems, specifically institutions of finance, and transections of legal papers, etc. There are two types of handwritten signature verification methods: online verification (dynamic) and offline verification (static) methods. Besides, signature verification approaches can be categorized into two styles: writer dependent (WD), and writer independent (WI) styles. Offline signature verification methods demands a high representation features for the signature image. However, lots of studies have been proposed for WI offline signature verification. Yet, there is necessity to improve the overall accuracy measurements. Therefore, a proved solution in this paper is depended on deep learning via convolutional neural network (CNN) for signature verification and optimize the overall accuracy measurements. The introduced model is trained on English signature dataset. For model evaluation, the deployed model is utilized to make predictions on new data of Arabic signature dataset to classify whether the signature is real or forged. The overall obtained accuracy is 95.36\% based on validation dataset.

ER-NeRF++: Efficient region-aware Neural Radiance Fields for high-fidelity talking portrait synthesis

Despite conditional Neural Radiance Fields (NeRF) achieving great success in modeling audio-driven talking portraits, the generation quality is increasingly hampered by the lack of efficient use of space information. This paper presents ER-NeRF, a novel conditional NeRF-based architecture for talking portrait synthesis, and its variant version ER-NeRF++ to concurrently achieve fast convergence, real-time rendering, and state-of-the-art performance with small model size. Inspired by the unequal contribution of spatial regions, we propose two modules in ER-NeRF to guide the talking portrait modeling: (1) A compact and expressive Tri-Plane Hash Representation to improve the accuracy of dynamic head reconstruction by pruning empty spatial regions with three planar hash encoders. (2) A Region Attention Module for the audio–visual feature fusion, including a novel cross-modal attention mechanism to connect audio features with different spatial regions explicitly for local motion priors. Additionally, to tackle the difficulty in learning large facial motions, we propose a deformable variant ER-NeRF++ by including a Deformation Grid Transformer to enable the reuse of cross-regional spatial features for large motion representation. Compared to ER-NeRF, our ER-NeRF++ framework achieves a significant improvement in facial motion quality while maintaining the ability of fast training and real-time rendering. For the torso part, a directAdaptive Pose Encoding is introduced to simplify the pose information for a better head-torso connection. Extensive experiments demonstrate that both of our proposed frameworks can efficiently render lifelike talking portrait videos with rich realistic details, performing better in image quality and audio-lip synchronization compared to previous methods.

Sub-features orthogonal decoupling: Detecting bone wall absence via a small number of abnormal examples for temporal CT images

The absence of bone wall located in the jugular bulb and sigmoid sinus of the temporal bone is one of the important reasons for pulsatile tinnitus. Automatic and accurate detection of these abnormal singes in CT slices has important theoretical significance and clinical value. Due to the shortage of abnormal samples, imbalanced samples, small inter-class differences, and low interpretability, existing deep-learning methods are greatly challenged. In this paper, we proposed a sub-features orthogonal decoupling model, which can effectively disentangle the representation features into class-specific sub-features and class-independent sub-features in a latent space. The former contains the discriminative information, while, the latter preserves information for image reconstruction. In addition, the proposed method can generate image samples using category conversion by combining the different class-specific sub-features and the class-independent sub-features, achieving corresponding mapping between deep features and images of specific classes. The proposed model improves the interpretability of the deep model and provides image synthesis methods for downstream tasks. The effectiveness of the method was verified in the detection of bone wall absence in the temporal bone jugular bulb and sigmoid sinus.

An End-to-end Deep Clustering Method with Consistency and Complementarity Attention Mechanism for Multisensor Fault Diagnosis

Deep clustering methods have found successful applications in single-sensor data fault diagnosis. However, most of these methods employ separate optimization strategies that overlook the interaction between feature learning and clustering. Moreover, conventional deep learning methods for fault diagnosis often disregard the consistent and complementary information inherent in the multisensor data, resulting in unsatisfactory multisensor fault diagnosis performance. In this study, we introduce a novel End-to-end Deep Clustering Method with Consistency and Complementarity Attention Mechanism, termed EDCM-CCAM, tailored for multisensor fault diagnosis. Firstly, multiple deep autoencoder networks are utilized to concurrently extract the deep representation features from various sensor inputs. Secondly, we introduce a Consistency and Complementarity Attention Mechanism (CCAM) to facilitate multisensor feature fusion, accompanied by the design of two distinct loss functions to fully exploit the consistent and complementary information within multisensor data. Finally, fault pattern recognition in multisensor data is accomplished through Kullback-Leibler (KL) divergence-based clustering, while a joint optimization strategy is employed to simultaneously optimize all components of the EDCM-CCAM. The efficacy of the proposed method is validated on a gearbox dataset, demonstrating superior performance in multisensor fault diagnosis compared to alternative methods.

Global semantic-guided graph attention network for Siamese tracking with ranking loss

Most Siamese trackers based on graph attention models focus on the topology between nodes when extracting features. These approaches ignore contextual information regarding the overall structure of the target, resulting in a diminished capacity for target discrimination. In this paper, to resolve this issue, we propose a global semantic-guided graph attention network for Siamese tracking. Firstly, in order to balance the node constraints with the overall information of the target, a global semantic awareness module (GSA) is proposed. It utilizes the attention strategy to focus on the overall characteristics of the target, aiming to mine the relationship of the target context. Incorporating it into the semantic-guided graph attention network through cross-correlation operations can avoid the problem of focusing only the information between nodes without being able to effectively characterize the target. Secondly, to enhance the accuracy of single-point feature regression results, a novel extreme point feature aggregation module (EPA) is proposed. We obtain feature information of the extreme points on the bounding box and combine them with single-point features for the bounding box feature representation to improve the regression accuracy. Finally, to establish a connection or bridge between the classification and regression subnetworks, a categorical regression alignment network is designed. An IOU-guided ranking loss is introduced to align the classification confidence with the IOU of the corresponding positive sample localization prediction to solve the problem of misalignment of the classification branch with the regression branch effectively. Experiments conducted on five challenging benchmarks datasets, namely GOT-10K, OTB-100, LaSOT, UAV123, and TC128, demonstrate that our approach exhibits superior performance compared to other tracking methods in terms of tracking accuracy and execution efficiency.

Frequency Spectrum Is More Effective for Multimodal Representation and Fusion: A Multimodal Spectrum Rumor Detector

Multimodal content, such as mixing text with images, presents significant challenges to rumor detection in social media. Existing multimodal rumor detection has focused on mixing tokens among spatial and sequential locations for unimodal representation or fusing clues of rumor veracity across modalities. However, they suffer from less discriminative unimodal representation and are vulnerable to intricate location dependencies in the time-consuming fusion of spatial and sequential tokens. This work makes the first attempt at multimodal rumor detection in the frequency domain, which efficiently transforms spatial features into the frequency spectrum and obtains highly discriminative spectrum features for multimodal representation and fusion. A novel Frequency Spectrum Representation and fUsion network (FSRU) with dual contrastive learning reveals the frequency spectrum is more effective for multimodal representation and fusion, extracting the informative components for rumor detection. FSRU involves three novel mechanisms: utilizing the Fourier transform to convert features in the spatial domain to the frequency domain, the unimodal spectrum compression, and the cross-modal spectrum co-selection module in the frequency domain. Substantial experiments show that FSRU achieves satisfactory multimodal rumor detection performance.

MolTailor: Tailoring Chemical Molecular Representation to Specific Tasks via Text Prompts

Deep learning is now widely used in drug discovery, providing significant acceleration and cost reduction. As the most fundamental building block, molecular representation is essential for predicting molecular properties to enable various downstream applications. Most existing methods attempt to incorporate more information to learn better representations. However, not all features are equally important for a specific task. Ignoring this would potentially compromise the training efficiency and predictive accuracy. To address this issue, we propose a novel approach, which treats language models as an agent and molecular pretraining models as a knowledge base. The agent accentuates task-relevant features in the molecular representation by understanding the natural language description of the task, just as a tailor customizes clothes for clients. Thus, we call this approach MolTailor. Evaluations demonstrate MolTailor's superior performance over baselines, validating the efficacy of enhancing relevance for molecular representation learning. This illustrates the potential of language model guided optimization to better exploit and unleash the capabilities of existing powerful molecular representation methods. Our code and appendix are available at https://github.com/SCIR-HI/MolTailor.

Uncertainty-Aware Yield Prediction with Multimodal Molecular Features

Predicting chemical reaction yields is pivotal for efficient chemical synthesis, an area that focuses on the creation of novel compounds for diverse uses. Yield prediction demands accurate representations of reactions for forecasting practical transformation rates. Yet, the uncertainty issues broadcasting in real-world situations prohibit current models to excel in this task owing to the high sensitivity of yield activities and the uncertainty in yield measurements. Existing models often utilize single-modal feature representations, such as molecular fingerprints, SMILES sequences, or molecular graphs, which is not sufficient to capture the complex interactions and dynamic behavior of molecules in reactions. In this paper, we present an advanced Uncertainty-Aware Multimodal model (UAM) to tackle these challenges. Our approach seamlessly integrates data sources from multiple modalities by encompassing sequence representations, molecular graphs, and expert-defined chemical reaction features for a comprehensive representation of reactions. Additionally, we address both the model and data-based uncertainty, refining the model's predictive capability. Extensive experiments on three datasets, including two high throughput experiment (HTE) datasets and one chemist-constructed Amide coupling reaction dataset, demonstrate that UAM outperforms the state-of-the-art methods. The code and used datasets are available at https://github.com/jychen229/Multimodal-reaction-yield-prediction.

Heterogeneous Graph-Convolution-Network-Based Short-Text Classification

With the development of online interactive media platforms, a large amount of short text has appeared on the internet. Determining how to classify these short texts efficiently and accurately is of great significance. Graph neural networks can capture information dependencies in the entire short-text corpus, thereby enhancing feature expression and improving classification accuracy. However, existing works have overlooked the role of entities in these short texts. In this paper, we propose a heterogeneous graph-convolution-network-based short-text classification (SHGCN) method that integrates heterogeneous graph convolutional neural networks of text, entities, and words. Firstly, the model constructs a graph network of the text and extracts entity nodes and word nodes. Secondly, the relationship of the graph nodes in the heterogeneous graphs is determined by the mutual information between the words, the relationship between the documents and words, and the confidence between the words and entities. Then, the feature is represented through a word graph and combined with its BERT embedding, and the word feature is strengthened through BiLstm. Finally, the enhanced word features are combined with the document graph representation features to predict the document categories. To verify the performance of the model, experiments were conducted on the public datasets AGNews, R52, and MR. The classification accuracy of SHGCN reached 88.38%, 93.87%, and 82.87%, respectively, which is superior to that of some existing advanced classification methods.

Detection Transformer with Multi-Scale Fusion Attention Mechanism for Aero-Engine Turbine Blade Cast Defect Detection Considering Comprehensive Features.

Casting defects in turbine blades can significantly reduce an aero-engine's service life and cause secondary damage to the blades when exposed to harsh environments. Therefore, casting defect detection plays a crucial role in enhancing aircraft performance. Existing defect detection methods face challenges in effectively detecting multi-scale defects and handling imbalanced datasets, leading to unsatisfactory defect detection results. In this work, a novel blade defect detection method is proposed. This method is based on a detection transformer with a multi-scale fusion attention mechanism, considering comprehensive features. Firstly, a novel joint data augmentation (JDA) method is constructed to alleviate the imbalanced dataset issue by effectively increasing the number of sample data. Then, an attention-based channel-adaptive weighting (ACAW) feature enhancement module is established to fully apply complementary information among different feature channels, and further refine feature representations. Consequently, a multi-scale feature fusion (MFF) module is proposed to integrate high-dimensional semantic information and low-level representation features, enhancing multi-scale defect detection precision. Moreover, R-Focal loss is developed in an MFF attention-based DEtection TRansformer (DETR) to further solve the issue of imbalanced datasets and accelerate model convergence using the random hyper-parameters search strategy. An aero-engine turbine blade defect X-ray (ATBDX) image dataset is applied to validate the proposed method. The comparative results demonstrate that this proposed method can effectively integrate multi-scale image features and enhance multi-scale defect detection precision.