To address the challenges of large scale variations and strong background interference in dense crowd images, this paper proposes a crowd counting model named Multi-scale Selective Density Aggregation Network (MSDA-Net). The model introduces three collaborative modules: the Multi-scale Tuning Module (MTM), the Selective Attention Module (SAM), and the Density Context-aware Module (DCF), which collectively enhance feature representation and density map quality, thereby improving counting accuracy in complex congested scenes. Specifically, the MTM captures multi-scale contextual information through multi-branch atrous convolutions, alleviating the impact of drastic scale variations. The SAM employs spatial and channel attention mechanisms to suppress background noise and highlight foreground crowd regions. The DCF models the contextual dependencies of density maps using convolutional LSTM architecture, enhancing continuity between adjacent regions to generate smoother and more accurate density maps. A joint loss function combining Euclidean distance loss and structural similarity (SSIM) loss is adopted to optimize pixel-level regression and local structural consistency simultaneously. Extensive experiments are conducted on four public benchmarks, including ShanghaiTech Part_A, ShanghaiTech Part_B, UCF_CC_50, and UCF_QNRF. The proposed MSDA-Net achieves superior performance on all datasets, e.g., MAE of 58.1 on Part_A, 6.9 on Part_B, 205.3 on UCF_CC_50, and 84.8 on UCF_QNRF, outperforming many state-of-the-art methods. Ablation studies further validate the individual contribution of each module. Overall, MSDA-Net demonstrates strong generalization ability and practical application value in real-world crowded scenes. 针对密集人群图像中目标尺度变化大、背景干扰强等问题，本文提出一种基于多尺度选择性密度聚合网络（MSDA-Net）的人群计数模型。该模型引入了三个协同模块：多尺度调制模块（MTM）、选择性注意力模块（SAM）和密度上下文感知模块（DCF），三者共同增强特征表达能力和密度图质量，从而提高拥挤复杂场景下的计数精度。具体而言，MTM模块通过多分支空洞卷积捕获不同尺度的上下文信息，缓解目标尺度剧烈变化带来的影响；SAM模块利用空间与通道注意力机制抑制背景噪声，突出前景人群区域；DCF模块采用卷积长短时记忆网络对密度图的上下文依赖关系进行建模，增强相邻区域间的连续性，以生成更平滑、准确的密度图。实验中采用了联合损失函数，结合了欧氏距离损失与结构相似性（SSIM）损失，同时优化像素级回归和局部结构一致性。在四个公开数据集上进行了充分验证，包括ShanghaiTech Part_A、ShanghaiTech Part_B、UCF_CC_50和UCF_QNRF。所提出的MSDA-Net在所有数据集上均取得了优异性能，例如在Part_A上MAE为58.1，Part_B上为6.9，UCF_CC_50上为205.3，UCF_QNRF上为84.8，超过了多种最先进的方法。消融实验进一步验证了每个模块的单独贡献。总体而言，MSDA-Net在真实拥挤场景中表现出强大的泛化能力和实际应用价值。

Abstract Aero-engines generate a large amount of complex time-series signals during operation, involving multimodal information and specialized knowledge. Traditional fault diagnosis methods suffer from feature redundancy and insufficient interpretability in processing high-dimensional, multi-source data. To address these issues, Fixed-Option Question Answering (QA) transforms the fault diagnosis task into a candidate option selection task. This not only helps to accurately identify fault types but also facilitates understanding the model’s judgment criteria, improving the interpretability of the diagnostic process. To meet this challenge, this paper proposes a Fixed-Option QA model, which transforms the fault diagnosis task into a ‘question-fixed option-selection’ reasoning process. This explicitly characterizes the semantic comparison relationships between different fault types, enhancing the model’s discriminative power and interpretability. The model first employs a low-rank fusion method to jointly represent temporal signals and textual information, reducing interference from invalid information while preserving cross-modal complementary information and enhancing semantic expression. Subsequently, an option-aware mechanism is introduced to adaptively extract discriminative features among candidate options to improve model performance. Then, a knowledge graph is constructed to characterize the semantic relationships between signals, faults, and components, and structured information transmission and fusion are achieved in graph attention networks. Finally, an explicit reasoning process is introduced, allowing the model to demonstrate key reasoning steps while making decisions, making the diagnostic results more transparent and interpretable. We evaluate our method on the EngineMT-QA dataset, comparing it with Time-LLM, AutoTime, and ITFormer-0.5B (and its variants) using Accuracy and F 1 scores across multiple diagnostic tasks. Experimental results demonstrate that our model achieves state-of-the-art performance, and ablation studies validate the effectiveness of each module in multimodal feature fusion and structured reasoning.

The health condition of rolling bearings is crucial for the safe and stable operation of rotating machinery. However, bearing vibration signals acquired in industrial settings typically exhibit nonstationarity, strong nonlinearity, and extremely weak early fault characteristics, which significantly limit the accuracy and practicality of traditional diagnostic methods. To address these core challenges, this study proposes a novel end-to-end fault diagnosis model named LNFormer. It enhances the modeling of raw vibration signal sequences through an input transposition operation, eliminating the need for additional positional encoding. Furthermore, by removing the Softmax function from the self-attention mechanism, the model's computational complexity is substantially reduced, improving processing efficiency for long, high-sampling-rate vibration signals common in industry and supporting real-time or near-real-time monitoring. An innovative multi-head layer normalization (MHLN) structure is also designed as a nonlinear operator to effectively capture complex fault patterns while suppressing overfitting, thereby enhancing the model's robustness and generalization capability under noisy and varying operational conditions. Extensive experiments on eight public bearing datasets demonstrate that LNFormer achieves superior diagnostic accuracy and F1 score compared to existing advanced methods, along with excellent generalization and computational efficiency. A practical case study further confirms its significant potential for direct industrial application without complex data preprocessing.

Real-world multivariate time-series classification often faces structured domain shifts that preserve class semantics while systematically altering spectral characteristics, including frequency scaling, band shifting, and spectral coloring. This work aims to improve the robustness of multivariate time-series classification under such structured spectral shifts, rather than focusing only on standard in-domain evaluation. Existing approaches commonly seek robustness through generic multi-view learning or by treating domain variation as a set of discrete identifiers. As a result, they do not explicitly model the mechanism-driven spectral perturbations that frequently arise in practice. To address this limitation, we propose TriView-TA, a multi-domain and transformation-aware framework for robust multivariate time-series learning under spectral shift. TriView-TA constructs three complementary views in the time, frequency, and time-frequency domains to jointly capture temporal dynamics, global spectral structure, and localized spectral evolution. Building on this design, we introduce a transform-aware cross-frequency consistency objective defined over a parameterized shift/scale/color transformation family. This objective encourages the learned representation to remain stable under semantics-preserving spectral perturbations, without relying solely on discrete domain labels. TriView-TA optionally employs a Shared-QK attention mechanism to promote cross-view structural alignment while preserving view-specific expressivity. The proposed framework is evaluated on seven multivariate time-series benchmark datasets under a controlled, robustness-oriented protocol. Representative diagnostic analyses characterize how performance degrades as spectral-shift severity increases. Across the seven-dataset benchmark, TriView-TA achieves 93.44% clean accuracy and maintains strong worst-case robustness under safe spectral shifts. These results show that TriView-TA preserves strong discriminative performance on clean data while improving robustness to structured spectral mismatch. The findings indicate that explicit transformation-aware regularization is an effective strategy for robust mul tivariate time-series classification under deployment-relevant spectral shifts. For reproducibility, the source code, configuration files, Safe-A calibration scripts, perturbation-generation scripts, and main experiment reproduction scripts are publicly available at https://github.com/Oubrenn/ Triview-TA-Framework.

Industrial bearing diagnosis is challenged by two persistent obstacles: scarce fault exemplars and domain shift between laboratory and field measurements. We propose a simulation-augmented, physics-consistent diagnostic pipeline in which a statistically calibrated grey-box vibration synthesizer aligns synthetic data with real recordings to narrow the simulation-to-real gap. The calibration is performed by ECDF/KS-based distribution alignment on six task-agnostic descriptors, so that low-order amplitude statistics and fault-relevant indicators (e.g. envelope-line amplitude Aline and sideband SNR) match at the distribution level. Both synthetic and real segments are then processed by the same front end-spectral-kurtosis-guided resonance-band selection, Hilbert-envelope demodulation, and maximum correlated kurtosis deconvolution (SK -> envelope -> MCKD)-under a consistent frequency resolution, exposing cyclostationary defect signatures (BPFI/BPFO) and +/- kfr sidebands for fair cross-domain comparison. The resulting representations are fed into a lightweight residual 1D-CNN-BiLSTM-multi-head attention classifier. Across ten cross-condition transfer tasks spanning CWRU, SEU, and PU benchmarks, the proposed pipeline achieves an average target-domain accuracy of 97.31% (min 96.25%, max 98.05%), outperforming a strong deep baseline (DWCResNet) by 5.22 percentage points on average. Overall, simulation-augmented distribution alignment combined with a unified physics-guided front end provides a reproducible and deployment-oriented recipe for bearing-fault diagnosis when real-world labels are scarce.

The core of bearing fault diagnosis lies in the accurate feature extraction from non-stationary vibration signals; however, unavoidable multi-source engineering noise often leads to severe feature distortion. To address this issue, this paper proposes a two-stage robust feature extraction framework based on adaptive time-varying filtering fast ensemble empirical mode decomposition (ATVF-FEEMD) and hierarchical time-shift composite multi-scale fuzzy entropy (HTSCMFE). First, by introducing an adaptive filter and an 'information-energy' adaptive dual-criteria screening strategy (Log-maximum information coefficient and Log-Gini index), ATVF-FEEMD effectively eliminates noise components, achieving high-fidelity signal reconstruction under intense background noise. Second, through the hierarchical decomposition, multi-offset time-shifting, and composite multi-scale coarse-graining of the reconstructed signal via HTSCMFE, the derived feature vectors exhibit outstanding discriminative capability. Experiments on simulated and real-world datasets demonstrate that, owing to the exceptionally high quality of the extracted features, the proposed framework maintains excellent recognition accuracy and generalization ability across varying signal-to-noise ratios, even when utilizing only fundamental classifiers such as artificial neural networks and transformers. Furthermore, this study draws a crucial conclusion: high-quality front-end features can significantly reduce the algorithmic dependency of diagnostic models on complex graph topologies (e.g., graph convolutional networks/graph attention network), thereby providing a highly robust new paradigm for bearing fault diagnosis under severe noise conditions.

For the problems of limited feature representation, unstable parameter learning, and insufficient reliability of classification decisions in online handwritten signature verification under small-sample conditions, a verification method integrating lightweight representation, multi-scale feature modeling, and probabilistic discrimination is proposed. Aiming at noise interference, feature redundancy, and insufficient utilization of dynamic information in online signature sequences, the method first performs smoothing on the original signature sequences, and then combines feature importance evaluation with Principal Component Analysis to conduct feature selection and dimensionality reduction, thereby constructing an input representation that contains both global statistical attributes and local dynamic variation information. In the feature extraction stage, Ghost feature mapping is adopted for the initial representation of input information, and the InceptionNext-TF module together with the DASE module is used for multi-scale deep feature modeling to characterize variation patterns of signature samples at different scales. In the classification stage, a variational Bayesian fully connected layer is introduced to model the output weights in a distributional manner, and a Bayesian optimization-based adaptive parameter search strategy is further employed to adjust the relevant key hyperparameters. Experimental results on the public MCYT-100 and SVC-2004 Task2 datasets show that, under the 10-shot setting, the Equal Error Rates are 1.46% and 3.05%, respectively. 针对在线手写签名认证在小样本条件下存在特征表达受限、参数学习不稳定以及分类决策可靠性不足等问题，提出了一种融合轻量化表征、多尺度特征建模与概率判别机制的认证方法。该方法面向在线签名序列中噪声干扰、特征冗余以及动态信息利用不足等情况，首先对原始签名序列进行平滑处理，并结合特征重要性评估与主成分分析完成特征筛选与降维，构建同时包含全局统计属性和局部动态变化信息的输入表征。在特征提取阶段，采用Ghost特征映射对输入信息进行初始表征，并结合InceptionNext-TF模块与DASE模块开展多尺度深层特征建模，以刻画签名样本在不同尺度下的变化特征。在分类阶段，引入变分贝叶斯全连接层对输出权重进行分布化建模，并结合基于贝叶斯优化的自适应参数搜索策略，对相关超参数进行调整。基于MCYT-100和SVC-2004 Task2公开数据集的实验结果表明，在10-shot条件下，等错误率分别为1.46%和3.05%。

Sustainable scheduling is getting more and more attention with economic globalization and sustainable manufacturing. However, fewer studies on the batch scheduling problem consider energy consumption. This paper conducts an investigation into the multi-objective hybrid flow shop batch-scheduling problem with the objectives of minimizing both the makespan and electrical energy consumption. The study aims to select the optimal scheduling solution for the problem by considering batch splitting for all products. In this paper, we propose an improved black widow optimization (IBWO) algorithm to study the problem, which incorporates procreation, cannibalism, and mutation behaviors to maintain the population’s diversity and stability. To achieve our objectives, we use the dynamic entropy weight topsis method to select individual spiders. Finally, we use the nature theorem construction method, which relies on the property theorem, to solve the Pareto solution set and derive the optimization scheme for the hybrid flow shop batch scheduling problem. We verify the effectiveness of the proposed IBWO on instances of varying sizes. When we keep all other factors and cases constant, we compare the IBWO to the NSGA2 algorithm and find that it converges faster for both goals and has lower goals than the NSGA2.

In the field of rotating machinery fault diagnosis, domain adaptation (DA) methods have emerged as one of the most prevalent approaches for handling cross-scene diagnosis challenges. However, traditional fault diagnosis methods based on DA rely on a large number of unlabeled target-domain samples. In real-world applications, rotating machinery usually produces only healthy signals. The lack of fault samples significantly restricts the application of these methods. To tackle this issue, this article proposes a novel cycle-consistent generative adversarial network (Cycle-GAN) for data augmentation, named FDCycle-GAN. Specifically, it utilizes all source-domain signals and solely healthy target-domain signals as inputs to generate highly realistic target-domain fault signals. These generated signals are then employed to train the target fault diagnosis network. Moreover, FDCycle-GAN incorporates a series of feature enhancement strategies and introduces semantic consistency and time-frequency consistency constraints. When compared with state-of-the-art DA methods across three datasets, FDCycle-GAN demonstrates its superiority by achieving higher fault diagnosis accuracy in tasks where fault samples are scarce. This research provides a more practical and efficient solution for rotating machinery fault diagnosis. It holds great significance for safeguarding the stable operation of industrial equipment and minimizing economic losses resulting from equipment malfunctions.

Online Signature Verification (OSV), as a personal identification technology, is widely used in various industries. However, it faces challenges, such as incomplete feature extraction, low accuracy, and computational heaviness. To address these issues, we propose a novel approach for online signature verification, using a one-dimensional GhostACmix Residual Network (1D-ACGRNet), which is a Ghost-ACmix Residual Network that combines convolution with a self-attention mechanism and performs improvement by using Ghost method. The Ghost-ACmix Residual structure is introduced to leverage both self-attention and convolution mechanisms for capturing global feature information and extracting local information, effectively complementing whole and local signature features and mitigating the problem of insufficient feature extraction. Then, the Ghost-based Convolution and Self-Attention (ACG) block is proposed to simplify the common parts between convolution and self-attention using the Ghost module and employ feature transformation to obtain intermediate features, thus reducing computational costs. Additionally, feature selection is performed using the random forest method, and the data is dimensionally reduced using Principal Component Analysis (PCA). Finally, tests are implemented on the MCYT-100 datasets and the SVC-2004 Task2 datasets, and the equal error rates (EERs) for small-sample training using five genuine and forged signatures are 3.07% and 4.17%, respectively. The EERs for training with ten genuine and forged signatures are 0.91% and 2.12% on the respective datasets. The experimental results illustrate that the proposed approach effectively enhances the accuracy of online signature verification.

Aiming at the interference problem of outliers in point set matching, we propose an optimized variational Bayesian point set matching algorithm using coordinate ascent and simulated annealing. For the complex terms in the objective function, we introduce the coordinate ascending strategy and use the Monte Carlo integral stochastic approximation and control variable method to directly optimize the parameters. To enhance the generalization ability of the algorithm and avoid falling into the local optimal solution, we use the simulated annealing scheme to optimize the spatial mapping of the point set. Finally, the affine transform point set matching experiment and the non-rigid point set matching experiment represent our algorithm have a good matching effect and high accuracy.

Online signature verification (OSV) is widely used in finance, law and other fields, and is one of the important research projects on biological characteristics. However, its data set has a small scale and has high requirements for generalization of certification models. Therefore, how to overcome these problems is of great value to improve the practicality and security of online handwriting signature technology. We propose a writer-independent online handwritten signature verification method, which adopts the relative position matrix method to convert the traditional temporal features into images for processing. This method enriched the features of the signatures, serving the purpose of data augmentation. Then two-dimensional multi-scale feature fusion based Siamese neural network (2D-MFFnet) is built for representing and learning the importance of each channel adaptively combined with the attention mechanism. Finally, a temporal convolutional network is designed to construct the classifier. The results illustrate that compared with traditional time series models, the algorithm has reduced the equal error rate by at least 2.52% on the open datasets MCYT-100 and SVC2004 task2.

Due to the challenges posed by large-scale variability in crowd images and overlapping and occlusion of people in high-density regions, traditional CNNs with fixed-size convolution kernels or transformers lacking 2D locality and channel adaptation need to struggle to cope with this challenge. While Transformers have a global receptive field for long sequence tasks, CNNs exhibit better generalization and 2D locality. In order to combine the advantages of both approaches, this paper proposes a dual-branch multi-scale attention network (DBM-SA-Net). First of all, we propose a multi-scale channel attention convolution module to extract features at different scales while enhancing channel adaptation. Furtherly, local features are augmented using a feed-forward neural network that is more suitable for visual tasks. Then an efficient lightweight multi-scale regression head is employed to predict density maps. Finally, progressive cross-head supervision is introduced as a loss function to dynamically supervise instance labels noise and mitigate its effect. Extensive experiments are conducted on three crowd counting datasets (ShanghaiTech Part A, ShanghaiTech Part B, UCF-QNRF) to validate the effectiveness of the proposed method and the results show that DBMSA-Net outperforms state-of-the-art methods.

As for the uncertainty problem of detection and estimation of robot localization and mapping using ultrasonic sensor, this paper proposes the improvement of Strong Tracking UKF-SLAM approach using three-position ultrasonic detection. A three-position ultrasonic detection model is first of all built for reducing these uncertainties through topological relationship screening and environmental contour estimation. Then Strong Tracking UKF-SLAM approach is improved by using multiple fade factors to fuse the ultrasonic measurement data and motion model information of robot for obtaining more accurate localization and mapping. Finally, we construct simulation and indoor experimental environments and design the mobile robot system with ultrasonic sensor for verification. The simulation represents that the improved algorithm has less error and more accurate effect than original algorithms in localization and mapping of mobile robot. The indoor environmental experiment is performed for illustrating the feasibility and effectiveness of the proposed method. The proposed method has certain reference value for research of Simultaneous Localization and Mapping.

In order to solve the problem of multi-scale in a single image gathering crowd counting, a new crowd counting network based on the fusion of dilating convolution pyramid and context attention mechanism (DCPCANet) is proposed. With the first ten convolutional layers of VGG16 as the front-end network, an dilated convolutional pyramid fusion attention mechanism module (AMP) is proposed, which is introduced into the three-level upsampling feature fusion module to extract fused multi-scale features, and the AMP module stack is used as the back-end network to capture and fuse multiscale features, The context attention module (CAM) is used to generate the feature map with weight, and high-quality crowd density map is output at the same time. Three mainstream public data sets are adopted, ShanghaiTech PartA,ShanghaiTech PartB,UCF_CC_50. Compared with the previous algorithm, the MAE of the UCF_CC_50 dataset is reduced by 11%, which preliminarily verifies the accuracy and robustness of the model.

为了提高在线手写签名认证的准确率,设计了一种挖掘签名稳定笔段用于认证的方法.本文提出对签名笔段计算累计差异值矩阵进行匹配;其次采用动态时间规划(DTW)算法计算笔段稳定度;在此基础上,通过以笔段的特征输入双向长短期记忆网络(BLSTM)加注意机制进行处理,从而得到每个用户的稳定签名段集合;最后提取该集合的特征进行分类.该方法在svc2004 数据库上进行验证并得到了97.08%的认证率,并在40 个用户上取得了1.16%的等误率.该结果表明本文方法能够提高认证精度,并且验证了BLSTM与稳定笔段结合方法的有效性.

Ship detection has always been an important and challenging task. Small ship targets and complex backgrounds in optical remote sensing images can both lead to false alarms and missed alarms in detection. The YOLO series algorithms have been widely used in optical remote sensing ship target detection, which has the advantage of fast detection speed. However, the YOLO series algorithms have poor detection performance when facing small targets. Therefore, we propose a new algorithm, SMMA-YOLOv5, which can improve the model detection accuracy without significantly increasing the model size. We first introduce a self-attention mechanism to replace some of the convolutional layers to capture the global information of the feature map. Second, we integrate an efficient channel attention (ECA) model to the self-attention mechanism to enable information interaction between channels without adding additional computational overhead. Furthermore, we propose a new similarity mask structure to filter out the invalid regions in the feature map based on the elements’ similarity. The experiments on the public MASATI ship dataset indicate that SMMA-YOLOv5 improves Precision by 3.9%, Recall by 5.3%, and AP by 5.4%, and prove the effectiveness of the algorithm while maintaining real-time detection.

How to adaptively learn important signature features and use a lightweight model to achieve high-precision signature verification is still a challenge in the field of online signature verification. In response to this challenge, we proposed an attention mechanism depth-wise separable convolution residual network(A-DWSRNet) for online signature verification. First of all, the weight of signature features is adaptively learned through the convolutional attention module to improve the representation learning ability of the network. Next, the depth-wise separable convolution and depth convolution modules are introduced to improve the standard residual structure to construct a depth-wise separable residual unit, which reduces the overall parameter amount of the model and alleviates the loss of feature information of the multi-step residual structure. Finally, the proposed method has achieved 2.88% equal error rate (EER) and 4.17% EER in MCYT100 database and SVC2004-task2 database, respectively. The results show that the proposed method can effectively improve the accuracy of the Online signature verification system.