To address the challenges of construction vehicle detection caused by complex illumination and severe occlusion in transportation infrastructure scenarios, as well as the requirements for emergency rescue under sudden hazards, this paper proposes a detection system based on a multi-model decision fusion architecture. The system adopts a “coarse-screening and fine-inspection” strategy, utilizing YOLOv8 for rapid global screening to ensure real-time performance. For regions with low confidence, an improved DPFL-RetinaNet is introduced to perform secondary refined detection. Concurrently, this paper proposes the Decoupled Polynomial Focal Loss based on Taylor series expansion, which enhances the mining capability for weak targets through gradient reshaping. Experimental results demonstrate that the proposed system achieves an AP50 of 87.1% on the TCVD dataset, representing an improvement of nearly 30 percentage points over the standalone YOLOv8. This provides an efficient visual perception solution for construction safety and emergency rescue in complex environments, effectively balancing detection precision and inference speed. The proposed Confidence-Guided Cascade Decision System (CG-CDS) effectively addresses the inherent trade-off between computational efficiency and detection accuracy. By dynamically routing challenging samples to a specialized fine-grained detector, the system achieves robust performance across diverse environmental conditions including low illumination, severe occlusion, and complex background clutter. Extensive ablation studies confirm that each component of the cascaded architecture contributes positively to the overall detection performance, with the proposed Decoupled Polynomial Focal Loss (DPFL) providing the most significant improvement for weak target mining.
The entity alignment task in Multi modal Knowledge Graph (MKG) faces challenges such as strong modal heterogeneity, large semantic gap, and inflexible fusion strategies. Existing methods generally rely on static weighted fusion and large amounts of annotated data, making it difficult to cope with complex multimodal environments. This article proposes a novel multimodal entity alignment model- TextFusionEA, which implements structural embedding based on GraphSAGE, uses MobileNetV3 to obtain visual features, introduces an adaptive fusion mechanism of semantic, structural, and visual features, and designs a semi supervised iterative learning strategy to extend the seed subset to alleviate the problem of data scarcity. Experiments on five datasets have shown that TextFusionEA performs well in MRR Hits@1. It significantly outperforms existing mainstream methods in terms of metrics, verifying its alignment performance and robustness in complex multimodal scenarios.
In this study, we aimed to develop a position-oriented evaluation model for football talent selection by applying the Intuitionistic Fuzzy Vlsekriterijumska Optimizacija Kompromisno Resenje method to analyze and rank ability indicators for three on-field positions: forwards, midfielders, and defenders. Based on an extensive literature review, 19 ability indicators were identified across five primary dimensions: technical, tactical, physical, psychological, and social. A total of 53 football coaches with practical experience and official coaching licenses participated in the evaluation. The results revealed significant positional differences in the prioritization of ability indicators. Forwards placed the greatest emphasis on shooting, midfielders on passing and ball reception, and defenders on one-on-one defensive ability. Furthermore, several competencies—namely autonomous decision-making, spatial awareness and positioning, tactical execution, and responsiveness to ball movement—emerged as core capabilities shared across all positions. These findings support the scientific and systematic development of football talent selection frameworks and provide practical guidance for coaches in designing training strategies and long-term player development plans.
With the development of digital learning, increasing students’ interest in learning and enhancing interactivity in the learning process have become central issues in educational research. By combining the hedonic-motivation system adoption model and the reward-penalty mechanism, this study analyzed learning behaviors in a gamified learning environment from the perspective of hedonic motivation. Correlation and regression analyses were employed to systematically explore the influence of multiple psychological factors, including perceived ease of use, perceived usefulness, curiosity, pleasure, and perceived control on the learning process, intention to use, and learning outcomes. The findings suggest that perceived ease of use and perceived usefulness significantly influence learner engagement and intention to use. Curiosity and a sense of pleasure promote continuous engagement and active learning. Perceived control enhances learners’ sense of control over what they are learning, focus and immersion, and learning effectiveness. Meanwhile, the reward-penalty mechanism exerts a positive moderating effect on the relationship between perceived ease of use and the learning process, thereby fostering learning motivations and promoting the continuity and depth of learning behaviors. The study demonstrates that the joint effects of psychological factors and the reward-penalty mechanism have significant implications for optimizing learning environments and enhancing learning outcomes.
The permanent magnet synchronous motor (PMSM) has been extensively applied in high-performance drive systems because of its high efficiency, compact construction, and superior dynamic performance. However, the design and implementation of traditional sliding mode control (SMC) systems have to be very dependent on the parameter selection, which could possibly lead to the increase of chattering and decreased adaptability under different operating conditions. In order to solve these problems, an SMC controller parameter optimization method based on reinforcement learning (RL) is put forward in this paper. The proposed RL-based SMC controller parameter optimization algorithm integrates RL into the sliding mode control process, and the adaptive adjustment of the control parameters can be realized through feedback from the performance of the controlled system. Specifically, the system states consist of the speed tracking error, the current error, and the sliding surface characteristic; the optimization goal contains the tracking accuracy, the convergence performance, and the chattering suppression. Simulation experiments on MATLAB/Simulink are performed to validate the effectiveness of the proposed method. The results show that, compared with the traditional SMC, the RL-based adaptive SMC control scheme can achieve good dynamic performance, less steady-state error, chattering reduction, and disturbance rejection.
