Entity search,which is a critical task in information retrieval,aims to accurately identify target entities to a user query from a vast collection of documents.It plays a key role in enhancing user experience,enabling cross-domain applications,facilitating big data analysis,and supporting intelligent services.With the development of large language models (LLMs),they have demonstrated outstanding performance across various fields.The powerful capabilities of semantic understanding and generation of LLMs can significantly improve the accuracy of entity search.However,the evaluation of LLMs specifically for entity search tasks has not yet been fully explored.Therefore,an evaluation framework for LLMs tailored to entity search tasks are proposed，which can not only improve the evaluation framework but also provide valuable insights for further optimization and application of these models.By constructing and publicly releasing a cross-domain Chinese entity search test set,nine open-source LLMs are tested and their practical performance in entity search is demonstrated.Through comparative experiments,the performance of LLMs are evaluated and analyzed from multiple perspectives,providing empirical evidence for their application in the entity search domain and offering new insights for future research.

Target action information is an important component of intelligence data.A key criterion for the reliability of its action information is whether the target actions are spatio-temporally consistent.By formally defining a target action information repository as the detection object and a target action spatio-temporal knowledge base as the detection basis,a modeling method for the problem of spatio-temporal consistency detection of target actions is proposed.Based on this,the spatio-temporal consistency detection problem for target actions is defined,and a metric system for measuring spatio-temporal consistency is established.Finally,an algorithm for detecting the spatio-temporal consistency of target actions is designed using Answer Set Programming (ASP).The experimental results of quality detection for typical military target activity information indicate that the research findings have practical reference value.

The rapid development of artificial intelligence has deeply penetrated various industries,with deep neural networks (DNN) being widely applied in multiple fields.However,the popularity of the application exposes the serious threat of adversarial attacks on the vulnerability of intelligent models.Adversarial attacks can lead to model failure,particularly in critical areas such as commercial and military security,potentially resulting in severe consequences.These attacks work by crafting carefully designed inputs to disrupt the normal functioning of models,thus compromising system security and reliability.To comprehensively and scientifically evaluate the robustness of different model algorithms,a quantitative evaluation framework is proposed based on Principal Component Analysis (PCA) for the first time,which encompasses several key evaluation metrics,including misclassification rate,imperceptibility,and attack efficiency.By testing over 20 model algorithms and using PCA to reduce the dimensionality of high-dimensional data,the main evaluation factors are extracted,simplifying the data structure and ultimately producing an overall score for each algorithm.Experimental results demonstrate that the proposed evaluation method is effective and reliable,providing scientific guidance for research on model robustness.

Artificial Intelligence (AI) technologies are increasingly applied in metrology,bringing new vitality to traditional measurement methods.The applications and development trends of AI in metrology is systematically reviewed from micro,meso and macro levels.At the micro level,the innovative applications of AI in optical measurement,mechanical measurement,quantum metrology,and automated calibration，is analyzed.At the meso level,the integration of intelligent sensor networks,edge AI,and blockchain technologies are discussed.At the macro level,the practical applications of AI in environmental monitoring and industrial manufacturing is summarized.Based on literature analysis,it is found that AI metrology research has shown rapid growth over the past decade,with a significant increase in research intensity after 2018.Focusing on key issues such as standardization,uncertainty quantification,and model interpretability,current challenges and prospects for future research directions are explored.

Adversarial robustness is a crucial component of evaluating the security of intelligent models.Currently,research on transfer attacks is primarily limited to fixed-budget attacks,with a lack of studies focusing on minimal-norm attacks.Furthermore,there is no unified and comprehensive evaluation framework for assessing transfer attacks or measuring attack transferability,which poses significant challenges to the security assessment of intelligent models.To address these issues,a minimal-distance-based framework for assessing adversarial attack transferability is proposed.By the framework，a search method is used to identify the minimum budget for successful transfer of attacks firstly.Then,the transferability of different attacks are compared using an overall score that combines perturbation size and attack success rate,as well as an optimality measure.The overall score reflects the varying degrees of transferability among different attacks and allows for comparison of target models' robustness against transfer attacks.The optimality measure quantifies the proximity of an attack to an optimal solution.Texting results demonstrate that our method outperforms existing state-of-the-art approaches and yields several empirical conclusions.

In order to improve the flexibility and rapidity of 3D laser projection system,a calibration method of 3D laser projection system based on object spatial sampling is proposed.A point on the object to be projected is directly mapped to the pixel plane of the monocular camera through perspective transformation,and then the homography matrix between the camera pixel plane and the digital quantity plane of the galvanometer is used to obtain the corresponding digital quantity control signal of the point.The whole 3D laser projection system can be calibrated by taking two images.The experimental results show that the calibration time of the projection system calibrated by this method is less than 5 min,and the projection positioning error is less than ±0.4 mm in the range of 3 m,which meets the application requirements of industrial scenarios.Meanwhile,the method is meaningful for the cooperative work of multiple galvanometers in large objects to be projected.

The traditional dynamic calibration methods of heat flow sensors,such as the thermal wind tunnel method and the excitation tube method,can no longer adapt to the future demand for dynamic calibration of high-speed heat flow sensors,and there are problems such as small output heat flow,uneven heat flow,and single signal.Using a semiconductor laser as the heat flow source,the built dynamic test platform is capable of generating excitation signals such as step,pulse,square wave and swept sine.On time domain,using pulse signals for dynamic calibration of the sensor,the time constant was measured at a maximum of 87 μs,and the change of time constant is small when the pulse width changes within a certain range;on frequency domain,using a periodic square wave signal sensor for dynamic calibration,through the FFT（Fast Fourier Transform）,the sensor and photodetector spectral characteristic curve to calculate the amplitude ratio is obtained and the normalization is carried out to get the frequency of its response at 2 160 Hz.

Angle is the basic parameter of geometric measurement,and the results of angle measurement affect the accuracy of measurement,detection and calibration.With the progress of precision machining and manufacturing technology,optical gyroscopes are widely used in the measurement and calibration of dynamic angles,and play an important role in inertial measurement with the advantages of high precision,good reliability and fast response speed.The basic working principle of optical gyroscope is introduced,the research status of optical gyroscope angle measurement technology at home and abroad is expounded,and the application of ring laser dynamic goniometer and angular velocity standard device in angle measurement is introduced.Nowadays,optical gyrometer goniometry is developing rapidly in the direction of digitalization,automation and intelligence,and future research will further improve the sensitivity,stability and dynamic performance of angle measurement.

To ensure the diaphragm-based F-P hydrophone shares the sensitivity and resistance to net water pressure,a MEMS fiber F-P hydrophone is proposed,fabricated and tested.A silver membrane is fabricated using a sacrificial layer based membrane process as the sound pressure sensitive unit,and air is used as the material in the F-P cavity,thereby the sound pressure sensitivity of the fiber optic hydrophone is improved.The performance of MEMS fiber optic F-P hydrophone is tested and analyzed.The results show that the phase sensitivity of the fiber optic F-P hydrophone is -157±3 dB re 1rad/μPa in the range of 80～800 Hz,and the dynamic range at 400 Hz is 56.21 dB.This hydrophone has advantages such as simple structure and high sensitivity.

Angular velocity sensitivity is an important item for evaluating the accelerometer error caused by angular motion.Current testing standards of accelerometer are lack of specific testing procedures and makes it hard to compensate the angular motion error.At first,testing principles are proposed based on definitions of angular velocity sensitivity in standards home and abroad.Then,by reference to rotating calibration method of accelerometer,co-axis dual-turntable centrifuge is introduced to measure angular velocity sensitivity.At last,a specific testing procedure of angular velocity sensitivity is described by taking single-axis accelerometer as an example.

In response to the problems of long computation time and slow convergence speed in existing registration algorithms,a registration algorithm is proposed based on point cloud skeleton and edge contour for measured point cloud data of headgear workpieces.Firstly,point cloud data is downsampled based on voxel filtering algorithm.Then,the L1 center skeleton extraction algorithm and edge extraction algorithm are used to calculate the skeleton points and edge feature points of the point cloud as key points of the point cloud model.Finally,the reference point cloud model and target point cloud key points are used as inputs for the registration algorithm.Point cloud registration is performed based on principal component analysis algorithm and ICP algorithm.Through simulation experiments on the publicly available dataset Cactus and measured headcover point cloud data,the proposed algorithm has high registration accuracy and less computational time.

Corresponding verification methods for different verification objects and professional characteristics are provided in GJB 9399-2018 General Methods for Military Laboratory Verification.The verification methods and steps of the measurement process are focused on,the verification of the multi-objective theodolite verification device is taken as an example to experimentally verify the verification of the measurement process,the verification method is proved to be effective in ensuring the confidence of the verification or calibration status of standard devices,and the measures to deal with the loss of control of measurement standards are proposed.