哈尔滨剑桥学院

Using some new techniques, criteria for strongly exposed points of Orlicz sequence spaces generated by arbitrary Orlicz function and equipped with the p-Amemiya (1 < p < infinity) norm are obtained. Next, strongly exposed property of these spaces is deduced. The obtained results broaden the knowledge about these notions in Orlicz sequence spaces equipped with the p-Amemiya norm.

This study examines how maturity-mismatched investments influence firms’ morality performance and the role of social credit system reform policy and social trust. Using the sample of Chinese A-share listed firms, we find that firms’ maturity-mismatched investments could impede their morality performance, which is consistent with the passive mismatch view. Specifically, maturity-mismatched investments hinder firms’ positive morality performance, instead promoting their negative morality performance. Social credit system reform policy and social trust weaken the negative relationship between firms’ maturity-mismatched behavior and morality performance. Meanwhile, social credit system reform policy and social trust also have a synergistic moderating role in alleviating the negative influence of firms’ maturity-mismatched investments on morality performance. In addition, the potential mechanism analysis suggests that maturity-mismatched investment may influence firms’ morality performance through inefficient investment and financing problems. Finally, heterogeneity analysis shows that the adverse influence of firms’ maturity-mismatched investment on their morality performance is more pronounced in firms with higher managerial power and lower collaborative culture.

Image degradation due to haze poses significant challenges to computer vision systems. Traditional methods based on convolutional neural networks (CNNs) are limited by their local receptive fields and struggle to model global haze distribution effectively. Furthermore, haze non-uniformly affects low-frequency energy and suppresses high-frequency details, a characteristic often underutilized by existing methods. To address these limitations, we propose CFCNet, a cross-spatial frequency collaborative network that jointly optimizes spatial and frequency representations. CFCNet introduces a cross-spatial aggregation mechanism for long-range dependency establishment and an adaptive frequency decomposition paradigm for fine-grained enhancement. Experimental results on synthetic (ITS, OTS, Haze4K) and real-world datasets (Dense-Haze, NH-Haze) demonstrate superior performance, with CFCNet achieving a PSNR of 42.92 dB on SOTS-Indoor and 35.40 dB on Haze4K, using only 7.15M parameters. This approach effectively restores degraded information, showcasing its potential for real-world applications.

Mass personalization manufacturing demands intelligent robotic systems with real-time decision-making and dynamic adaptability, yet deep learning (DL) deployment faces critical challenges: terminal devices lack computational resources while centralized cloud processing incurs prohibitive latency. This paper proposes a Cloud-Edge-Terminal collaborative framework that strategically distributes DL intelligence across hierarchical tiers. We present vertical distributed DL with early-exit mechanisms and partitioned DL strategies, integrated with incremental learning and model compression techniques. Experimental validation demonstrates 84% response time reduction, 88% communication overhead decrease, and 16-fold faster model updates compared to baseline approaches. A real-world smartphone assembly case study confirms practical viability, achieving 31% cycle time reduction and 72% downtime decrease. This framework establishes a foundational architecture for intelligent manufacturing systems capable of efficiently producing highly personalized products at scale.

Centimeter-level precise localization stands as a fundamental prerequisite for the safe and reliable operation of autonomous vehicles, especially in intricate urban intersection scenarios where conventional localization approaches often struggle with performance degradation. Existing methods face significant challenges at intersections due to environmental shielding and dynamic object occlusion. To address this critical issue, we propose a marker assisted vehicle-road cooperative localization methodology. This paper focuses on the core methodology design and the key technology of vehicle top marker detection. The proposed methodology satisfies feasibility, reliability, explainability, social, and economic requirements by leveraging special markers installed on bus tops and existing roadside cameras at intelligent intersections. A implementation of marker detection, including adaptive thresholding, edge processing, cycle extraction, quadrilateral verification, and marker decoding, is presented. Simulation results demonstrate that the detection method can effectively identify vehicle top markers even under severe occlusion by large surrounding vehicles. The methodology and detection technology provide a robust foundation for solving intersection localization problems in autonomous bus operations.