The rapid development of global manufacturing has made distributed flexible workshop scheduling problems increasingly complex, involving multiple dimensions such as factory task allocation, internal machine scheduling, and job sequence sorting, requiring efficient collaboration. However, in the actual production environment, the arrival times of jobs are usually dynamic. To solve the distributed flexible job shop scheduling problem with dynamic arrival of jobs, a mixed-integer programming model is developed to minimize both the maximum completion time and total energy consumption. Furthermore, a dual objective optimization algorithm based on an improved dual deep Q-network is proposed. To enhance learning efficiency and model generalization, 9 normalized state representation functions are extracted. In addition, considering the characteristics of the problem, 9 composite scheduling rules and multi-objective adaptive reward functions are specifically designed. The results showed that the proposed algorithm can more effectively optimize the maximum completion time and total energy consumption targets, demonstrating significant advantages in solving the dynamic scheduling problem under study.
Accurate species identification of marine microalgae is crucial for water-quality monitoring and harmful algal bloom (HAB) early warning. However, existing microscopic image detection methods struggle with dense, small targets and low-contrast backgrounds. This study presents PHYLA-Det11, an enhanced model based on YOLOv11n, specifically optimized for microalgae detection. Our architecture integrates a four-scale detection head for hierarchical cell modeling, an Attention-guided Feature Fusion and Recalibration Module (AFFRM) to enhance feature extraction and suppress background noise, and a CARAFE upsampling operator to improve reconstruction in low-contrast regions. Benchmarked on the algae-gtnnf and VisAlgae 2023 datasets, PHYLA-Det11 achieved mAP@0.5 improvements of 3.4% and 3.0%, respectively, over the baseline. Notably, recall on the algae-gtnnf dataset increased from 84.8% to 90.7%. With its lightweight design and superior sensitivity to small taxa, PHYLA-Det11 is highly suitable for deployment on shipborne or buoy-based embedded monitoring platforms.
This paper introduces a novel image encryption algorithm that integrates the one-dimensional Logistic chaotic system, the three-dimensional Lorenz chaotic system, compressed sensing (CS), and a dynamic S-box. Initially, the Arnold transform is applied to scramble the pixel positions of the color image, thereby increasing its complexity. Subsequently, compressed sensing technology is utilized for the first-stage encryption, achieving simultaneous measurement and compression. A hybrid chaotic sequence, generated by combining the one-dimensional Logistic chaotic map and the three-dimensional Lorenz chaotic system, is then used to construct a dynamic S-box, which performs secondary encryption on the image data. This approach not only enhances encryption security but also significantly reduces storage and transmission requirements. Experimental results show that the proposed method achieves a high entropy value of 7.9991 bits per pixel, a peak signal-to-noise ratio (PSNR) of up to 40.743 dB, and a structural similarity index (SSIM) close to 1, demonstrating its effectiveness in preserving image quality while providing robust encryption. This work presents a promising solution for securing image data in various applications, including those in the military and medical industries.
This paper presents a comprehensive performance analysis of single, double, and triple material gate (SMGJL, DMGJL, TMGJL) Junctionless SOI FinFETs for Analog and RF applications using Technology Computer-Aided Design (TCAD) simulations at a 30 nm gate length. This study investigates key performance metrics, including Drain Induced Barrier Lowering (DIBL), Subthreshold Swing (SS), ON current (I-on), OFF current (I-off), I-on/I-off ratio, transconductance (g(m)), output conductance (g(d)), cut-off frequency (f(T)), and maximum oscillation frequency (f(max)) for three different gate structures. The junctionless structure, along with SOI technology, enables simplified fabrication, improved electrostatic control, and reduced leakage currents. The simulation results indicate that the TMGJL FinFET exhibits superior performance, achieving the lowest SS (59.99mV/dec), highest I-on/I-off ratio (6.11 & times;10), and enhanced RF figures of merit (f(T) = 121.98GHz, f(max) = 471.16GHz), making it highly suitable for next-generation low-power and high-frequency integrated circuits. Furthermore, the inverter circuit designed using the TMGJL FinFET exhibits high noise margin (N-MH) of 0.500 V and low noise margin (N-ML) of 0.287 V, from the voltage transfer characteristics, indicating robust noise immunity of the simulated TMGJL FinFET. The comparative analysis demonstrates the advantages of multi-material gate engineering, with the TMGJL structure providing the best trade-off among switching behaviour, current drive capability, and RF/Analog performance.
Asymmetric high-voltage pulse track circuits are widely applied in track sections with poor shunting conditions. To address the lack of mature mathematical or circuit models for pulse signals, this study develops a wideband simulation model of the asymmetric high-voltage pulse track circuit. This model allows for exhaustive simulation of various operational scenarios, thereby providing foundational support for simulation-based analysis of pulse track circuits. Firstly, the construct a simulation circuit for the pulse signal source. Next, theoretical calculations determine the impedance parameters of the frequency-dependent components like rails within the track circuit system. A wideband model is then established using vector fitting method combined with circuit synthesis theory. Subsequently, a wideband model for asymmetric high-voltage pulse track circuits is constructed based on the system architecture for simulation analysis. The model's accuracy and robustness are validated through field measurement data and rail surface voltage tests. Finally, transmission characteristics analysis is performed based on this simulation model, including the effects of ballast resistance and transmission distance on the receiving-end voltage and track surface voltage along the line.
This study presents an adaptive control strategy for achieving generalized synchronization in networks composed of n nodes with arbitrary topological configurations. Each node is modeled as a FitzHugh-Nagumo type dynamical system. Numerical simulations are provided to demonstrate and validate the effectiveness of the proposed theoretical results.
Extreme scale variation and severe occlusion in complex driving environments significantly degrade the detection performance of small traffic signs. To address these challenges, this paper presents EFFE-YOLO, an accurate detection model based on YOLOv11n. Specifically, the Cross-Stage Partial Network and Parallel Multi-Scale Feature Fusion Attention (CSP-PMSFA) module enhances deep multi-scale semantic feature extraction, and the Cross-scale Alignment Zone Neck (CAZ Neck) achieves seamless cross-scale feature alignment. The Adaptive Downsampling (ADown) algorithm reduces information loss and preserves fine-grained details critical for small targets. Furthermore, the integration of Scale Sequence Feature Fusion (ScalSeq) and the P2 detection layer effectively fuses multi-scale feature maps, retains high-resolution information for small targets, and strengthens the model's multi-scale detection ability. We propose an Inner-GIoU loss function equipped with auxiliary bounding boxes and a scaling factor mechanism to improve bounding box regression accuracy. Experimental results show that EFFE-YOLO achieves 77.99% mAP@50 and 60.24% mAP@50:95 on TT100K, with improvements of 1.82% mAP@50 and 3.07% mAP@50:95 on CCTSDB over the baseline model. Specifically, it yields a 24.98% mAP@50 improvement for small targets on TT100K, validating its superiority in small-scale traffic sign detection.
Depression is a prevalent psychological disorder with serious consequences, making its early identification crucial for timely intervention and effective mental health management. Speech signals are closely associated with emotional states and have therefore been widely explored for automated depression detection. Although Transformer-based models exhibit strong global modeling capabilities, they often struggle with long-duration speech, particularly when capturing pauses, prosody, and sparsely distributed depressive cues. As a result, these models can be distracted by redundant speech segments. To address these issues, we propose ASTAR-Former, a speech-based framework for depression detection. In the encoding stage, we integrate gated relative positional bias with multi-scale convolutional branches to better model dynamic speech rhythm. In the aggregation stage, we design an adaptive multi-instance learning scheme that combines Top-K selection, attention-based aggregation, and prototype guidance, allowing the model to focus on a small number of key segments containing depressive cues. Experimental results show that ASTAR-Former achieves a Macro-F1 score of 0.747 on the DAIC-WOZ dataset, outperforming representative baselines, and 0.721 on the CMDC dataset, demonstrating its robustness and generalization across different languages and datasets.
Cost control and risk management are pivotal factors determining the success or failure of a project. The final project cost is intrinsically linked to the efficiency of resource allocation and schedule planning, while Value at Risk (VaR) serves to quantify the maximum potential loss a project may incur over different time periods. Due to the influence of multiple factors such as technology, resources, and external environment, the actual completion time of subprojects often exhibits uncertainty, thereby escalating overall completion risk. Research indicates that shortening project duration may accelerate progress but could lead to concentrated resource investments and increased unit costs; conversely, extending the duration may alleviate resource pressure but tends to accumulate uncertainties, thereby elevating systemic risk and VaR. Achieving an optimal balance between minimizing completion costs and controlling VaR thus constitutes a central challenge in project management. To address this issue, this paper introduces an enhanced multi-objective optimization algorithm that integrates the Non-dominated Sorting Genetic Algorithm II (NSGA-II) with Q-learning reinforcement learning, incorporating two innovative initialization strategies: Blue Noise Sampling (BNS) and Front-Aware Inverse Mapping Seeding (FAIMS). Experimental results demonstrate that the proposed approach outperforms traditional algorithms such as NSGA-II in terms of convergence speed and solution diversity, offering substantial practical applicability.
Accurate management of maize diseases and pests plays a vital role in ensuring stable grain production. Conventional detection methods dependent on manual experience exhibit limited efficiency and unsatisfactory accuracy. To enhance the detection performance of maize diseases and pests, this study proposes an improved object detection model, named ASCC-Net, which is optimized based on the YOLOv9 framework. The new model innovatively integrates Spatial and Channel Reconstruction Convolution (SCConv) with AConv in the base model, forming ASCConv. By constructing a spatial context path and a channel reconstruction mechanism, the model significantly alleviates spatial and channel redundancy in image features, improving its robustness and generalization capability for detecting maize diseases and pest targets of various sizes and shapes. Meanwhile, the Convolutional Block Attention Module (CBAM) is introduced to further improve the model's attention to key target areas, reduce background noise interference, and strengthen semantic feature extraction. Experiments were conducted on four common maize disease types, including Leaf Spot, Corn Brown Spot, Corn Pest, and Corn Downy Mildew. The results indicate ASCC-Net achieves notable improvements compared with the baseline model, with detection precision improved by 1.9%, mAP@50 improved by 2.2%, and mAP@50-95 improved by 1.1%. These findings provide an efficient and deployable model reference for intelligent agricultural disease and pest detection.
Aiming at the problems of high false detection rate, high missed detection rate, low precision, and poor robustness in traffic target detection within computer vision tasks, a model called MFSF-YOLO11 (YOLO11 with Multi-directional information Flow and Scale-adaptive Fusion) is proposed. First, an improved feature extraction method is proposed. By designing the SPPCAKO module to replace the original SPPF module, the capability and accuracy of feature extraction are effectively enhanced. In terms of feature fusion, an innovative SDI-Damo Neck structure is put forward, which significantly improves the detection accuracy and robustness of YOLO series models in practical applications. To further boost the detection performance, a brand-new OASFFHead structure is introduced in the detection head part. This structure can simultaneously take into account the directionality and scale variation of targets, thereby optimizing the overall performance of target detection. Finally, the InnerMPDIoU loss function is adopted to replace the traditional CIoU loss function, which not only improves the accuracy of target localization but also remarkably optimizes the fitting effect of detection boxes. Experimental results show that the improved YOLO11 model achieves a 4.8% increase in mAP50, an 8.2% improvement in precision, and a 15.8% rise in recall on the TT100K dataset, demonstrating higher accuracy and robustness in complex environments.
In this article, we present several characterizations of the Minkowski inverse. It is observed that, unlike the Moore-Penrose inverse, the existence of the Minkowski inverse is not always guaranteed. We derive several necessary and sufficient conditions for the existence of the Minkowski inverse. In addition, a determinantal formula for the Minkowski inverse is obtained.
In fatigue driving detection, conventional methods based on video or behavioral features are easily disturbed by lighting, occlusion, and other environmental factors, making stable and reliable recognition difficult. In contrast, electroencephalogram (EEG) signals directly reflect the driver's neural activity and provide stronger anti-interference capability and higher temporal resolution, and thus have become an important data source for fatigue modeling. However, most existing EEG-based approaches still rely on features from a single domain and therefore lack the ability to jointly characterize rhythmic dynamics and structural rhythm changes. Meanwhile, mainstream public datasets contain relatively few samples and fail to adequately capture inter-individual variability. To address these limitations, this paper proposes a unified architecture, UniTF-ADA, which fuses multi-dimensional fatigue-related features via time-frequency dual-domain modeling. Considering the limited number of samples, we further design a pseudo-distribution generation strategy to expand the training variation space and improve model robustness. In addition, a dynamic weighted multi-objective loss function is introduced to adaptively balance different discrimination targets, stabilizing training and enhancing cross-subject generalization. Experimental results on two public datasets show that, compared with the Deformer baseline, UniTF-ADA improves ACC and Macro-F1 by 1.38% and 1.69% on Dataset I, and by 8.31% and 9.06% on Dataset II, respectively.
Waste tyre crumb aggregates (WTCAs) have adverse effects on the durability because they are not compatible with the other components of the concrete. Despite the fact that rubberized concrete is more porous and weaker because of the reduced bond with rubber particles, it also has such features as high ductility and impact resistance. Therefore, the purpose of the investigation was to improve the sulphuric acid resistance of rubberized concrete by improving the bond of rubber-cement paste and adding more silica-alumina to minimize the deterioration of concrete through the pozzolanic additives. Waste Glass Bottle Powders (WGBPs) and Ground Granulated Blast Furnace Slag (GBFS) are employed to substitute cement partially, whereas WTCAs are utilized as a fine or/and coarse substitute for natural aggregate. The durability performance was analyzed in terms of acid attack using residual strength, weight loss, virtual inspection, ultrasonic pulse velocity (UPV), and impact resistance tests, as well as X-ray Diffraction test (XRD) as microstructure tests. The utilization of the response surface methodology (RSM) enabled the optimization of strength loss, weight loss, UPV, and impact energy at both the initial and final crack stages, based on the binder mass percentage. Incorporating 30% WTCAs and 6% WGBPs improved durability, with higher residual strength after prolonged acid exposure compared to OPC concrete.
Domain generalization remains a fundamental challenge in medical image segmentation due to the domain shift. In this paper, we propose a novel segmentation framework, termed Memory-augmented Hierarchical Perception (MHP), which integrates strong prior knowledge from foundation models. Specifically, we design a frozen SAM-assisted encoder to extract high-quality local features using the SAM without additional fine-tuning. To further enhance domain-invariant representation learning, we introduce a hierarchical perception module that fuses complementary features from three pathways: local semantic cues, global contextual dependencies via Vision Mamba, and anatomical boundary sensitivity through edge-aware attention. Besides, a memory-augmented imagination generator is proposed to simulate unknown-domain features by the variational generation process with a learnable codebook, enabling the model to generalize beyond the training distribution. The encoder extracts multi-level features, which are integrated by the perception module and augmented by the imagination generator. These components form the core of our MHP framework. Extensive experiments on Fundus and Prostate segmentation datasets demonstrate that our MHP outperforms existing SOTA domain generalization methods.
In LCD-based vat photopolymerization 3D printing, excessive separation force between the cured layer and the release film can cause part warpage, film damage, and failure to print fine structures. Separation force is therefore a critical determinant of printing efficiency and success rate. In this study, the key factors governing separation force are systematically investigated through theoretical modeling and experimental validation. Based on solid-solid separation mechanics and viscous fluid dynamics, six primary parameters are first identified. An orthogonal experiment incorporating parameter interactions is then conducted. The results show that the interaction between separation speed and light intensity/exposure time is the dominant contributor to separation-force behavior. Using range analysis of the test matrix, the optimal parameter set is determined as follows: layer thickness 0.07 mm, light intensity 70%, exposure time 2.5 s, release-film thickness 0.2 mm, separation speed 0.5 mm/s, and leveling time 1.5 s. Under these conditions, the separation force decreases by 19.38%, and the successful print rate for complex fine structures increases from 65% to 80%. These results demonstrate that the proposed method improves print quality and process reliability.
With the continuous advancement of intelligent transportation systems, traffic object detection has become a key technology for addressing traditional traffic management challenges. However, in complex traffic environments, factors such as dense occlusion, significant scale variation, and the limited pixel proportion of small objects often lead to insufficient detection accuracy and increased false detection rates for small targets. To address these challenges, this paper proposes a multi-scale feature fusion-based object detection algorithm, termed DRSTD-YOLO11. First, a novel convolutional module, C3k2 LDConv, is constructed to enhance feature extraction capability and generalization performance under scale variation. Second, a multi-branch SENetV2 attention mechanism is introduced to strengthen inter-channel dependencies and global information aggregation, thereby improving small-object feature representation. Third, an Adaptive Spatial Feature Fusion Head (ASFFHead) is adopted to learn spatially adaptive fusion weights across different scale levels, effectively alleviating feature inconsistency among hierarchical representations. In addition, a P2 detection layer is incorporated to preserve fine-grained details of small objects, further improving small-object detection accuracy. Extensive experiments conducted on the KITTI benchmark demonstrate that DRSTD-YOLO11 achieves an mAP@0.5 of 90.5%, representing a 4.6% improvement over the baseline YOLO11, with a recall of 85.2%. Notably, the proposed method improves small-object AP by 13.9%, while maintaining stable performance on medium-and large-scale targets. Compared with YOLOv12 and YOLOv13, the proposed approach exhibits superior detection performance in small-object and occluded scenarios, validating its effectiveness for multi-scale object detection in complex traffic environments.
To address the challenges of low accuracy, missed detections, and false positives in unmanned aerial vehicle (UAV) object detection, particularly under complex scenes and dense small-object distributions, this paper proposes a lightweight small-object detection framework named WFD-YOLOv11, built upon YOLOv11. First, a Feature Fusion Attention Network (FFA-Net) is introduced to preprocess UAV imagery, effectively mitigating image distortion and preserving discriminative features. Second, a novel C3k2-iAFF module with an iterative attention mechanism is designed to enhance multi-scale spatial and channel feature representation. Third, a dynamic head is employed to improve small-object localization accuracy while reducing model parameters. Finally, the WIoUv3 loss function is adopted to dynamically adjust bounding-box regression weights, further enhancing detection robustness. Experimental results on the VisDrone2019 dataset demonstrate that, compared with YOLOv11n, WFD-YOLOv11n improves mAP@0.5 and mAP@0.5:0.95 by 4.5% and 3.0%, respectively, while reducing parameter count by 11.6%. The proposed method exhibits superior performance in detecting densely distributed small objects under complex backgrounds, meeting the lightweight requirements of UAV-based applications.
Accurate prediction of metro passenger flow is crucial for effective urban rail management. Origin-destination (OD) flow prediction offers a more comprehensive view of travel behavior, yet faces significant challenges, including matrix sparsity, complex spatio-temporal dependencies, and dynamic travel patterns. To address these challenges, this study proposes a spatio-temporal fusion OD traffic prediction model, termed the Single Station Multi-Route Interaction Graph Neural Controlled Differential Equation (SMI-GNCDE), guided by pattern differences. The model constructs a Single Station Multi-Route Interaction Graph (SMI-MRIG) and integrates it with Neural Controlled Differential Equations (NCDEs) to capture OD flow dynamics. It characterizes passenger flow differences from two perspectives: variations in travel patterns between weekdays and holidays, and intermediate station flow correlations based on station functional characteristics. Experiments on the two datasets demonstrate that SMI-GNCDE outperforms state-of-the-art baseline models in terms of Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Root Mean Square Error (RMSE), while maintaining more stable accuracy in long-horizon predictions. The metro passenger flow forecasting method proposed in this study holds promise for further enhancing the management and optimization of urban rail transit systems.
In this paper, the transfer matrix method is adopted to investigate the shear lag effect parameters of a single-span constant-section groove beam with different tensile and compressive moduli. By controlling the elastic modulus ratio, the distribution laws of the shear lag effect for simply supported and cantilever groove beams at different section positions under uniformly distributed loads and concentrated loads are systematically analyzed. The longitudinal shear lag effect only exists in the case of positive shear lag, and the shear lag coefficient reaches its maximum value at the mid-span section. As the concentrated load position approaches the fulcrum and the elastic modulus ratio increases, the shear lag effect becomes more pronounced, which should be given special consideration in design. The influences and general trends of the main structural parameters (such as the span-to-width ratio, width-to-height ratio, width-to-thickness ratio, and plate-to-thickness ratio) on the shear lag effect of the groove beam are investigated. Among various factors, the span-to-width ratio is the most important in influencing the shear lag effect of groove beam. The smaller the span-to-width ratio, the more severe the shear lag effect. As the width-to-height ratio increases, the shear lag coefficient of the section decreases.Meanwhile, as the width-to-thickness ratio increases, the shear lag effect will also decrease. However, as the plate-to-thickness ratio increases, the shear lag coefficient increases. This indicates that, with the span-to-width ratio remaining unchanged, an increase in the stiffness of the roof slab will lead to an increase in the shear lag effect. Through the study of the shear lag effect in groove beam, the influence laws of these structural parameters on shear lag were obtained, providing a valuable reference for the design and application of groove beam.
