Search

Machine learning is a promising method to model production processes and predict product quality. It is challenging to accurately model complex systems due to data scarcity, as mass customisation leads to various high-variety low-volume products. This study conceptualised knowledge accumulation, extraction, and transfer (KAET) to exploit the knowledge embedded in similar entities to address data scarcity. A sequential cross-product KAET (SeqTrans) is proposed to conduct KAET, integrating data preparation and preprocessing, feature selection (FS), feature learning (FL), and transfer learning (TL). The FS and FL modules conduct knowledge extraction and help address various practical challenges such as changing operating conditions and unbalanced datasets. In this paper, sequential TL is introduced to production modelling to conduct knowledge transfer among multiple entities. The first case study of auxetic material performance prediction demonstrates the effectiveness of sequential TL. Compared with conventional TL, sequential TL can achieve the same test mean square errors with 300 fewer training examples when facing data scarcity. In the second case study, balancing anomaly detection models were constructed for two gas turbines in the same series using real-world production data. With SeqTrans, the F1-score of the anomaly detection model of the data-poor engine was improved from 0.769 to 0.909. [ABSTRACT FROM AUTHOR]

The dynamic characteristics (i.e. autocorrelation and cross-correlation) in modern industrial systems have raised great challenges to process fault detection. To cope with the dynamics and uncertainty of dynamic industrial processes, this paper proposes a new process control method, dynamic convolutional gated recurrent unit attention auto-encoder (DCGRUA-AE) for fault detection in dynamic processes. Firstly, DCGRUA-AE integrates a convolutional gated recurrent unit (CGRU) with a local convolution layer to learn both global and local features of dynamic process data in an unsupervised fashion. Secondly, a dual attention module is embedded in the deep network to preserve effective features. Finally, DeconvGRU combined with a dense layer is used as the encoder to reconstruct the original process data. Two statistics (i.e. T-square (T2) and squared prediction error (SPE)) based on DCGRUA-AE are used to set the control limits for fault detection. The feasibility and superiority of DCGRUA-AE-based fault detection method have been verified on four industrial processes. The experimental results indicate that the CGRU and dual attention mechanism can significantly improve the fault detection performance of DCGRUA-AE in dynamic processes. The hybrid of CGRU, attention mechanism and auto-encoder provides a new method for fault detection in dynamic industrial processes. [ABSTRACT FROM AUTHOR]

Dynamic latent scale GAN is an architecture‐agnostic encoder‐based generative model inversion method. This paper introduces a method to efficiently integrate perceptual VAE into dynamic latent scale GAN to improve the performance of dynamic latent scale GAN. When dynamic latent scale GAN is trained with a normal i.i.d. latent random variable and the latent encoder is integrated into the discriminator, a sum of a predicted latent random variable of real data and a scaled normal noise follows the normal i.i.d. random variable. Since this random variable is paired with real data and follows the latent random variable, it can be used for both VAE and GAN training. Furthermore, by considering the intermediate layer output of the discriminator as the feature encoder output, the VAE can be trained to minimise the perceptual reconstruction loss. The forward propagation & backpropagation for minimising this perceptual reconstruction loss can be integrated with those of GAN training. Therefore, the proposed method does not require additional computations compared to typical GAN or dynamic latent scale GAN. Integrating perceptual VAE to dynamic latent scale GAN improved the generative and inversion performance of the model. [ABSTRACT FROM AUTHOR]

As image–text matching (a critical task in the field of computer vision) links cross-modal data, it has captured extensive attention. Most of the existing methods intended for matching images and texts explore the local similarity levels between images and sentences to align images with texts. Even though this fine-grained approach has remarkable gains, how to further mine the deep semantics between data pairs and focus on the essential semantics in data remains to be quested. In this work, a new semantic filtering and adaptive approach (FAAR) was proposed to ease the above problem. To be specific, the filtered attention (FA) module selectively focuses on typical alignments with the interference of meaningless comparisons eliminated. Next, the adaptive regulator (AR) further adjusts the attention weights of key segments for filtered regions and words. The superiority of our proposed method was validated by a number of qualitative experiments and analyses on the Flickr30K and MSCOCO data sets. [ABSTRACT FROM AUTHOR]

Copyright of Journal of Computer Engineering & Applications is the property of Beijing Journal of Computer Engineering & Applications Journal Co Ltd. and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

While multimodal remote sensing data processing has garnered growing interests in the geoscience field, how to effectively extract, represent, and fuse heterogeneous features remains challenging. In this article, a novel heterogeneous feature learning network (HFLN) architecture is investigated for multimodal remote sensing collaborative classification, which takes both advantages of long-range feature modelling capability of transformer as well as local feature extraction of convolutional neural network (CNN). First, the global spectral dependencies are extracted from hyperspectral images using the spectral transformer structure. Then, local spatial invariant characteristics are extracted from multimodal remote sensing images by convolutional operation. Next, heterogeneous spectral and spatial characteristics, having significant structural and semantic differences, are dynamically integrated through the feature coupling module in an interactive manner. Finally, the multi-stage network architecture is utilized for extracting hierarchical characteristics, where the numbers of feature maps in CNN and heads in transformer are gradually increased to represent more complex features. Four benchmark remote sensing datasets (i.e. hyperspectral images and multimodal remote sensing datasets) are utilized for experimental analysis, and extensive experimental results have certified the progressiveness of the investigated approach. [ABSTRACT FROM AUTHOR]

Accurate and timely diagnosis of COVID-like virus is of paramount importance for lifesaving. In this work, deep learning techniques are applied to lung CT image segmentation for accurate disease diagnosis. We discuss the limitations of current diagnostic methods, such as RT-PCR, and highlights the advantages of deep learning, including its ability to automatically learn features and handle complex lesion morphology and texture. We, therefore, propose a novel deep learning framework, GAHFNet, specifically designed for automatic segmentation of COVID-19 lung CT images. The proposed method addresses the challenges in lung CT image segmentation, such as the complex image structure and difficulties of distinguishing COVID-19 pneumonia lesions from other pathologies. We provide the detailed description of the proposed GAHFNet. Finally, comprehensive experiments are carried out to evaluate the performance of GAHFNet, and the proposed method outperforms other traditional and the state-of-the-art methods in various evaluation metrics, demonstrating the effectiveness and the efficiency of the proposed method in this task. GAHFNet is able to facilitate the application of artificial intelligence in COVID-19 diagnosis and achieve accurate automatic segmentation of infected areas in COVID-19 lung CT images. [ABSTRACT FROM AUTHOR]

Abstract As image–text matching (a critical task in the field of computer vision) links cross-modal data, it has captured extensive attention. Most of the existing methods intended for matching images and texts explore the local similarity levels between images and sentences to align images with texts. Even though this fine-grained approach has remarkable gains, how to further mine the deep semantics between data pairs and focus on the essential semantics in data remains to be quested. In this work, a new semantic filtering and adaptive approach (FAAR) was proposed to ease the above problem. To be specific, the filtered attention (FA) module selectively focuses on typical alignments with the interference of meaningless comparisons eliminated. Next, the adaptive regulator (AR) further adjusts the attention weights of key segments for filtered regions and words. The superiority of our proposed method was validated by a number of qualitative experiments and analyses on the Flickr30K and MSCOCO data sets.

Weak target signal processing is the cornerstone and prerequisite for radar to achieve excellent detection performance. In complex practical applications, due to strong clutter interference, weak target signals, unclear image features, and difficult effective feature extraction, weak target detection and recognition have always been challenging in the field of radar processing. Conventional model-based processing methods do not accurately match the actual working background and target characteristics, leading to weak universality. Recently, deep learning has made significant progress in the field of radar intelligent information processing. By building deep neural networks, deep learning algorithms can automatically learn feature representations from a large amount of radar data, improving the performance of target detection and recognition. This article systematically reviews and summarizes recent research progress in the intelligent processing of weak radar targets in terms of signal processing, image processing, feature extraction, target classification, and target recognition. This article discusses noise and clutter suppression, target signal enhancement, low- and high-resolution radar image and feature processing, feature extraction, and fusion. In response to the limited generalization ability, single feature expression, and insufficient interpretability of existing intelligent processing applications for weak targets, this article underscores future developments from the aspects of small sample object detection (based on transfer learning and reinforcement learning), multidimensional and multifeature fusion, network model interpretability, and joint knowledge- and data-driven processing.

This paper provides a review on representation learning for videos. We classify recent spatio-temporal feature learning methods for sequential visual data and compare their pros and cons for general video analysis. Building effective features for videos is a fundamental problem in computer vision tasks involving video analysis and understanding. Existing features can be generally categorized into spatial and temporal features. Their effectiveness under variations of illumination, occlusion, view and background are discussed. Finally, we discuss the remaining challenges in existing deep video representation learning studies. [ABSTRACT FROM AUTHOR]

SAR image target recognition relies heavily on a large number of annotated samples, making it difficult to classify the unseen class targets. Due to the lack of effective category auxiliary information, the current zero-shot target recognition methods for SAR images are limited to inferring only one unseen class rather than classifying multiple unseen classes. To address this issue, a conditional generative network with the category features from the simulated images for zero-shot SAR target recognition is proposed in this paper. Firstly, the deep features are extracted from the simulated images and fused into the category features that characterize the entire class. Then, a conditional VAE-GAN network is constructed to generate the feature instances of the unseen classes. The high-level semantic information shared in the category features aids in generalizing the mapping learned from the seen classes to the unseen classes. Finally, the generated features of the unseen classes are used to train a classifier that can classify the real unseen images. The classification accuracies for the targets of the three unseen classes based on the proposed method can reach 99.80 ± 1.22% and 71.57 ± 2.28% with the SAMPLE dataset and the MSTAR dataset, respectively. The advantage and validity of the proposed architecture are indicated with a small number of the seen classes and a small amount of the training data. Furthermore, the proposed method can be extended to generalized zero-shot recognition. [ABSTRACT FROM AUTHOR]

Existing point cloud feature learning networks often learn high-semantic point features representing the global context by incorporating sampling, neighborhood grouping, neighborhood-wise feature learning, and feature aggregation. However, this process may result in a substantial loss of granular information due to the sampling operation and the widely-used max pooling feature aggregation, which neglects information from non-maximum point features. Consequently, the resulting high-semantic point features could be insufficient to represent the local context, hindering the network's ability to distinguish fine shapes. To address this problem, we propose PointStack, a novel point cloud feature learning network that utilizes multi-resolution feature learning and learnable pooling (LP). PointStack aggregates point features of various resolutions across multiple layers to capture both high-semantic and high-resolution information. The LP function calculates the weighted sum of multi-resolution point features through an attention mechanism with learnable queries, enabling the extraction of all available information. As a result, PointStack can effectively represent both global and local contexts, allowing the network to comprehend both the global structure and local shape details. PointStack outperforms various existing feature learning networks for shape classification and part segmentation on the ScanObjectNN and ShapeNetPart datasets, achieving 87.2% overall accuracy and instance mIoU. [ABSTRACT FROM AUTHOR]

Finding the most interesting areas of an image is the aim of saliency detection. Conventional methods based on low-level features rely on biological cues like texture and color. These methods, however, have trouble with processing complicated or low-contrast images. In this paper, we introduce a deep neural network-based saliency detection method. First, using semantic segmentation, we construct a pixel-level model that gives each pixel a saliency value depending on its semantic category. Next, we create a region feature model by combining both hand-crafted and deep features, which extracts and fuses the local and global information of each superpixel region. Third, we combine the results from the previous two steps, along with the over-segmented superpixel images and the original images, to construct a multi-level feature model. We feed the model into a deep convolutional network, which generates the final saliency map by learning to integrate the macro and micro information based on the pixels and superpixels. We assess our method on five benchmark datasets and contrast it against 14 state-of-the-art saliency detection algorithms. According to the experimental results, our method performs better than the other methods in terms of F-measure, precision, recall, and runtime. Additionally, we analyze the limitations of our method and propose potential future developments. [ABSTRACT FROM AUTHOR]

Anomaly detection, a critical aspect of data analysis and cybersecurity, aims to identify unusual patterns that deviate from the expected norm. In this study, we propose a hybrid approach that combines the strengths of Autoencoder neural networks and Multiclass Support Vector Machines (SVM) for robust anomaly detection. The Autoencoder is utilized for feature learning and extraction, capturing intricate patterns in the data, while the Multiclass SVM provides a discriminative classification mechanism to distinguish anomalies from normal patterns. Specifically, the Autoencoder is trained on normal data to acquire a compact and efficient representation of the underlying patterns, with the reconstruction errors serving as indicative measures of anomalies. Concurrently, a Multiclass SVM is trained to classify instances into multiple classes, including an anomaly class. The anomaly scores from the Autoencoder and the decision function of the Multiclass SVM, along with that of the Random Forest Neural Network (AE-RFNN), are combined, leveraging their complementary strengths. A thresholding mechanism is then employed to classify instances as normal or anomalous based on the combined scores. The performance of the hybrid model is evaluated using standard metrics such as precision, recall, F1-score, and the area under the Receiver Operating Characteristic (ROC) curve. The proposed hybrid anomaly detection approach demonstrates effectiveness in capturing complex patterns and discerning anomalies across diverse datasets. Additionally, the model offers flexibility for adaptation to evolving data distributions. This study contributes to the advancement of anomaly detection methodologies by presenting a hybrid solution that combines feature learning and discriminative classification for improved accuracy and generalization. [ABSTRACT FROM AUTHOR]

Convolutional neural networks have been effective in several applications, arising as a promising supporting tool in a relevant Dermatology problem: skin cancer diagnosis. However, generalizing well can be difficult when little training data is available. The fine-tuning transfer learning strategy has been employed to differentiate properly malignant from non-malignant lesions in dermoscopic images. Fine-tuning a pre-trained network allows one to classify data in the target domain, occasionally with few images, using knowledge acquired in another domain. This work proposes eight fine-tuning settings based on convolutional networks previously trained on ImageNet that can be employed mainly in limited data samples to reduce overfitting risk. They differ on the architecture, the learning rate and the number of unfrozen layer blocks. We evaluated the settings in two public datasets with 104 and 200 dermoscopic images. By finding competitive configurations in small datasets, this paper illustrates that deep learning can be effective if one has only a few dozen malignant and non-malignant lesion images to study and differentiate in Dermatology. The proposal is also flexible and potentially useful for other domains. In fact, it performed satisfactorily in an assessment conducted in a larger dataset with 746 computerized tomographic images associated with the coronavirus disease. [ABSTRACT FROM AUTHOR]

The authentication of writers through handwritten text stands as a biometric technique with considerable practical importance in the field of document forensics and literary history. The verification process involves a meticulous examination of the questioned handwriting in comparison to the genuine handwriting of a known writer, aiming to determine whether a shared authorship exists. In real-world scenarios, writer verification based on the handwritten text presents more challenges compared to signatures. Signatures typically consist of fixed designs chosen by signers, whereas textual content can vary and encompass a diverse set of letters, numbers, and punctuation marks. Moreover, verifying a writer based on limited handwritten texts, such as a single word, is recognized as one of authentication's open and challenging aspects. In this paper, we propose a Customized Siamese Convolutional Neural Network (CSCNN) for offline writer verification based on handwritten words. Additionally, a combined loss function is employed to achieve more accurate discrimination between the handwriting styles of different writers. The designed model is trained with pairs of images, each comprising one authentic and one questioned handwritten word. The effectiveness of the proposed model is substantiated through experimental results obtained from two well-known datasets in both English and Arabic, IAM and IFN/ENIT. These results underscore the efficiency and performance of our model across diverse linguistic contexts. [ABSTRACT FROM AUTHOR]

The goal of visual place recognition (VPR) is to determine the location of a query image by identifying its place in a collection of image databases. Visual sensor technologies are crucial for visual place recognition as they allow for precise identification and location of query images within a database. Global descriptor-based VPR methods face the challenge of accurately capturing the local specific regions within a scene; consequently, it leads to an increasing probability of confusion during localization in such scenarios. To tackle feature extraction and feature matching challenges in VPR, we propose a modified patch-NetVLAD strategy that includes two new modules: a context-aware patch descriptor and a context-aware patch matching mechanism. Firstly, we propose a context-driven patch feature descriptor to overcome the limitations of global and local descriptors in visual place recognition. This descriptor aggregates features from each patch's surrounding neighborhood. Secondly, we introduce a context-driven feature matching mechanism that utilizes cluster and saliency context-driven weighting rules to assign higher weights to patches that are less similar to densely populated or locally similar regions for improved localization performance. We further incorporate both of these modules into the patch-NetVLAD framework, resulting in a new approach called contextual patch-NetVLAD. Experimental results are provided to show that our proposed approach outperforms other state-of-the-art methods to achieve a Recall@10 score of 99.82 on P i t t s b u r g h 30 k , 99.82 on FMDataset, and 97.68 on our benchmark dataset. [ABSTRACT FROM AUTHOR]

Lysine glycation is one of the most significant protein post-translational modifications, which changes the properties of the proteins and causes them to be dysfunctional. Accurately identifying glycation sites helps to understand the biological function and potential mechanism of glycation in disease treatments. Nonetheless, the experimental methods are ordinarily inefficient and costly, so effective computational methods need to be developed. In this study, we proposed the new model called iGly-IDN based on the improved densely connected convolutional networks (DenseNet). First, one hot encoding was adopted to obtain the original feature maps. Afterward, the improved DenseNet was adopted to capture feature information with the importance degrees during the feature learning. According to the experimental results, Acc reaches 66%, and Mathews correlation coefficient reaches 0.33 on the independent testing data set, which indicates that the iGly-IDN can provide more effective glycation site identification than the current predictors. [ABSTRACT FROM AUTHOR]

Features of data are much critical to the classification. However, when only small data are available, suitable features can not be easily obtained, easily leading to the bad classification performance. This paper propose a novel approach to automatically learns features from the irrelevant domain with much discriminative features for the given classification task. It first computes as the learning objectives the central vectors of each class in the irrelevant domain, and then uses machine learning method to automatically learn features for each sample in the target domain from these objectives. The merits of our method lie in that unlike the transfer learning, our method does not require the similarity between two domains. It can learn features from much discriminative domains. Its learned features are not limited to its original ones, unlike feature selection and feature extraction methods, so that the classification performance with the learned features can be better. Finally, our method is much general, simple, and efficient. Lots of experimental results validated the proposed method. [ABSTRACT FROM AUTHOR]

Tool condition monitoring (TCM) is crucial to ensure good quality products and avoid downtime. Machine learning has proven to be vital for TCM. However, existing works are predominately based on supervised learning, which hinders their applicability in real-world manufacturing settings, where data labeling is cumbersome and costly with in-service machines. Additionally, the existing unsupervised solutions mostly handle binary decision-based TCM which is unable to fully reflect the dynamics of tool wear progression. To address these issues, we propose different unsupervised and semisupervised five-class tool wear recognition frameworks to handle fully unlabeled and partially labeled data, respectively. The underlying methods include Laplacian score, sparse autoencoder (SAE), stacked SAE (SSAE), self-organizing map, Softmax, support vector machine, and random forest. For the semisupervised frameworks, we considered designs where labeled data influence only feature learning, classifier building, or both. We also investigated different training configurations of SSAE regarding the supervision level. We applied the frameworks on two run-to-failure datasets of milling tools, recorded using a microphone and an accelerometer. Single sensor and multisensor data under different percentages of labeled training data were considered in the evaluation. The results showed which of the frameworks led to the best predictive performance under which data settings, and highlighted the significance of sensor fusion and discriminative feature representations in combating the unavailability and scarcity of labels, among other findings. The highest macro-F1 achieved for the two datasets with fully unlabeled data reached 87.52% and 75.80%, respectively, and over 90% when only 25% of the training observations were labeled.

The joint use of hyperspectral image (HSI) and light detection and ranging (LiDAR) data has gained significant performance on land-cover classification. Although spatial–spectral feature learning methods based on convolutional neural networks and transformer networks have achieved prominent advances, contextual information described by fixed convolutional kernels and all self-attention heads selected have limited ability to characterize the detailed information and nonredundant features of land-covers on multimodal data. In this article, a multiscale head selection transformer (MHST) network, is proposed to fully explore detailed and nonredundant features in spatial and spectral dimensions of HSI and LiDAR data. To better acquire detailed information of spatial and spectral features at different scales, a multiscale spectral–spatial feature extraction module, including cascaded multiscale 3-D and 2-D convolutional layers, is inserted into MHST. Simultaneously, an adaptive global feature extraction module based on head selection pooling transformer is given after transformer encoder module for alleviating token redundancy in an adaptive computation style. Finally, we develop a multimodal–multiscale feature fusion classification module with local features and global class token, to exploit a powerful global–local fuse style. The extensive experiments on three popular datasets demonstrate that MHST significantly outperforms other related networks.

Electrocardiographic (ECG) signals have been successfully used in biometric recognition. However, the accuracy of ECG-based biometric systems is generally lower than systems based on other physiological traits. This study introduces a local feature learning method aimed at enhancing the performance of ECG-based biometric recognition systems. Specifically, we first extracted the multi-scale differential feature (MDF) for each point in the training ECG heartbeats using the difference between each point and its neighboring points. Second, we learn feature mapping to project these MDFs into low-dimensional descriptors in an unsupervised manner, where 1) the errors between the original MDF and reconstructed MDF are minimized. 2) The total variation in the reconstructed MDFs is minimized. Third, we represented each ECG heartbeat as a histogram feature using clustering and pooling descriptors. Finally, we adopted global feature learning methods to obtain a representation of an ECG heartbeat. Experiments on the MIT-BIH Arrhythmia, ECG-ID, and Physikalisch Technische Bundesanstalt databases verified the performance of the proposed method over existing ECG biometric recognition methods using within-session analysis. Moreover, we evaluated the performance of the proposed method using an across-session analysis of the ECG-ID database.

Accurate and efficient landslide identification is an important basis for landslide disaster prevention and control. Due to the diversity of landslide features, vegetation occlusion, and the complexity of the surrounding surface environment in remote sensing images, deep learning models (such as U-Net) for landslide detection based only on optical remote sensing images will lead to false and missed detection. The detection accuracy is not high, and it is difficult to satisfy the demand. Synthetic aperture radar (SAR) has penetrability, and SAR images are highly sensitive to changes in surface morphology and structure. In this study, a multi-input channel U-Net landslide detection method fusing SAR, optical, and topographic multisource remote sensing data is proposed. First, a multi-input channel U-Net model fusing SAR multisource remote sensing data is constructed, then an attention mechanism is introduced into the multi-input channel U-Net to adjust the spatial weights of the feature maps of the multisource data to emphasize the landslide-related features, and finally, the proposed model is applied to the experimental scene for validation. The experimental results demonstrate that the proposed model combined with SAR multisource remote sensing data improves the perception ability of landslide features, focuses on learning landslide-related features, improves the accuracy of landslide detection, and reduces the rate of false detections and missed detections. Compared with the traditional U-Net landslide detection method based on SAR multisource remote sensing data and the traditional U-Net method that disregards SAR multisource remote sensing data, the proposed method has the best quantitative evaluation indicators. Among them, the proposed method obtained the highest F1 value (66.18%), indicating that fused SAR remote sensing data can provide rich and complementary landslide feature information, simultaneously setting up a multichannel U-Net model to input multisource remote sensing data can effectively process landslide feature information. The proposed method can provide theoretical and technical support for landslide disaster prevention and control.

Clustering-based approach has proved effective in dealing with unsupervised domain adaptive person re-identification (ReID) tasks. However, existing works along this approach still suffer from noisy pseudo labels and the unreliable representation ability during the whole training process. In order to solve these problems, in this paper, we propose a new approach to learn the feature representation with better generalization ability through limiting noisy pseudo labels. At first, we propose a Sample Dropout (SD) method to prevent the training of the model from falling into the vicious circle caused by samples that are frequently assigned with noisy pseudo labels, our method can correct the noisy labels and boost the representation ability. In addition, we put forward a new method referred as to Feature Diversity Learning (FDL) under the classic mutual-teaching architecture, which can significantly improve the generalization ability of the feature representation on the target domain in an unsupervised fashion. Experimental results show that our proposed FDL-SD achieves the state-of-the-art performance on multiple well-known benchmark datasets. [ABSTRACT FROM AUTHOR]

Clustering-based unsupervised domain adaptive person re-identification methods have achieved remarkable progress. However, existing works are easy to fall into local minimum traps due to the optimization of two variables, feature representation and pseudo labels. Besides, the model can also be hurt by the inevitable false assignment of pseudo labels. In order to solve these problems, we propose the Doubly Stochastic Subdomain Mining (DSSM) to prevent the nonconvex optimization from falling into local minima in this paper. And we also design a novel reweighting algorithm based on the similarity correlation coefficient between samples which is referred to as Maximal Heterogeneous Similarity (MHS), it can reduce the adverse effect caused by noisy labels. Extensive experiments on two popular person re-identification datasets demonstrate that our method outperforms other state-of-the-art works. The source code is available at . [ABSTRACT FROM AUTHOR]

Point cloud analysis is an important part of 3D geometric processing. It has been widely used in many fields, such as automatic driving and robots. Although great progress has been made in recent years, there are still some unresolved problems. For example, current methods devote employing MLP to extract local features after search k neighbor points, they cannot effectively model the dependency relationship between the anchor point and k neighboring points. In addition, the prevailing models may not exploit the inherent structural similarities present in the global scope. To solve these issues, we propose a feature extraction model named DetailPoint to get detailed local information and long-range global dependency of point clouds. DetailPoint possess three units: the shallow local learning unit, the deep local learning unit and the deep global learning unit. We first use the SLL to extract shallow local features, and then use the DLL to learn deep local features. In these two units, we design a dual-path extraction method to acquire detail local features with dependencies. Finally, the DGL unit is employed to improve the generalization ability of local features and establish global interaction. These three units are connected in series to form our DetailPoint. We evaluated the performance of our model on four datasets, ScanObjectNN and ModelNet40 for shape classification, the ShapeNet dataset for part segmentation, and the S3DIS dataset for sementatic segmentations. The experimental results demonstrate that DetailPoint is capable of expressing point clouds more effectively, resulting in superior performance compared to existing methods. [ABSTRACT FROM AUTHOR]

Accurate localization in GPS-denied environments has always been a core issue in computer vision and robotics research. In indoor environments, vision-based localization methods are susceptible to changes in lighting conditions, viewing angles, and environmental factors, resulting in localization failures or limited generalization capabilities. In this paper, we propose the TransCNNLoc framework, which consists of an encoding–decoding network designed to learn more robust image features for camera pose estimation. In the image feature encoding stage, CNN and Swin Transformer are integrated to construct the image feature encoding module, enabling the network to fully extract global context and local features from images. In the decoding stage, multi-level image features are decoded through cross-layer connections while computing per-pixel feature weight maps. To enhance the framework's robustness to dynamic objects, a dynamic object recognition network is introduced to optimize the feature weights. Finally, a multi-level iterative optimization from coarse to fine levels is performed to recover six degrees of freedom camera pose. Experiments were conducted on the publicly available 7scenes dataset as well as a dataset collected under changing lighting conditions and dynamic scenes for accuracy validation and analysis. The experimental results demonstrate that the proposed TransCNNLoc framework exhibits superior adaptability to dynamic scenes and lighting changes. In the context of static environments within publicly available datasets, the localization technique introduced in this study attains a maximal precision of up to 5 centimeters, consistently achieving superior outcomes across a majority of the scenarios. Under the conditions of dynamic scenes and fluctuating illumination, this approach demonstrates an enhanced precision capability, reaching up to 3 centimeters. This represents a substantial refinement from the decimeter scale to a centimeter scale in precision, marking a significant advancement over the existing state-of-the-art (SOTA) algorithms. The open-source repository for the method proposed in this paper can be found at the following URL: github.com/Geelooo/TransCNNloc. [ABSTRACT FROM AUTHOR]

Visible-infrared Cross-modality Person Re-identification (VI-ReID) is a critical technology in smart public facilities such as cities, campuses and libraries. It aims to match pedestrians in visible light and infrared images for video surveillance, which poses a challenge in exploring cross-modal shared information accurately and efficiently. Therefore, multi-granularity feature learning methods have been applied in VI-ReID to extract potential multigranularity semantic information related to pedestrian body structure attributes.However, existing researchmainly uses traditional dual-stream fusion networks and overlooks the core of cross-modal learning networks, the fusion module. This paper introduces a novel network called the Augmented DeepMulti-Granularity Pose-Aware Feature FusionNetwork (ADMPFF-Net), incorporating theMulti-Granularity Pose-Aware Feature Fusion (MPFF)module to generate discriminative representations. MPFF efficiently explores and learns global and local features with multi-level semantic information by inserting disentangling and duplicating blocks into the fusion module of the backbone network. ADMPFF-Net also provides a new perspective for designing multi-granularity learning networks. By incorporating the multi-granularity feature disentanglement (mGFD) and posture information segmentation (pIS) strategies, it extracts more representative features concerning body structure information. The Local Information Enhancement (LIE) module augments high-performance features in VI-ReID, and the multi-granularity joint loss supervises model training for objective feature learning. Experimental results on two public datasets show that ADMPFF-Net efficiently constructs pedestrian feature representations and enhances the accuracy of VI-ReID. [ABSTRACT FROM AUTHOR]

This study aims to use the theory of iterative control algorithm to learn and extract the semantic structural features of large-scale network data by analyzing and exploring the semantic structural features of large-scale networks with the help of data visualization analysis methods. In addition, it conducts relevant research on the high-quality simplified representation of computer network graphs based on the effective extraction of structural features and combined with graph sampling techniques. An iterative learning control algorithm based on reference trajectory update is proposed for the tracking control problem of discrete linear control system output subject to non-repetitive perturbation. First, the controller is parametrically optimized by constructing a performance index function to track the system output quickly and accurately at the desired point of reference trajectory update. Second, when the system output is affected by a batch of non-repetitive perturbations, a new performance indicator function is further constructed by introducing a Lagrange multiplier algorithm to establish multi-objective performance indicators to optimize the robust iterative learning controller. Finally, the algorithm is applied to a computer network graph design system, and the simulation results verify the reasonableness and effectiveness of the algorithm. These results show that the suggested technique surpassed the existing approaches with accomplishments such as an average degree of 12.5%, a clustering coefficient of 65%, network efficiency of 91%, and higher modularity, community identification, and network embedding scores. [ABSTRACT FROM AUTHOR]

Speech emotion recognition (SER) is an active research area in affective computing. Recognizing emotions from speech signals helps to assess human behaviour, which has promising applications in the area of human-computer interaction. The performance of deep learning-based SER methods relies heavily on feature learning. In this paper, we propose SCAR-NET, an improved convolutional neural network, to extract emotional features from speech signals and implement classification. This work includes two main parts: First, we extract spectral, temporal, and spectral-temporal correlation features through three parallel paths; and then split-convolve-aggregate residual blocks are designed for multi-branch deep feature learning. The features are refined by global average pooling (GAP) and pass through a softmax classifier to generate predictions for different emotions. We also conduct a series of experiments to evaluate the robustness and effectiveness of SCAR-NET which can achieve 96.45%, 83.13%, and 89.93% accuracy on the speech emotion datasets EMO-DB, SAVEE, and RAVDESS. These results show the outperformance of SCAR-NET. [ABSTRACT FROM AUTHOR]

Traffic crashes cause serious traffic delay and have some unobserved heterogeneity in different areas. Using Texas accident data in 2020, this article aims to predict the level of delay caused by crashes (LDC) accurately and efficiently and discuss the difference between metropolitan and non-metropolitan areas. A framework based on Random Forests (RF) and LightGBM (LGBM) is developed to measure the association between LDC and its possible risk factors. At first, the most relevant variables in different areas were recognised through recursive feature elimination based on logistic regression. Then, LDC were forecasted by classifiers after grid search hyper parameters. To resolve data imbalance, two threshold moving methods of maximisation G-mean and F1-score were used. Finally, SHapley Additive explanation was employed to interpret the best model. The results indicate that the improved RF performs better in metropolitan areas and the improved LGBM performs better in non-metropolitan areas. In addition, Highway, spring and sunrise are the main risk factors of higher LDC in the two areas. And excessive wind speed and temperature in metropolitan areas can lead to higher LDC while in non-metropolitan areas it is pressure and apparent temperature. [ABSTRACT FROM AUTHOR]

Talking face generation is widely used in education, entertainment, shopping, and other social practices. Existing methods focus on matching the speaker's mouth shape with the speech content. Still, there is a lack of research on automatically extracting potential head motion features from speech, resulting in a lack of naturalness. This paper proposes SATFace, a subject agnostic talking face generation method with natural head movement. To model the talking face's complicated and critical features (identity, background, mouth shape, head posture, etc.), we construct SATFace by taking encoder-decoder as the primary network architecture. Then, we design a long short-time feature learning network to better reference the global and local information in audio for generating reasonable head movement. Besides, a modular training process is proposed to improve explicit and implicit features' learning effects and efficiency. The experimental comparison results show that SATFace improves by at least about 9.8% in cumulative probability of blur detection and 8.2% in synchronization confidence compared with the mainstream methods. The mean opinion scores show that SATFace has advantages in terms of lip sync quality, head movement naturalness, and video realness. [ABSTRACT FROM AUTHOR]

The time series and multi‐scale characteristics of complex industrial process data are always important factors affecting the performance of fault diagnosis. In this study, a new fault diagnosis model based on multi‐scale attention dilated causal residual convolution (ADCRC) is proposed. Aiming at the temporal nature of industrial data, the ADCRC module is developed to extract time series features, in which the ADCRC module is composed of dilated causal convolution (DCC), attention mechanism (AM), and residual connection, DCC is used to extract time series features, AM adjusts the weight of features according to attention distribution to obtain more important feature information, and residual connection is used to enhance the training accuracy of model. For the multi‐scale characteristics of original data, MS‐ADCRC model based on ADCRC module is developed for multi‐scale feature extraction, in which multiple ADCRC modules extract multi‐scale features of data in parallel. Finally, the proposed MS‐ADCRC model is tested on the Tennessee‐Eastman data set. Compared with other existing models, the results show that the proposed MS‐ADCRC model has more advantages in fault diagnosis feature learning. [ABSTRACT FROM AUTHOR]

Existing point cloud feature learning networks often learn high-semantic point features representing the global context by incorporating sampling, neighborhood grouping, neighborhood-wise feature learning, and feature aggregation. However, this process may result in a substantial loss of granular information due to the sampling operation and the widely-used max pooling feature aggregation, which neglects information from non-maximum point features. Consequently, the resulting high-semantic point features could be insufficient to represent the local context, hindering the network’s ability to distinguish fine shapes. To address this problem, we propose PointStack, a novel point cloud feature learning network that utilizes multi-resolution feature learning and learnable pooling (LP). PointStack aggregates point features of various resolutions across multiple layers to capture both high-semantic and high-resolution information. The LP function calculates the weighted sum of multi-resolution point features through an attention mechanism with learnable queries, enabling the extraction of all available information. As a result, PointStack can effectively represent both global and local contexts, allowing the network to comprehend both the global structure and local shape details. PointStack outperforms various existing feature learning networks for shape classification and part segmentation on the ScanObjectNN and ShapeNetPart datasets, achieving 87.2% overall accuracy and instance mIoU.

SAR image target recognition relies heavily on a large number of annotated samples, making it difficult to classify the unseen class targets. Due to the lack of effective category auxiliary information, the current zero-shot target recognition methods for SAR images are limited to inferring only one unseen class rather than classifying multiple unseen classes. To address this issue, a conditional generative network with the category features from the simulated images for zero-shot SAR target recognition is proposed in this paper. Firstly, the deep features are extracted from the simulated images and fused into the category features that characterize the entire class. Then, a conditional VAE-GAN network is constructed to generate the feature instances of the unseen classes. The high-level semantic information shared in the category features aids in generalizing the mapping learned from the seen classes to the unseen classes. Finally, the generated features of the unseen classes are used to train a classifier that can classify the real unseen images. The classification accuracies for the targets of the three unseen classes based on the proposed method can reach 99.80 ± 1.22% and 71.57 ± 2.28% with the SAMPLE dataset and the MSTAR dataset, respectively. The advantage and validity of the proposed architecture are indicated with a small number of the seen classes and a small amount of the training data. Furthermore, the proposed method can be extended to generalized zero-shot recognition.