National University of Defense Technology

Logistics, as a tertiary industry, has developed rapidly and become an important part of the national economy. However, owing to the behindhand logistics pattern, the logistics vehicles drive empty-loaded on their return trip, resulting in wastage of half of the delivery resources. This paper proposes a logistics–client matching model under fourth-party logistics (4PL) to reduce the empty-loaded rate. First, a preference calculation model between logistics providers and clients was constructed. Next, for clients with small quantities of goods, a linear logistics−client one-to-one stable matching model was constructed based on the stable marriage matching model. Then, for clients with large quantities of goods, a linear logistics−client many-to-one stable matching model was constructed with a novel linear many-to-one stable matching constraint. Finally, the case study indicated that the linear model supports large-scale and global optimization. The real case study verified that the proposed stable matching model is effective in reducing the empty-loaded rate compared to the ordinary matching model, and the matching solution was fairer.

Hyperspectral images (HSIs) are gradually playing an important role in many fields because of their ability to obtain spectral information. However, sensor response differences and other reasons may lead to the generation of stripe noise in HSIs, which will greatly degrade the image quality. To solve the problem of HSIs destriping, a new iterative method via spectral-spatial factorization is proposed. We first rearrange the HSI data to get a new two-dimensional matrix. Then the original noise-free HSI is decomposed into a spectral information matrix and a spatial information matrix. The sparsity of stripe noise, the group sparsity of spatial information matrix, the smoothness of spectral information matrix can be used to achieve sufficient removal of stripe noise while effectively retaining spectral information and spatial details of the original HSI. Numerical tests on simulated datasets show that our method achieves an average PSNR growth above 4dB and a better SSIM result. The proposed method also obtains good results when processing real datasets polluted by Gaussian noise and stripe noise.

Large-scale models have gained significant attention in a wide range of fields, such as computer vision and natural language processing, due to their effectiveness across various applications. However, a notable hurdle in training these large-scale models is the limited memory capacity of graphics processing units (GPUs). In this paper, we present a comprehensive survey focused on training large-scale models with limited GPU memory. The exploration commences by scrutinizing the factors that contribute to the consumption of GPU memory during the training process, namely model parameters, model states, and model activations. Following this analysis, we present an in-depth overview of the relevant research work that addresses these aspects individually. Finally, the paper concludes by presenting an outlook on the future of memory optimization in training large-scale language models, emphasizing the necessity for continued research and innovation in this area. This survey serves as a valuable resource for researchers and practitioners keen on comprehending the challenges and advancements in training large-scale language models with limited GPU memory.

Objective Diabetic peripheral neuropathy (DPN) is a common complication of diabetes, posing a significant risk for foot ulcers and amputation. Corneal confocal microscopy (CCM) is a rapid, noninvasive method to assess DPN by analysing corneal nerve fibre morphology. However, selecting high-quality representative images remains a critical challenge. Methods In this study, we propose a fully automated CCM image-selection algorithm based on deep learning feature extraction using ResNet-18 and unsupervised clustering. The algorithm consistently identifies representative images by balancing non-redundancy and representativeness, ensuring objectivity and reproducibility. Results When validated against manual selection by researchers with varying expertise levels, the algorithm demonstrated superior performance in distinguishing DPN and reduced inter-observer variability. It completed the analysis of hundreds of images within 1 s, significantly enhancing diagnostic efficiency. Compared with traditional manual selection, the proposed method achieved higher diagnostic accuracy for key morphological parameters, including corneal nerve fibre density, length, and branch density. Conclusion The algorithm is open source and compatible with standard CCM workflows, offering researchers and clinicians a robust and efficient tool for DPN diagnosis. Further, multicentre studies are needed to validate these findings in diverse populations.

The dual-fiber optical tweezers have become widespread in trapping, assembling, and sensing due to their simple fabrication process and flexible operation. However, the miniaturization and integration of their displacement measurement optical paths remain challenging. Here, we propose and experimentally demonstrate an integration of structured-light displacement (SLD) measurement method tailored for dual-fiber optical tweezers. A key component split-waveplate is integrated onto the fiber end via coating and etching in the SLD method. The etched fiber and another single mode fiber form optical tweezers, which enables to trap particle and measure its position simultaneously without additional optics. More importantly, it demonstrates a superior signal-to-noise ratio after filtering out the trapping field by the etched fiber. Our results demonstrate a displacement sensitivity reaching the 0.1 pm/Hz1/2 level, which surpasses the performance of most results using the quadrant photodiode method. Ultimately, we discussed the possibilities of using two etched fibers to detect displacements in different directions, or integrating this method into a single optical fiber. This method has significant potential applications in precision sensing, contributes to the integration of optical tweezers and fosters the development of lab-on-fiber applications.

When discussing inter-satellite links, a unified space–time reference needs to be defined in advance. This section will briefly introduce the background knowledge of the space–time coordinate system where artificial earth satellites are located.

Position information is a fundamental aspect of spacecraft, playing a crucial role in user orbit maintenance, formation flight, configuration maintenance, and business operations. User positions can be predicted from orbital parameters or directly measured and calculated; similarly, continuous positioning can also obtain user orbits. Orbit determination methods can be divided into ground-based, space-based, and astronomical orbit determination based on the location of the ranging signal source. Ground-based orbit determination mainly includes space telemetry and control (such as USB, unified S-band telemetry and control), Satellite Laser Ranging (SLR), Doppler Orbitography and Radiopositioning Integrated by Satellite (DORIS), and Very Long Baseline Interferometry (VLBI); space-based orbit determination mainly includes Global Navigation Satellite Systems (GNSS) and Tracking and Data Relay Satellite Systems (TDRSS); astronomical orbit determination mainly includes satellite sensors and pulsars. Although there are many orbit determination methods, each has its advantages and disadvantages and cannot effectively solve the positioning problem of spacecraft, especially for medium and high orbit spacecraft. Medium and high orbit spacecraft have high orbital altitudes, and the ground observation angle is limited, with orbit determination accuracy at the hundred-meter level: for example, for a navigation satellite with an orbital altitude of 22,000 km, the angle from the satellite to the tangent of the Earth's edge is less than 26°; the tangent angle of the geosynchronous orbit satellite is even less than 18°. In actual station layout, due to political factors and terrain restrictions, it is impossible to set up stations globally and achieve full-arc coverage. Space-based positioning, which raises the reference position and is not restricted by region, has become the ideal positioning method for medium and high orbit spacecraft. The space-based navigation system represented by the GPS system is widely used in low-orbit satellite orbit determination, but its main lobe only covers 4.5% of GEO/HEO orbit users, and even with the addition of navigation signal sidelobes, the quadruple coverage is less than 30%. The Tracking and Data Relay System is mainly used for orbit determination of medium and low orbit, with a theoretical accuracy of up to 20 m, but it cannot be used for GEO/HEO satellite orbit determination.

The spacetime metric in the framework of general relativity includes two types of time, namely proper time and coordinate time. The proper time can be realized by a high-precision atomic clock according to the definition of the second, and is carried by the observer for the time measurement of the local reference system. The local time maintained by the navigation satellite clock belongs to the category of proper time. The coordinate time is determined by the spacetime metric of the global coordinate system. The International Astronomical Union (IAU) officially recommended the use of general relativity as the theoretical basis for the spacetime reference system in 1991, introduced the Geocentric Coordinate Time (TCG), and defined the coordinate time in terms of SI second (Li in Relativistic time comparison theory and high-precision time synchronization technology. PLA Information Engineering University, Zhengzhou, 2004 [1]).

The inter-satellite link system includes a data plane and a management plane. The data plane uses a layered structure to provide basic inter-satellite ranging, time synchronization, and packet data transmission services, and supports future extended application services. The network protocol of the layered structure of the navigation constellation inter-satellite link refers to the ISO/OSI reference model, and has been simplified according to the network characteristics of the navigation constellation. The network protocol stack is divided into five parts: physical layer, link layer, network layer, transport layer, and application layer.

The inter-satellite link is realized through a radio frequency transmission and reception device, using the corresponding frequency and communication system, to achieve relative measurement and data exchange between constellation satellites. The relative measurement process within the constellation is composed of a series of relative distance measurements and communication processes between two satellites, mainly to complete the precise measurement of inter-satellite pseudorange and clock time difference. High-precision inter-satellite distance and time difference measurement is the basis for navigation constellation orbit determination and time synchronization, and inter-satellite bi-directional measurement is currently the most commonly used and highest precision method (Jing Yi et al. in Rev Sci Inst 89(6):06,450 (2018)).

The key elements of the inter-satellite link involve measurement systems and topology analysis. The constellation configuration of space satellites is complex, and the link topology structure under different measurement systems is different. Inter-satellite observation is not conducted between any two satellites, but needs to be designed according to the measurement system and inter-satellite visibility analysis. The observation data of inter-satellite and the observation data between satellites and ground stations are the basic observation quantities for realizing satellite autonomous orbit determination and system time synchronization, which have important significance for the verification of satellite navigation system functions, performance and satellite orbit determination, system time synchronization methods.

The time slot planning of inter-satellite links directly determines the performance of the satellite network. Under the needs of inter-satellite ranging and communication, how to allocate inter-satellite links is the key to solving the problem. In the STDMA access mode, the division of inter-satellite network resources is not only the allocation of time slots, but also the allocation of satellite nodes. From the analysis of inter-satellite visibility, it can be known that each MEO satellite, excluding the currently invisible MEO satellites, plus the currently visible GEO/IGSO high-orbit satellites, the average total number is about 20.

Angle of arrival (AOA) estimation and instantaneous frequency measurement (IFM) are important aspects of radar, communication, and electronic warfare, and there is an urgent need to find intelligent solutions that are faster and more accurate than traditional methods. Currently, deep learning (DL) has increasingly strong learning and feature extraction capabilities, which brings new opportunities to the field of electromagnetic signal processing. However, the models in DL have a large number of parameters, high storage requirements, and long computational latency, and not practical for deploying applications on resource-limited devices. In this paper, we build a complete AOA and IFM system for the first time, taking advantage of microwave photonics (MWP) technology, DL, and features of field programmable gate array (FPGA). Then, a novel hardware-friendly pruning-quantization-iteration compression (PQIC) algorithm is proposed for the post-processing of MWP signal measurement applications. By reducing the data width from 32 bits to 4 bits, the compressed algorithm reduces parameter storage requirements and hardware implementation complexity with negligible performance loss. Finally, we deploy the acceleration system and test the actual collected signal data on FPGA while running 200 MHz. The results show that with a 14× model compression and a 3.99× reduction in operations, the accuracy of AOA reaches 98.82%, with the mean absolute error (MAE) of 0.27°. More importantly, the running latency is only 20.78 μs, meeting the real-time processing requirements of MWP signal processing. It is also applicable to real-world signal intelligence processing and demonstrates superior performance compared to other existing algorithms in this field.

Road scene parsing is a crucial capability for self-driving vehicles and intelligent road inspection systems. Recent research has increasingly focused on enhancing driving safety and comfort by improving the detection of both drivable areas and road defects. This article reviews state-of-the-art networks developed over the past decade for both general-purpose semantic segmentation and specialized road scene parsing tasks. It also includes extensive experimental comparisons of these networks across five public datasets. Additionally, we explore the key challenges and emerging trends in the field, aiming to guide researchers toward developing next-generation models for more effective and reliable road scene parsing.

Metalens has shown its significantly ultra-light and ultra-thin features. However, large-aperture achromatic metalens is constrained by both maximum dispersion range and computational memory. Here, we propose a fully device optimizing framework that engineers phase dispersion and amplitude transmittance to create centimeter-size achromatic metalens operating in long-wave infrared regime (8–12 μm). Via wrapping group delay within a defined range and optimizing dispersion phase of desired wavelengths, chromatic aberrations can be effectively corrected. We verify our design by characterizing all-silicon 3.18-cm-diameter and 6.36-cm-diameter LWIR achromatic metalenses. Diffraction-limited tight-focusing can be achieved, and the normalized focal length shift is less than 3.3 × 10 ⁻⁴ . Thermal imaging performance is verified on targets of holes or letters with a diameter or line width exceeding 2 mm. These findings facilitate the development of large-aperture achromatic metalenses and open up possibilities for lightweight imaging systems in long-wave infrared.