Abstract：With the widespread use of encrypted communication, malicious attackers increasingly exploit encrypted traffic to conceal their activities, posing challenges to traditional signature-based and rule-based detection methods. Machine learning provides a novel solution for detecting malicious encrypted traffic. This paper reviews the applications of supervised learning, unsupervised learning, deep learning, and ensemble learning in this domain. Supervised learning identifies known attacks using labeled data, while unsupervised learning uncovers new attack patterns in unlabeled data. Deep learning enhances feature extraction capabilities in large-scale data environments, and ensemble learning strengthens system robustness through model fusion. The findings indicate that machine learning significantly improves the accuracy of malicious behavior detection, particularly in complex feature extraction and the identification of new attack patterns.

Abstract：As a training support base for aviation troops, the airfield station is responsible for organizing, commanding and providing professional services for the flight and equipment maintenance of the mission forces. The communication and navigation support throughout the flight mission is the core work of the airfield station's information and communication support. With the continuous acceleration of the development and update of information and communication, new problems that need to be urgently solved have emerged. In order to better adapt to the current development of the information and communication system construction, this paper studies the network information integration construction, adopts different simulation calculation models, and obtains the system development ideas for realizing efficient communication and navigation support and comprehensive control methods in the airfield station. It lays a solid theoretical foundation for the aviation combat command and support system that adapts to the conditions of information warfare and improves the information support capability of the system combat.

Abstract：Electrocardiogram (ECG) provides essential clues for detecting heart diseases, and more and more smart devices are developed to monitor the ECG signals. However, due to the convenience of use, only limited leads of ECG signals are measured from such smart devices, which may significantly affect the validity of disease judgement. To enhance the diagnose performance of these smart devices, this study suggests a Generative Adversarial Network (GAN)-based approach that combines Transformer and U-Net structures to autonomously reconstruct the whole 12-lead ECG signals from a single lead ECG signals. This study evaluates the proposed model on the PTB-XL and the Shaoxing People's Hospital 12-lead ECG dataset, then compares it with several state of art approaches. The code for this study can be found in https://github.com/Chaoquan-123/12-lead-ECG-reconstruction.

Abstract：Aiming at the cumbersome process of collection and labeling of commodity data set caused by rapid change of commodity packaging, this paper designs a commodity data set generation network based on Segment Anything Model (SAM) and Pixel to Pixel (pix2pix). The network uses multi-angle images of a single commodity as input to generate a data set similar to the actual settlement scene. The data set generation test was carried out on Retail Product Checkout Dataset(RPC) set, and the improvement of the generated data set on target detection effect was further verified on YOLOv7, Fast R-CNN and AlexNet target detection networks. The experimental results show that the generated data set can effectively improve the accuracy of commodity recognition, and has better substitution compared with the actual data set. Compared with the original data set, the recognition accuracy of the three networks generated by fusion data set is improved by 7.3%, 4.9% and 7.8%, respectively. Through this method, the efficiency and practicability of model training are significantly improved, and the manpower and material input required for traditional commodity data collection and labeling is reduced.

Abstract：Aiming at the existing asphalt mixing plant mixture loading semantic segmentation methods with low Mean Intersection over Union(mIoU) values and slow detection speed, a lightweight network RCS-UNet is proposed for semantic segmentation of asphalt mixing plant mixture loading state.Firstly, residual connections are integrated into the U-Net network to mitigate the gradient vanishing issue, promoting stability during training, enhancing convergence speed, and improving generalization abilities. Secondly, the Coordinate Attention(CA) mechanism is incorporated to boost the perception of positional and channel information, refining feature extraction and enabling a sharper focus on critical regions within the image. Finally, the standard convolution in the U-Net network is modified to depth-separable convolution in order to reduce the size and parameters of the model, so that the model has a lower resource consumption and a faster inference speed while maintaining a higher performance. The experimental results show that the accuracy, mIoU, and FPS of the improved model are 99.20%, 98.41% and 22.98, respectively, which are the highest compared with the classical model and the current state-of-the-art model. The best segmentation results are obtained.

Abstract：This article summarizes, analyzes, and interprets the main testing standards for electrostatic discharge of device level charged device models (CDM) both domestically and internationally. The differences and correlations between various standards were sorted out, and the application scope and technical points of each standard were clarified. The impact factors and control methods that affect the testing results of device level charged device models (CDM) were deeply studied. Relevant technical skills were proposed to ensure the consistency and accuracy of test results in the standard application process, providing guidance for the selection, testing, and engineering application of device level charged device model testing standards.

Abstract：The failure mechanisms and influencing factors of Si-based GaN HEMT devices under ESD events were systematically investigated. The study analyzed the failure voltage distribution of the devices under various bias conditions and utilized EMMI to pinpoint abnormal regions responsible for device degradation, followed by micro-area characterization. Furthermore, the impact of key process parameters, such as the geometric dimensions of the T-gate, including gate foot width, gate head width, and total gate width, on the anti-static robustness of the devices was examined. Mechanism analyses were conducted to elucidate these effects. The findings provide valuable insights and practical guidance for optimizing the ESD performance and enhancing the reliability of Si-based GaN HEMT devices.

Abstract：The traditional power divider structure is difficult to meet the requirements of high frequency and miniaturization of microwave systems due to obvious high-frequency crosstalk and large volume of impedance conversion lines. To solve this problem, this paper proposes a Ka-band power divider based on multi-line slit coupling structure. Starting from the circuit topology, the circuit design, modeling simulation and experimental verification of the device are carried out. The test results show that the device has good high-frequency performance and shows the filtering response with propagation zeros on both sides of the pass-band, which realizes the design of thin-film power divider with multi-wire gap coupling.

Abstract：The stability of the wavelength and output power of the fiber laser is susceptible to fluctuations in the pump drive current, and a high-precision, high-stability constant-current drive system has been designed to cope with this problem. The constant current source part adopts the closed-loop negative feedback principle to control the consistent current output through the Metal-Oxide-Semiconductor Field-Effect Transistor (MOSFET), realizing the precise control of the semiconductor laser (LD) drive current. The current-limiting protection circuit protects the LD from overcurrent shocks by utilizing the diode's limiting effect, and the current slow start uses a software-controlled method to protect LD from inrush current damage. The ADA4530 chip is used in the optical power detection circuit to ensure accurate detection of power output even at deficient light levels. The experimental results show that the circuit can achieve continuous adjustment of the drive current from 0 A to 1.2 A, with short-term (120 min) and long-term (24 h) current stability of 0.001 mA and 0.002 mA, respectively. The wavelength fluctuation of the laser center does not exceed 0.03 nm, and the stability of the output power does not exceed 0.084 3%, with an accuracy of less than ±0.15 mW.

Abstract：Addressing the current issues with elderly service robots, such as limited intelligent voice functions, insufficient perception capabilities, and low integration levels, this paper designed a robot with features including facial recognition, voice interaction, and fall detection. The robot is developed on the Robot Operating System (ROS) platform, equipped with a Jeston Nano main control board, a ZED2i stereo camera, and various other sensors. By employing facial recognition based on Haar features and LBPH algorithm, voice interaction using Alpaca-2 model, and fall detection based on ZED2i stereo camera, it provides more comprehensive and intelligent care services for the elderly. Experimental results demonstrate excellent performance of the robot in facial recognition, voice interaction, and fall detection.

Abstract：In recent years, unmanned aerial vehicles (UAVs) have been widely used in the military and civilian fields, especially since the Russia-Ukraine conflict, their advantages of low cost and strong functions are particularly prominent and have played a huge role. However, the usage environment of drones is complex and often uncontrollable, facing more serious security threats such as drone hijacking, network attacks, wireless communication interference, malicious access, and data theft. In response to these risks, traditional security measures such as wireless network security and data encryption are currently mainly adopted in drone systems. The zero trust architecture, as a new security model, emphasizes dynamic trust and provides new ideas for the security protection of unmanned aerial vehicle systems. This paper researches on zero trust security protection architecture, combined with typical applications of unmanned aerial vehicle systems, analyzes the security risks faced by unmanned aerial vehicle measurement and control, data transmission, and proposes a zero trust based unmanned aerial vehicle system security protection architecture to address these risks. It designs implementation solutions for zero trust technology in unmanned aerial vehicle measurement and control, data communication, and typical scenarios, providing strong support for the specific implementation of zero trust architecture in military and civilian unmanned aerial vehicle systems.

Abstract：In the inspection process of photovoltaic system, unmanned aerial vehicles need to accurately and quickly identify the defects of photovoltaic modules. Therefore, a photovoltaic module defect segmentation network based on multi-scale convolution and attention mechanism is proposed. Firstly, a multi-scale convolution module is added to each Stage of the traditional U-Net network to segment the defects of photovoltaic modules, and the pixel accuracy rate reaches 98.61%. Compared with the traditional U-Net and FCN networks, the accuracy rate is increased by 0.32% and 1.17% respectively, and the algorithm consumes 0.054 s. Compared with the comparison, the segmentation algorithm improves by 0.006 s~0.013 s; Then the split defect mask and the original image are operated. Finally, the defects (heat spots, cracks, bird feces) of photovoltaic modules are detected and classified by lightweight network MobileNetV3, with an accuracy of 98.82%. Compared with SqueezeNet, ShuffleNet V2, and GhostNet, the average detection time is increased by 0.43%, 1.08%, and 0.8%, respectively. Compared with Squeezenet, ShuffleNet V2, and GhostNet, the average detection time is increased by 0.002 s~0.036 s. The experimental results show that the defect segmentation network based on multi-scale convolution and attention mechanism has high accuracy and recognition rate.

Abstract：Based on a technology of 0.15 μm GaAs high electron mobility transistor (pHEMT), a 19~21 GHz high linearity power amplifier MMIC was demonstrated in this paper. In the scenario of signal transmission with high peak-to average power ratio（PAPR）,the efficiency and linearity of power amplifier（PA）have a critical impact on RF front-end performance. The amplifier incorporates a cold-mode linearization circuit cascaded at the power amplifier gate to compensate for the amplifier's phase distortion characteristics, thereby enhancing linearity and efficiency. To address the voltage sensitivity issue of the cold mode, an on-chip active voltage stabilization and temperature compensation bias circuit is employed to expand the dynamic range, reduce large dynamic distortion, and mitigate the deterioration of linearity caused by process variations and external discreteness. The room temperature measured results demonstrate that the achieved saturated output power of 22.3~22.8 dBm with 35.3%~36.5% saturated power-added efficiency (PAE) between 19 GHz and 21 GHz. Especially, the measured IMD3 at 19 GHz, 20 GHz and 21 GHz is below -30 dBc with an output power level of 19 dBm. For a 64-QAM orthogonal frequency division multiplexing signal with 6 dB peak-to-average power ratio and 100 MHz bandwidth，the PA exhibits an adjacent channel power ratio of -31.9 dBc、-33.2 dBc and -31.2 dBc when the error vector magnitude are 4.32%、4.13% and 5.3%.

Abstract：The traditional frequency selective surface（FSS）radome has the problem such as high structural profile and insufficient out of band cut-off performance. A design method for low-profile and band-pass FSS radome with steep out of band cut-off in K-band is presented. By using multilayers non-resonant constituting elements and the electromagnetic strong coupling techniques, a third-order band-pass radome in K-band is realized. It is worth noting that it not only has an extremely low profile，but also achieves a rapid steep transition from pass to stop band. Further，the K-band curved FSS radome prototypes was fabricated by using multi-layer circuit board pressing technology and radome molding technology, and the fabricating accuracy of FSS pattern as well as the radome surface shape is effectively ensured. Finally，the transmission characteristics of the radome and its influence on the antenna pattern are tested and verified. The measurement results agree very well with the simulated results, and the feasibility of the design method is demonstrated.

Abstract：To meet the demand for high-performance silicon-based low noise amplifier (LNA) with the development of communication systems, a gm-boost dual path noise-canceling circuit topology is proposed, which can reduce transistor noise and achieve low noise figure (NF). A gm-boost structure is introduced in the common-gate path to further improve the gain and noise performance. A K-band LNA is designed based on the topology mentioned above, consisting of two paths: a common-gate path and a common-source path, each of which acts as feedback branch of the other. Two paths are synthesized to achieve noise cancellation. The LNA is designed under a 90 nm CMOS SOI process, with a chip core size of 500 μm×280 μm. The simulation results show that a NF of 1.61~1.87 dB and a peak gain of 15.8 dB are achieved in 17~22 GHz. The input power with 1 dB gain compression is -4.3 dBm, demonstrating good noise and linear gain performance.

Abstract：A two-dimensional linear displacement microwave sensor based on split ring resonator (SRR) is designed in this paper using microwave technology to measure displacement. The sensor is composed of a stator and a mover. The stator is composed of two groups of SRRs with different sizes coupled with a transmission microstrip line. The mover is made of FR-4 dielectric substrate with single copper coating. When the mover moves in two dimensions, the resonant frequency of the two SRRs will change, and the transmission zero will also shift accordingly, thus establishing the relationship between the displacement and the transmission zero. In addition, the detection sensitivity is improved by loading the defect structure on SRR. Through electromagnetic modeling and simulation, the sensor produces two transmission zeros in the range of 1 GHz to 3.2 GHz, which can characterize the displacement of 0~6 mm in the x and y directions, and the sensitivity is 122 MHz/mm and 82 MHz/mm, respectively. The sensor is fabricated and tested, and the measured and simulated data are basically consistent, which confirms the effectiveness of the sensor design.

Abstract：In this manuscript, an angle estimation algorithm based on FOC-based MUSIC and LAD (FOCMLAD) is presented for millimeter wave (MMW) FMCW radar under spatially and temporally correlative clutter noise. The FOC-based MUSIC method is used to resolve and reconstruct the intermediate frequency (IF) signal. Then, it regards every sensor's signal expression as an independent regression form and uses LAD to achieve the angle. This algorithm effectively suppresses the correlative clutter-noise. The FOCMLAD algorithm mitigates the adverse effects of temporal correlation in correlative clutter-noise by incorporating the FOC-based MUSIC method and employing LAD to counteract its non-Gaussian nature. The FOCMLAD algorithm effectively suppresses the correlative clutter-noise's spatial correlation, as the clutter-noise distribution remains consistent across sensors regardless of spatial correlation. This effectiveness is validated through experimental results.

Abstract：In order to solve the problem of large volume and weight of high-precision integrated navigation system, a new 3D micro system packaging structure is proposed based on System in Package(SiP) technology such as multi-layer ceramic substrate, high-density wiring and 3D assembly. Using SIP processes such as Low Temperature Cofired Ceramic(LTCC) substrate high-density wiring and mounting, substrate stacking and high orthogonal assembly, the triaxial accelerometer, triaxial gyroscope, satellite navigation, geomagnetic meter and barometric altimeter are integrated into one packaging unit. The overall size of the integrated navigation microsystem product is only 3.1 cm×2.9 cm×0.96 cm, and the weighing is only 18 grams, and greatly expanded the application field. Through data fusion algorithms, navigation accuracy and reliability can be effectively improved, presenting broad application prospects.

Abstract：The navigation enhancement signals in communication and navigation integration signals are characterized by limited spectrum resources and non-continuous transmission. For the high-dynamic, short-burst navigation enhancement signals in the L-band of low-earth orbit (LEO) satellites, an improved segmented Fast Fourier Transform (FFT) two-stage time-frequency domain two-dimensional rapid acquisition method is proposed. Through the two-stage segmented coherent FFT acquisition process, the arrival time of the navigation enhancement signal and the signal Doppler shift estimation are obtained simultaneously, enabling fast and real-time signal acquisition processing in LEO scenarios. Simulation results demonstrate that when the signal carrier-to-noise ratio is above 48 dBHz, the rapid acquisition algorithm can adapt to signal Doppler shifts of ±40 kHz and accommodate short-burst transmission modes, with the signal Doppler estimation error controlled within ±80 Hz and the signal arrival time estimation error controlled within ±1 chip.

Abstract：The traditional Iterative Closest Point (ICP) method for point cloud registration has problems such as poor real-time performance, susceptibility to local extremum, and low registration accuracy. This paper proposes a three-step point cloud registration method based on feature point extraction, Principal Component Analysis (PCA) coarse registration, and ICP fine registration. Firstly, it defines the concept of local density in point cloud data and automatically selects points with higher local density as feature points. Then, it uses PCA to analyze the extracted feature points and calculates the required translation and rotation parameters for registration based on the principal component direction of PCA. Finally, it uses ICP to perform precise data registration. The experimental results show that the proposed method improves registration accuracy by more than 13.4% compared to the comparison methods, improves real-time performance by more than 38.2%, and exhibits higher adaptability under low signal-to-noise ratio conditions, with high application prospects.