Firstly， this paper points out the challenges faced by the design of multi-target tracking algorithms and the limitations of traditional methods. Secondly， a literature review and analysis of two types of algorithms are conducted： detection-based-tracking and joint-detection-tracking. Then， the commonly used evaluation indicators and publicly available datasets in the multi-object tracking algorithms were summarized， and the performance indicators of the two types of methods were analyzed. Finally， based on the current research status， the predictions and outlooks on the problems to be solved and the focuses of the future researches are made.

Microcapsule technology in intelligent maintenance of asphalt pavements is comprehensive analyed， elucidating the self-healing mechanism of microcapsules in asphalt pavements， synthesis technology， physical and mechanical performance testing techniques， and the evaluation methods of self-healing effects. The development trend of this technology in asphalt pavement maintenance includes research on the self-healing mechanism of microcapsules-aging asphalt materials in macro-micro-microscopic levels， the synthesis of multiple sustained-release microcapsules equivalent to the size of cracks in asphalt mixtures， exploration of the relationship between microcapsule preparation parameters and asphalt pavement maintenance requirements， improvement of microcapsule preparation processes， and the establishment of a self-healing evaluation model after the application of microcapsules in asphalt pavement materials based on actual engineering conditions， providing a reference for the application and development of microcapsules in intelligent maintenance of asphalt pavements.

In order to investigate the influence mechanism of complex gas flow behaviour on the lubrication effect of cylindrical gear with variable hyperbolic circular-arc-tooth trace under oil injection lubrication environment due to its special tooth shape. The computational fluid dynamics is used to model the computational basin， simplify and simulate the calculation， analyse the interference between the high-speed gas flow field and the oil injection beam， and determine the optimal oil injection angle through the streamline diagram. The oil distribution， average oil volume fraction and oil pressure on the inspection surface were extracted， and the influence of different oil injection parameters on the lubrication of the tooth surface was obtained. The results show that the lubrication effect of the gear transmission can be improved by biasing the oil injection angle towards the main wheel by 10.73o， and by increasing the oil injection speed， decreasing the oil injection distance， and lowering the gear rotation speed appropriately. The findings of this study provide a theoretical basis for the design and parameter optimisation of the oil injection lubrication system and cooling system of the gear transmission.

In order to obtain the influence of corrosion on the shear performance of pure self-tapping screw connection and self-tapping screw-pull rivet hybrid connection of cold-formed thin-walled steel， the shear test of the above connection specimens under corrosion condition was carried out. The results show that the mechanical properties of the connecting steel plate decrease linearly with the increase of weight loss rate. Under the same weight loss rate， the shear bearing capacity of the pure self-tapping screw connection specimen decreases more than that of the self-tapping screw-pulling rivet hybrid connection specimen， and the longer the corrosion time， the more obvious the phenomenon. The corrosion factor is introduced to determine the bearing capacity calculation formula of this kind of connection， which provides a corresponding reference for the safety assessment of the bearing capacity of corroded cold-formed thin-walled steel connections.

In order to optimize the design of excavator bucket structure and reduce the excavation energy consumption， this paper builds a joint simulation and optimization platform based on SOLIDWORKS， ADAMS， EDEM and MATLAB， and establishes a global optimization method for a medium-sized backhoe hydraulic excavator， which takes the maximal excavation load and strength of the bucket as the constraints， and the quality of the bucket and the unit excavation energy consumption as the optimization objectives. Through combined simulation of multibody dynamics and particle dynamics， load characteristics of the excavator during operation under different working conditions were obtained. Pearson correlation coefficients show a very strong correlation between experimental and simulated loads on the key hinge points of the bucket rod and boom， confirming the feasibility of using combined simulation to replace actual excavation operations. Multilayer perceptron （MLP） is used as the surrogate model and the parameters and hyper-parameters of MLP are optimized using Adam， Hyperband algorithm. The coefficients of determination R² of MLP for bucket mass， maximum excavation load， excavation mass and excavation energy consumption are 0.999， 0.943， 0.933， 0.984 for Case Ⅱ； the R² of MLP are 0.999， 0.944， 0.918， 0.925 for Case Ⅳ. The optimized MLP is combined with the NSGA-Ⅱ algorithm to iteratively optimize the bucket structure. The results show that the optimized bucket achieves a mass reduction of 7.06 % and a unit excavation energy consumption reduction of 6.47 % under the premise of meeting the requirements of the maximum excavation load and strength.

In order to study the reasonable calculation method of the transverse bending moment of the composite box girder with corrugated steel webs and the influence of the distortion warping stress distribution range of the corrugated steel webs on the transverse bending moment， Based on the traditional TYL frame method， considering the deformation coordination relationship between distortion displacement and sheet frame displacement， this paper proposes a transverse bending moment frame analysis method for composite box girder with corrugated steel webs by using a new calculation model that meets the equilibrium conditions， and a general formula for calculating the transverse bending moment is established. In view of the three different distortion warping stress distribution range of corrugated steel webs， the transverse bending moments at each key point of the straight webs and inclined webs composite box girder with corrugated steel webs were calculated by two frame analytical methods and finite element methods. The results show that for the straight webs box girder， the method proposed in this paper has the same calculation accuracy as the existing calculation method. For the inclined webs box girder， the transverse bending moment calculated by the proposed method is consistent with the distribution law of the calculation results of the finite element method， and the proposed method has higher calculation accuracy than the existing calculation methods. The different webs distortion warping stress distribution ranges of the composite box girder with corrugated steel webs have little influence on the results of the transverse bending moment， and when the transverse bending moment is actually calculated， it can be considered that the corrugated steel webs does not bear the distortion warping stress.

To facilitate the dexterous operation of a mobile manipulator in a complex environment， we conducted configuration optimization based on task priority for a redundant wheeled mobile manipulator （WMM）. To achieve this objective， we initially established the kinematic models for the wheeled mobile base and the multi-degree-of-freedom manipulator separately. Subsequently， we constructed the kinematic model for the WMM system based on these foundations. Subsequently， we designed an inverse kinematic model for the WMM using the task priority framework. Configuration optimization was then performed for various targets， including vehicle-arm dual trajectory tracking， singularity avoidance， and joint limit avoidance. Finally， the performance of the WMM in a complex environment is simulated using SolidWorks and Matlab/Simulink. The simulation results validate the effectiveness of the proposed method.

To study the global stability and design method of a novel shuttle-shaped double-restrained buckling-restrained brace （SDR-BRB）， this article first uses the equilibrium method to derive the calculation formula for the elastic buckling load and the restraining ratio of SDR-BRB with hinged ends. Then， the distribution function of lateral deformation and bending moment of SDR-BRB is constructed considering the influence of global initial geometric imperfections， and based on the yielding criteria of the outmost fiber for the restraining member section， the calculation formula of the three types of the restraining ratio requirement of SDR-BRB is obtained. On this basis， the ABAQUS finite element model verified by experiments is established， and the elastic-plastic finite element numerical analysis of its load-carrying capacity under monotonic loading and hysteretic performance under cyclic loading is carried out respectively. Finally， the global stability design method of SDR-BRB based on the restraining ratio requirement is proposed. The results show that the restraining ratio has a significant effect on the global stability， load-carrying capacity， and hysteretic performance of SDR-BRB. Based on the three types of the restraining ratio requirement， SDR-BRB can be divided into delayed buckling member， load-carrying BRB， and energy-consuming BRB. The design method proposed in this article can accurately predict the buckling behavior of the load-carrying capacity and hysteretic performance of SDR-BRB， and has good applicability， which can be used as the global stability design criterion of SDR-BRB.

In order to solve the complex problem of reliability analysis of CNC machine tool system， this paper combines the failure mode influence and hazard analysis method with the dynamic Bayesian network to establish a reliability analysis model. Firstly， the failure causes， failure modes and main fault parts of CNC machine tools were analyzed， and the risk assessment of the failure causes was carried out in combination with expert opinions. Then， based on the failure mode table and fault data of the CNC machine tool， the structure and parameters of the dynamic Bayesian network reliability model were constructed. The state change of the system， the importance of the faulty parts， the importance of the subsystem and the state changes of each component are analyzed by the model， and the results of the model analysis are compared with the traditional Monte Carlo method.

Aiming at the problems of slow clamping force response and deteriorating control accuracy caused by uncertainty factors such as internal system variable coupling， parameter perturbation， and external time-varying interference in electronic mechanical brake， an improved sliding mode control method for clamping force of electronic mechanical brake based on super-twisting extended state observer is proposed in this paper. Firstly， an improved reaching law is designed， which introduces a variable function gain term and a sliding mode surface power term on the basis of the traditional exponential reaching law. This increases the convergence speed while weakening chattering phenomenon. Based on the improved reaching law and sliding mode surface， an improved clamping force controller is designed. Secondly， design an extended state observer based on the super-twisting algorithm to estimate system disturbances， while feedforward the estimated disturbance values to the controller for compensation. Thirdly， the stability of the system is proven through the Lyapunov theorem. Finally， compare the algorithm proposed in this paper with algorithms such as double power reaching law， fast power reaching law and exponential reaching law in the test bench experiments. The results show that the clamping force control of the electronic mechanical brake under the algorithm proposed in this paper has higher response speed， control accuracy， and anti-interference ability.

A method for relative measurement and grouping the diameter of large-sized bearing steel balls based on micro machine vision is proposed in this article. The method transformation from the difference in diameter of steel balls to the change in slit width is achieved through a micro-deformation elastic mechanism. The steel balls to be tested in the same batch are used as reference balls， and the difference in slit width between subsequent steel balls and reference balls is used as the grouping basis. CMOS cameras and coaxial microscopes are used to capture slit images， and preprocess the images with grayscale， mixed bilateral filtering， and threshold segmentation. Edges in the image are extracted by the improved Canny operator and Zernike moment sub-pixel algorithm to achieve accurate localization， the Ransac method is used to fit edge points， and the "one point， one line method" is proposed to achieve feature localization between different images， completing slit width calculation within the feature area. Five groups of micro-deformation elastic mechanisms are built using different materials， according to the repeatability comparison results， the hardened 304 elastic plate corresponding mechanism was selected to build a prototype. The experiments results show that the method can measure and group steel balls with a diameter of 30~100 mm， and the detection accuracy can reach 0.2 μm. The difference in diameter of steel balls in the same group after grouping is ≤0.4 μm.

This paper adopts the roll bending forming process to bend profiles with a "日" shaped cross-section into commercial vehicle bumper beams. During the actual processing and forming process， profiles often exhibit forming defects such as wrinkling and cross-sectional distortion. To address this， an in-depth numerical simulation of the coreless mandrel-supported profile forming process was conducted using ABAQUS finite element software， aimed at analyzing the mechanism of defect formation. Based on this analysis， a new type of mandrel—a novel magnetically controlled mandrel—was designed and studied. This mandrel not only exhibits excellent forming quality in bending "日" shaped structural profiles， but also significantly improves the performance of the traditional mandrel forming processes. Moreover， by altering forming parameters such as the bending radius and die gap， the paper explored the limits of process parameters for profile bending， seeking to find the extreme process parameters while ensuring accuracy.

The effects of extrusion temperature on the microstructure and mechanical properties of longitudinal welds in aluminum matrix composites were investigated using graphene-reinforced 7075 aluminum matrix composites （GNP/7075Al） to enhance the welding quality during shunt die extrusion process.The results show that with the increase of extrusion temperature， the diffusion rate of metal atoms increases， the coarse second phases are obviously reduced， and more second phase particles are gradually dissolved into the aluminum matrix as well as the longitudinal weld. Appropriately increasing the extrusion temperature is conducive to forming a good welding interface and improving the quality of longitudinal weld. However， if the temperature is too high， the soft processes for the matrix such as dynamic recrystallization will decrease the deforming hardening of the extrusion， the weld pressure decrease， and the structure integrity of the graphene will be destroyed， decreasing the properties of the extrudate. At the extrusion temperature of 480℃， the welding line is the narrowest and the weld property is the best. Moderate dynamic recrystallization， solid solution strengthening of the matrix and the best state of the enhanced phase graphene brought the best weld quality and comprehensive mechanical properties of the extrudate with the tensile strength of 279.98 MPa.

To explore the impact of mixed driving for human-driven vehicles （HDV） and Connected and autonomous vehicles （CAV） on traffic flow， a freeway heterogeneous traffic flow model was established. The enhanced intelligent driver model （EIDM） was utilized to describe the car-following behavior of HDV， and the EIDM model was improved to describe the car-following behavior of CAV considering the headway， speed difference and acceleration of multi-vehicles. Then， the minimizing overall braking induced by lane changes （MOBIL） model was utilized to describe the lane-changing behavior of HDV， and the game theory is introduced to consider multi-vehicle competition and cooperation， autonomous and cooperative lane-changing models of CAV were established. Through simulation experiments， the rationality of the model was evaluated and the operation characteristics of heterogeneous traffic flow were analyzed. The research shows that compared with existing models， the model built in this paper demonstrates significant advantages in traffic flow stability and driving comfort. The increase of CAV penetration rate is conducive to the improvement of traffic capacity， especially when the CAV penetration rate is greater than 0.4， the effect is more significant.

In order to better describe the structure and behavior of the shared lane keeping assist system （LKAS）， and hence to enable the intelligent system to predict the driver's steering intention during the co-driving， a novel shared steering control algorithm based on situation prediction factor is thus proposed. Firstly， a driver steering model based on visual preview characteristics and neuromuscular dynamics is introduced. The parameters of such model are identified by particle swarm optimization （PSO） algorithm. By the integration of driver steering model and vehicle-road model， the closed-loop driver-vehicle-road model is established. Secondly， by employing model predictive control （MPC） framework， the decision-making of desired steering torque for shared LKAS is transformed into an online quadratic programming （QP） problem formulated as a quadratic objective function with multiple linear inequality constraints. Then， in order to achieve smooth transition of driving control authority， a driver-automation control authority model is designed using comprehensive preview error approach. Finally， the comparative experimental results demonstrate that by introducing the driver steering model， the intervention from intelligent system， negative intervention especially， can be effectively reduced to a certain extent， but the improved driver-automation friendliness is at the cost of sacrificing the accuracy of lateral motion control.

When a six-lane freeway closes two adjacent lanes， two upstream transition zones need to be set up. However， the current regulations do not specify the specific value of the distance between the two. Considering the mixing of automatic cars （AC）， the problem of the value of this distance is studied. Firstly， the traffic efficiency in the warning zone is defined as the evaluation basis for the upstream transition zone spacing D， and a dynamic comprehensive evaluation function is constructed to classify the traffic efficiency in the warning zone. Secondly， through SUMO software modeling， simulation experiments with different values of D were conducted based on different traffic conditions， such as different traffic volumes V， human-driving truck mixing rate T， and AC penetration rate P， to analyze the changes in traffic efficiency in the warning zone. Then， based on the classification of traffic efficiency， the range of values for D under different traffic conditions is explored through experiments. Finally， analyze the impact of D on the work zone environment and traffic efficiency under different traffic conditions. The simulation experiment shows that the larger the D， the higher the traffic efficiency in the warning zone， and this change is non-linear， with the increase rate slowing down as D increases. As V or T increases， the traffic efficiency in the warning zone decreases， and increasing D is necessary to maintain the same traffic efficiency in the warning zone. As P increases， the traffic efficiency in the warning zone first increases and then decreases. There exists an optimal P that maximizes the traffic efficiency in the warning zone and minimizes D. Compared to taking the minimum value of D， increasing D can reduce CO₂ emissions. Under appropriate traffic conditions， increasing D can simultaneously improve the traffic efficiency in warning zone and reduce the average travel time.

To construct a driving risk assessment system for heavy goods vehicle drivers with good generalization and interpretability， based on trajectory data from heavy goods vehicle in Yunnan Province， this paper establishes driving risk assessment indicators covering driving style risk， fatigue risk， and driving environment risk. It then uses Pearson correlation analysis to select 43 key indicators from the initial set of driving risk assessment indicators. Lastly， by applying the CRITIC （criteria importance though intercriteria correlation） empowerment method to determine characteristic weights of various indicators， it proposes a method for assessing the driving risk of heavy goods vehicle drivers. The findings indicate that approximately 82.16% of drivers have driving style risk scores below 30 points， suggesting that the majority of heavy goods vehicle drivers in high-altitude mountainous areas have low risk assessment values. The fatigue risk assessment reveals that 42.3% of samples score below 12 points， indicating that heavy goods vehicle drivers in Yunnan Province engage less frequently in long-distance transportation， with fewer instances of continuous long-duration driving. The driving environment risk assessment is most densely distributed in the 15 to 35 points range， accounting for 75.46%， which suggests that cargo transportation mainly depends on riskier arterial roads.

Extreme weather in recent years has caused serious damage to the highway power grid system， how to reasonably deploy mobile power vehicles is the current highway fault repair in the urgent need to solve the problem. Therefore， this paper proposes an optimal scheduling method for mobile power vehicles on highways under extreme weather. Firstly， the Monte Carlo simulation method is used to construct mathematical models of ice-covering load， wind load and insulator flashover under extreme weather through historical weather data and actual line parameters， to obtain the vulnerability model of the transmission line， and then to determine the fault conditions in the whole system. Secondly， the Monte Carlo simulation method and the convergence condition of the objective function are used to find out the optimal access point of the mobile power vehicle， and the mobile power vehicle is dispatched to improve the system resilience. Finally， the simulation analysis is carried out by using the data of the self-consistent energy system of a highway service area in Xinjiang. The results show that compared with the commonly used method of fixing the access point of mobile power vehicles， the dispatching method proposed in this paper increases the proportion of loads that are restored to the power supply by 12%， which enhances the efficiency of mobile power vehicle and the resilience performance of the self-consistent energy system.

In order to address the stability and durability issues of coal gangue-red clay subgrade in hot and humid areas of southern China， this study investigates the strength and deformation characteristics of a coal gangue-red clay mixture through a series of laboratory tests， including compaction， UCS， CBR， and triaxial tests. The results demonstrate that the maximum dry density initially increases and then decreases with increase of clay content， while the optimum moisture content gradually increases with higher clay content. The UCS value and CBR value exhibit an initial increase followed by a decrease as the clay content rises until reaching their peak at 40%. Permanent deformation initially decreases but then increases with increasing clay content. It also increases with higher deviator stress， loading cycles， and wet-dry cycles but decreases with increased confining pressure. Based on these findings， a comprehensive prediction model for permanent deformation is established considering physical state， stress state， loading state as well as environmental conditions. These research findings can provide valuable references for construction practices in filling coal gangue-red clay subgrade.

In order to accurately analyze distortion effect of supported curved steel box composite girders with corrugated webs （CSBCG-CSWs） and flexible diaphragms， in this paper the distortion warping displacement mode of CSBCG-CSWs was established based on the relationship between distortional warping displacement， bending-torsion coupled displacement and internal force of composite girder， and the potential energy of the additional distortion frame deflection caused by the bending-torsion coupled effect was derived. The anti-distortion stiffness of flexible diaphragms was solved by using virtual work principle. The distortion differential equation was established based on the energy variational method， and the influencing factors of CSBCG-CSWs distortion effect were analyzed. The results if the study show that the analytical solution in this paper was in good agreement with the experimental value and Abaqus numerical solution. Compared with the results without the intermediate diaphragm of plate-type， the maximum distortion bi-moment is reduced by 77.7%， 85.6% respectively when 1~2 passes intermediate diaphragm of plate-type are set. Compared with the results of setting intermediate diaphragms of X-type and K-type， the maximum distortion bi-moment is reduced by 25% when diaphragm of plate-type is set. The distortion effect caused by concentrated load is mainly affected by the presence or absence of diaphragms at the loading point. The uniform load can better reflect the overall distortion effect of the composite box girder with intermediate diaphragm. When the distortion stress ratio of the steel box composite girder with corrugated webs is controlled within 10% and 5% respectively， the intermediate diaphragm spacing needs to be satisfied： the straight steel box composite girder is less than 8m and 6m， and the curved steel box composite girder is less than 6m and 5m respectively，and distortion stress ratio can be reduced when the self-gravity load is taken into account. The increase of intermediate diaphragm thickness has no obvious enhance on suppressing effect the distortion effect of the straight and curved steel box composite girder with corrugated web when the thickness of the intermediate diaphragm of plane-type is more than 6mm. When the central angle increases from 0° to 28°， the distortional warping normal stresses of 1/4-span and 1/2-span cross-setions of CSBCG-CSWs increase by 74.6% and 87.4%， respectively.

Affected by factors such as changes in lighting conditions and image distortion in highway tunnels， the difficulty of extracting features at different scales from lining crack images has increased， which has affected the recognition of small cracks in highway tunnel lining. Therefore， a method for identifying of small cracks in highway tunnel lining based on deep learning SSD algorithm is proposed in this paper. Firstly， using the weighted average method to perform grayscale processing on highway tunnel images， extracting line segment features from the images through line segment detection algorithms， obtaining edge detection results of the image in both horizontal and vertical directions using Sobel operators， and correcting distorted images based on perspective geometry principles； secondly， introducing deep learning SSD algorithm， using regression model to adjust the position of prior boxes in SSD algorithm， and predicting the category of each prior box on the effective feature layer， combining non maximum suppression technology， the optimal prediction box is selected to achieve accurate identification of cracks. Experimental results show that the lining cracks identiflcation by this method are basically consistent with the cracks in tunnel images， with an AUC value of 0.956， which can effectively improve the accuracy of identifying small cracks in highway tunnel lining and have better application performance.

Aiming at the problem that the existing bridge crack detection methods do not fully extract the rotation feature of bridge cracks and have low detection and segmentation accuracy， a bridge crack detection method based on improved Mask RCNN with rotation self-attention was proposed. Firstly， on the basis of the Mask R-CNN instance segmentation network， the ViTAE network based on Transformer learning is used as the backbone feature extraction network to improve the detection and segmentation accuracy of cracks. Then， a rotating variable window self-attention mechanism was designed to integrate into the bridge crack detection network to improve the detection ability of the feature extraction network for crack rotation features. Finally， the deformable convolution was used to further fit the irregular geometry of cracks to strengthen the recognition ability of crack feature information. Experimental results show that compared with the original Mask R-CNN detection and segmentation method， the accuracy of the proposed method is improved by 4.85%， the recall rate is increased by 13.95%， and the F₁-score can reach 91.66%. The proposed method can extract crack features more fully， achieve more accuratecrack detection， and is superior to the comparison methods in subjective and objective evaluation.

There are numerous multi-step relationship paths between entities in the knowledge graph to indicate semantic relationships between entities， as well as the neighborhood heterogeneity between relationship structures and attribute structures. In response to this problem， An entity reliable path information semantic augmentation model is proposed in this paper， which simultaneously captures and aggregates multi-source information of aligned entities and their heterogeneous neighbors， an initial reliable path reasoning algorithm to generate. The model aggregates the relationship structure， attribute structure， and entity name information reliable path of the entities for semantic augmentation， which solves the problem of domain heterogeneity in knowledge graph alignment. The paper evaluated the entity reliable path information semantic augmentation model on three datasets（WK31-15K， DBP-15K and DWY-100K）show that this model is improved by 1.5%~3.2%compared with the state-of-art entity alignment method Hits@1， which shows that the proposed method has better performance.

To address issues such as low recognition of lesion areas in chest X-ray images and the difficulty in accurately capturing the spatial positions of lesions， a multi-scale attention information multiplexing network that helps improve the dassification accuracy of chest X-ray images was proposed in this paper. Firstly， by introducing multiple spatial information multiplexing blocks， the network enhances the positional connections between disease regions on feature maps and across channels； Secondly， through a multi-scale integration attention blocks， the network integrates multi-scale image feature information to automatically capture disease location variations and flexibly focus on key pathological information； Finally， the problem of imbalanced distribution of chest disease samples was alleviated by using an asymmetric shift focus loss function. Multiple experiments on the publicly available datasets ChestX-ray14 and CheXpert have shown that the average area under curve （AUC） value of the proposed network on two datasets reached 0.847 and 0.901 respectively， which is superior the more advanced network models in recent years. This indicates that the network can effectively improve the classification accuracy of chest diseases.

The perception of surrounding objects by vehicle autonomous driving is an important means to ensure traffic safety. Object detection model with deep learning is adopted widely， but they requires a large amount of annotated data for training. In this paper， an active vision object detection model is proposed using Gaussian mixture distribution to estimate the uncertainty of unlabeled images， reduces the dependence of model training on labeled data. Firstly， the mixed density network is adopted as the detection head， taking the image feature extracted by the deep neural network as input， estimates the probability distribution of classification and location of the object predicted boxes. Secondly， the classification score of the object predicted boxes is mapped into the probability space， and the classification uncertainty of the object is calculated by edge uncertainty； the location variance of the predicted boxes is used to measure the location uncertainty of object. Finally， the most unstable samples were selected for labeling. The results on the VOC dataset show that compared with other typical active learning sampling strategies， the proposed model achieved the best performance. The proposed model using only 54% of the data annotation volume can achieve the 98.8% performance of YOLOX with supervised learning， which saves up nearly 45% of the data annotation volume.

In the semantic similarity model based on contrastive learning， insufficient information exchange between sentences with different positive examples and the scarcity of hard negative examples in traditional negative sample collection strategies make it difficult to capture the subtle feature differences between sentences， thereby inability to accurately capture the similarity between texts. This article proposes a semantic similarity method based on augmented positives and interlayer negative. By designing a dynamic neighborhood mechanism to fuse information between different positive examples and proposing a method for generating difficult negative examples， the correlation of semantic similarity judgment is significantly improved. Firstly， retrieve sentence embeddings with similar semantic features to positive examples from dynamic neighborhoods， concatenate them with positive examples， and then obtain augmented positive examples through self attention aggregation， thereby fusing information from different positive examples. Secondly， a method for generating difficult negative examples is proposed， which takes the sentence representation in the middle layer of the model as the original positive example of difficult negative examples， and intraduces cross entropy loss as punishment to improve the negative example sampling strategy. The experimental results show that in the semantic similarity task dataset STS2012~STS2016， STS-B， SICK-R， the method proposed in this paper has a significant effect， with Spearman correlation coefficients increasing by an average of 1.09 and 0.34 percentage points compared to advanced models on the basis of BERT-base and BERT-large， respectively.

At present， the text watermarking algorithms based on modification and generation generally have the problems of large semantic ambiguity， limited embedding capacity and insufficient security. In view of the above problems， this paper proposes an improved algorithm based on chaotic AES and synonym expansion. By designing a context matching mechanism based on Sentence-Bert， the algorithm can not only ensure that the context after synonym replacement is highly similar to the semantics of the original text， but also distinguish the semantic ambiguity caused by synonym tampering. The variant characters are extended as synonyms， and more synonyms are replaced without causing semantic ambiguity， thereby improving the embedding capacity. By using chaotic mapping mechanism， the traditional MD5 algorithm is upgraded， and the chaotic AES algorithm is constructed to realize the encryption protection of hidden information， thereby increasing the difficulty of cracking. The research results show that compared with similar algorithms， the algorithm has good performance in semantic ambiguity， embedding capacity and security.

Due to the presence of the same object with different spectrum and the same spectrum with different objects in hyperspectral images， spectral information alone cannot fully reflect the features of hyperspectral images， and spatial information can be introduced to capture the features of objects more accurately. Therefore， a hyperspectral image classification method based on hybrid spectral enhancement and multi-scale spatial aggregation is proposed in this paper. In this method， a hybrid spectral enhancement module is designed， multi-scale local features of the spectrum are constructed using wavelet transform， and global features of the spectrum are generated by Transformer architecture， so as to enhance the intra-class consistency of the spectral features. At the same time， a multi-scale spatial aggregation module is designed to extract the inherent multi-scale information of spatial features and establish the interaction between different scales， so as to generate a more robust land cover representation， thereby further improve the classification performance. The experimental results show that the proposed method is superior to other advanced networks， indicating that the method can effectively obtain more abundant spectral information and spatial feature representation.

Aiming at the challenges of unnecessary leap and significant undulations terrains with large steering angles in path planning of multimodal quadruped robots by Rapidly Exploring Random Tree algorithm， a path planning algorithm solution based on discrete sampling is proposed. The path is preprocessed to remove unnecessary leap and a solution set is obtained by discrete sampling and dynamic programming method. B-spline curves are used to define spline segments and quadratic programming method is used to optimize the final path. The simulation results show that paths planned by the proposed method exhibit an average reduction of 31.4% in the adjustment of robot's center of mass height， a 13.4% decrease in undulation of terrain， an 11.4% reduction in terrain slope angle and a 62.7% reduction in steering angle， which affirm the effectiveness of the proposed method.

In order to improve the tracking accuracy and anti-interference ability of the fast steering mirror（FSM） in the space laser communication tracking and aiming system， a new integral fast non-singular terminal sliding mode controller was designed. By designing an integral fast non-singular terminal sliding surface， the performance of the sliding mode stage is improved， and a new exponential convergence rate is designed to improve the dynamic quality of the convergence stage. The dragonfly algorithm（DA） was used to optimize the controller parameters， in order to improve the dynamic performance of the system and ensure strong robustness， in view of the shortcomings of the traditional DA， the adjustment methods of the inertia factor， queue weight， and aggregation weight of the DA are improved， it also introduces a dimension-by-dimension evaluation mechanism and an improved position update mechanism of the Sine cosine algorithm to further improve the optimization capability of the DA. Finally， the effectiveness of the proposed method in the fast mirror control process was verified through simulation.

In the real-world positioning environment， observation stations are often set on mobile platforms， leading to measurement noise errors in the motion state information of the target to be measared as detected by the observation stations. These errors can affect the information received by the observation stations， thereby causing significant deviations in the estimation of the target source's position. To address this issue， a time-difference frequency-difference （TDOA/FDOA） multi-aircraft passive localization algorithm is proposed， which fusion the attention mechanism （AM）， convolutional neural network （CNN）， and bidirectional long short-term memory （BiLSTM） netwrok under observation station parameter errors. Meanwhile， which integrates an improved two-step weighted least squares method and a CNN-BiLSTM-Attention model. By correcting measurement values and improving the issue of the estimation performance degradation of the two-step method in the presence of observation noise， the accuracy of localization is improved. The results of the simulation comparisons show that the algorithm proposed in this paper demonstrates good performance in the presence of observation noise.

A self-propelled orchard mist spray was designed and developed to solve the problems of limited operation in low and dense orchards， difficulty in entering traditional large-scale spraying equipment， low operating efficiency and serious waste of liquid medicine. The machine adopts the form of pulse spray， improves the penetration of spray through high-speed air flow， increases canopy coverage， reduces drift， completes the design of the overall structure of the spray machine and key components such as the walking system， spray system， and fog dispersion device， and determines that the vertical spacing between the nozzles of two adjacent fog dispersion devices is 420 mm， and the included angle is 13°. The three-dimensional flow field model of spray system is established by using ANSYS Fluent software， and the Realizable k-ε And DPM particle model， the change of spray velocity， particle concentration and flow field distribution were simulated and analyzed. The field test was carried out on the operation process of the spray machine. Compared with the simulation results， the relative error of spray speed was 5.75%， the relative error of spray distance was 7.52%， and the relative error of spray height was 6.84%. The research results can provide theoretical support for the structural optimization of orchard spray.

Aims to address the challenge of poor adhesion and easy detachment of existing suction cups on rough surfaces and small structural objects. Inspired by the structure of leech suckers， we designed a novel type of sucker， RSS （rough-small-sucker）， which has multiple folds and a bottom array of small pits. Modeling and simulation are employed to investigate the adhesion mechanism， while experiments explore the impact of fold quantity and pit size on vertical adhesion and horizontal shear resistance of the sucker on rough surfaces. We study the attachment mechanism of the sucker by modeling and simulation， and experimentally investigate the effects of different fold numbers and hole sizes on the vertical adhesion performance and horizontal shear resistance of the sucker on rough surfaces. Results show that the multi-folded structure significantly improves RSS's vertical adhesion （up to 87.9% improvement under -40 kPa vacuum）. Pit size has minimal effect on vertical adhesion. For various sandpaper grits， small pit sizes （0.5 mm and 1 mm） have minimal impact on horizontal shear resistance of double-folded and triple-folded suckers. However， with a hole size of 1.5 mm， there is a 40.5% increase in horizontal shear resistance. The horizontal shear resistance increases with pit size for 1 600-grit sandpaper. Pit size has no significant effect on horizontal shear resistance for single-folded suction cups.