Confidential computing is an effective method for addressing data security and privacy issues. However， with the increasing diversification and heterogeneity of computing devices， the implementation of traditional confidential computing on diverse heterogeneous devices becomes extremely complex and challenging. How to achieve heterogeneous confidential computing has become a current research hotspot. Based on the implementation technology of confidential computing on heterogeneous devices， a comprehensive analysis and evaluation of heterogeneous confidential computing were conducted. Firstly， the concept and challenges of heterogeneous confidential computing are introduced. Secondly， the software boundary expansion techniques and hybrid boundary expansion techniques in mainstream heterogeneous confidential computing technologies are discussed. The article also summarizes the evaluation analysis methods and standards in current heterogeneous confidential computing research. Finally， we points out the future research directions in the field of heterogeneous confidential computing.

There are some disadvantages in the traditional methodology of the mix proportion design of ultra-high performance concrete， such as high cost， low efficiency， and complex process. The application of artificial intelligence technology can help to overcome these shortcomings and intelligently predict various properties， thus intelligent and sustainable mix proportion designs can be delivered. By the review on the research progress of artificial intelligence technology applied in the properties prediction and mix proportion design of ultra-high-performance concrete， obstacles and challenges in current mainstream technologies were pointed out， including data quality， model validation， model interpretability， and multi-objective optimization. In order to resolve these issues， many specific proposals were put forward based on the combination of artificial intelligence technology and discipline theory of building materials.

To clarify the research progress of strengthening technology of recycled concrete aggregate， the strengthening methods of recycled concrete aggregate and its improvement effect were reviewed systematically. The effect of different strengthening treatments on the performance of recycled aggregate concrete was evaluated. The problems existing in the application of strengthening methods were summarized， and constructive suggestions were put forward to provide a scientific basis for researchers to design and develop green high-performance recycled aggregate concrete.

In order to further study the contact pressure distribution characteristics of non-pneumatic tires， a reverse analysis of TWEEL was carried out， the finite element model of which was established and verified by tests. Based on finite element analysis， the influence of the main components of TWEEL— shear band and spokes on the uniform distribution characteristics of contact pressure was analyzed. The results show that the great equivalent axial stiffness and small shear stiffness of the shear band can make the shear deformation more easily， which is the key point of design for TWEEL to achieve uniform distribution of contact pressure. The small radial compressive stiffness of the spokes allows TWEEL to be a top-loader and have a better bearing capacity， thus can reduce the influence of the load transferred by the spokes in the contact patch on the contact pressure distribution. The large tensile stiffness of the spokes is conducive to increase the contact area， and ensures that the TWEEL has a uniform and low contact pressure distribution. The results of the study provide an important reference for the structural design and optimization of non-pneumatic tires.

In response to the "range anxiety" issue of battery electric vehicles caused by cabin heating demands in cold winter climate conditions， the sensitivity relationship between vehicle speed and the energy consumption of a heat pump air conditioning system was explored. A new paradigm of a control-oriented high-precision prediction model for the heat pump air conditioning system was developed. Then，based on this model， the vehicle speed preview information provided in the connected environment was integrated into the design of the thermal management controller， creating a predictive eco-heating strategy （EHS） based on model predictive control （MPC）. The proposed EHS was validated using a high-fidelity physical simulation model based on Dymola. The simulation results show that by utilizing vehicle speed preview information to optimize the operation load of the air conditioning system， while satisfying the thermal comfort requirements， it further achieves a 18.46%-20.61% improvement in energy efficiency compared to the baseline controller.

Firstly， Spearman correlation analysis was used to design the characteristic parameters based on face-to-volume ratio， face-to-face ratio， side clearance and back clearance volumes， firepower shore volume， cylinder gasket dead zone volume， clearance volume， and the proportion of each volume， and investigate their effects on hydrocarbon emissions under cold start conditions. Then， the effect was explained from the viewpoint of combustion chamber geometry and combustion process by using combustion 3D simulation. The results show that the clearance volume and clearance volume proportion have the strongest correlation with cold start hydrocarbon emissions， and the other characteristic parameters have very little correlation with cold start hydrocarbon emissions. Meanwhile， Spearman correlation coefficient r_s and one-tailed test P-value of clearance volume proportion are 0.753 8 and <0.01 respectively， which mean the correlation is very significant. The pre-combustion chamber jet ignition engine in cold start condition is consistent with the conclusion of the above correlation analysis， but in theoretical air fuel ratio hot engine condition， the hydrocarbon emission of jet ignition is significantly lower than that of spark ignition because the high temperature jet from the pre-combustion chamber enters the combustion chamber clearance earlier， ignites the mixture in it， and consumes the fuel near the wall early in the combustion， and in lean combustion， it does not have higher hydrocarbon emission although it has lower combustion temperature. This also explains and confirms the mechanism of the effect of clearance volume and clearance volume proportion on hydrocarbon emission from another perspective. The research results can provide a theoretical basis and support for the design of gasoline engine combustion system structure.

In order to improve the flexibility of the mechanical gripper， a pressure mechanical gripper adsorption force balance control method based on the white shark optimization algorithm was proposed. The structure of the pressure mechanical gripper was divided， the axial stress and deformation of the internal cavity of the gripper was calculated， and the mechanism of action between the mechanical gripper and the object was clarified. Based on the adsorption and release states of the mechanical gripper， fully consider external influences such as vibration and airflow， the positive pressure between the adsorption target and the suction head was adjusted， and an adsorption force balance control model was output. Introducing the white shark optimization algorithm， the optimal parameters of the model were solved through initialization， position update， and optimal positioning to achieve the expected adsorption force equilibrium control goal. The experimental analysis results confirm that the proposed method's adsorption force curve is basically consistent with the optimal adsorption force curve at different angles of the mechanical gripper， controlled between 3~5 N， and stable control can be achieved within 3 seconds. The duty cycle is still higher than 0.2 at a pressure of 0.7 Pa. This method effectively improves the smoothness and accuracy of the gripper object adsorption， achieving high-quality adaptive adjustment of the gripper adsorption force.

Aiming at the problems that the existing ramp recognition algorithms have poor adaptability to the conditions and are unable to satisfy the application requirements of mass production vehicles， an electric vehicle slope recognition method based on vehicle mass estimation was proposed. Firstly， the vehicle longitudinal dynamics model was established， and the signal characteristics of the acceleration sensor under the actual vehicle conditions were analyzed. The least square vehicle mass estimation strategy with forgetting factor was constructed to obtain the vehicle mass directly under the starting condition. Ramp recognition algorithms were designed for static parking and dynamic driving scenarios. In the static scenario， the filter latch strategy was used to deal with the interference factors such as activities in the vehicle. In the dynamic scenario， the Kalman filter algorithm based on measurement noise adaptation was designed to realize the fusion estimation of dynamic observation and kinematic observation for the slope. The effectiveness of the method was verified by Simulink-CarSim joint simulation. Finally， the real vehicle test was completed on Chery new energy's mass production electric vehicle platform and domain controller. The road test results show that the mass estimation error is less than ±10 kg； the static estimation error of the slope is less than 0.001 rad， and the dynamic error is within 0.005 rad. The estimation accuracy and stability are greatly improved， which ensures the environmental adaptability of the intelligent electric vehicles.

A novel in-vehicle network intrusion detection method based on controllerarea network （CAN） bus data was proposed and named as IncepNet method. First， a real Car-hacking dataset was selected for data preprocessing. Image processing methods were used to reorganize the time-series data in this dataset according to certain rules， and converted it into image data suitable for use as the input of convolutional neural network. Next， the existing residual network （Inception-ResNet） was optimized and followed by the addition of a long short-term memory （LSTM） layer with multi-batch normalization and a Dropout layer. Finally， a confusion matrix including recognition rate， accuracy， F₁-score （F₁-score）， and number of false alarms （FAR） was used to demonstrate the superior accuracy and reliability of the model. The results show that the model has low false alarm rate， high detection accuracy and high detection rate， and its efficiency is significantly better than previous detection methods based on other machine learning.

High-temperature tensile tests were used to establish the material mechanics parameters of CL60 wheel steel in the 26~500 ℃ temperature range， considering the convective heat transfer， wheel-rail contact heat transfer， thermal radiation， driving resistance， taking advantage of work-energy conversion relationships with Hertz contact theory to obtain the heat flow load， normal load， and tangential load. A wheel finite element model was developed in Abaqus to predict the fatigue life of wheel treads during the braking process of C80 trains with lengthy downhill cycles. The subroutine is created to manage the load with heat transfer qualities for rotational loading of the tread surface in order to determine the temperature of the tread surface and the Mises stress-time history. It also examines the thermal-mechanical interaction of the braking process for the lengthy downhill cycle of the train. By investigating the effect of gate tile pressure on fatigue life， the fatigue life procedures were developed based on the N.E. damage and R.W. damage formulas to evaluate the fatigue life of the tread surface under two types of damage. The results show that after the same cycle， the tread temperature and Mises stress history stabilize. The tread hazard point is the wheel-rail contact center， with the two different types of damage， the percentage inaccuracy of wheel service revolutions is 1.28%. The higher the pressure of the brake shoe， the higher the temperature of the danger point， the higher the Mises stress， and the shorter the fatigue life.

In order to understand the effect of ice crystal frost heave on permeable asphalt mixture in cold region， the water-heat-force coupling damage characteristics of permeable asphalt mixture were studied in this paper. Based on the analysis of the initial temperature field and initial water field of typical permeable asphalt mixture， the displacement field， stress field and strain field under the dynamic expansion mode of ice crystal were studied， and the distribution characteristics of the occurrence time of in-situ dynamic expansion and separation dynamic expansion， frost heave stress and deformation in the horizontal and vertical directions were obtained， and the basic change law of the dynamic expansion effect of ice crystal in the permeable asphalt mixture was clarified. The research results provide technical support for the design of permeable asphalt mixture in cold area.

In order to accurately predict the effect of precipitation and ice condensation on the engineering performance of asphalt mixtures during the service of environmentally friendly salt storage asphalt pavements， the long-term road performance changes of asphalt mixture were studied based on immersion freeze-thaw cycle tests. Based on the theory of life reliability and damage principle， considering the influence of moisture and freeze-thaw cycle on salt storage asphalt mixture， the damage evolution model of asphalt mixture under immersion-freeze-thaw condition was established， and the evolution of engineering performance damage of ordinary asphalt mixture and environmentlly-friendly salt storage asphalt mixture was compared. The experimental results show that with the increase of the number of immersion-freeze-thaw cycles， the porosity of the salt storage asphalt mixture increases obviously， while the performance indexes of splitting strength and dynamic stability decrease obviously， and the variation range of the salt storage asphalt mixture is always greater than that of the ordinary asphalt mixture. According to the model calculation and analysis， under the damage of immersion-freeze-thaw， with the precipitation of snow melting salt， the destruction of environmently-friendly salt storage asphalt mixture is developed from the outside to the inside， and the immersion-freeze-thaw cycle has a more significant effect on the splitting tensile strength of the asphalt mixture， which will affect the pavement performance to a certain extent.

The surface texturing technology is combined with laser cladding to solve the problems of cracks， porosity and other defects that are prone to occur in the laser cladding layer through the establishment of multi-field coupled numerical simulation model. The preset texture can smooth the temperature gradient， reduce the residual stress， decrease the melt pool flow rate， effectively avoid the double eddy current effect， and improve the surface quality and grain size uniformity of the cladding layer. The experimental results verify the validity of the numerical simulation， and expand new ideas for optimizing the laser cladding parameters， improving the quality of the cladding layer， and enhancing the performance of the cladding layer.

To promote the application of austenitic stainless steel in structural systems， it is necessary to clarify the mechanical properties of austenitic stainless steel materials. By material types and loading systems as variables， a total of 33 bar-shaped specimens were subjected to testing， yielding stress-strain curves， stress-time curves， and skeleton curves. The influence of mechanical properties such as yield ratio， hysteretic energy dissipation and percentage elongation after fracture was thoroughly examined. The parameter values of the Johnson-Cook model （J-C model） were obtained by fitting， and the finite element model of austenitic stainless steel specimens was established using ABAQUS software. The numerical simulation results showed good agreement with the experimental results， which confirmed both the accuracy of the finite element modeling method and the applicability of the J-C model. The test results show that the stress-strain curves of austenitic stainless steel specimens exhibit including elastic phase， strengthening phase， and local deformation phase， but no obvious yield platform is observed. The yield ratio of 316 austenitic stainless steel specimens under cyclic loading reached 0.857， while the hysteretic energy dissipation and percentage elongation after fracture decreased significantly， and its plastic deformation ability gradually weakened. The mechanical properties such as yield ratio， hysteretic energy dissipation and percentage elongation after fracture of 303 and 304 austenitic stainless steel specimens under monotonic and cyclic loading are relatively similar， and both of them have good plastic deformation ability.

In order to assess the overtaking risk level of intelligent networked vehicles when different networked information was provided. To make up for the neglect of driver factors in traditional risk assessment and the insufficient evaluation capability of a single traffic conflict indicator for complex traffic scenarios， the block maxima （BM） and peak over threshold （POT） methods were introduced to fit the extreme value distributions for the two types of conflict scenarios （follow-me and frontal oncoming conflict） involved in overtaking events， so as to assess the risk of following and frontal collision accidents， respectively. In each conflict scenario， a bivariate extreme value model was proposed to integrate different traffic conflict indicators and a non-stationary extreme value model was proposed to take the driver into account for road safety estimation， and the models were validated with the intelligent and connected vehicle overtaking test data. Extracted the overtaking event from the original test data and calculated the conflict indicators： including the time to collision between the ego vehicle and a preceding vehicle GAP， the time to collision between the ego vehicle and an oncoming vehicle TTC_t₁， the deceleration DRAC， the time to collision TTC， headway with the preceding vehicle THW. The degree of collision risk was characterized by the event probability that the time conflict index is negative or the DRAC is greater than the MADR. The results show that the error results of the binary extremum model constructed by different conflict indicators are different in the head-on collisions， and the binary extremum model constructed by THW&DRAC has the most accurate evaluation results （standard error MAE=0.000 28）. The binary extreme value model constructed by TTC&DRAC is the most accurate （MAE=0.006） in the frontal oncoming collisions. In different conflict scenarios， the non-stationary extreme value model considering the driver factor significantly improves the risk assessment accuracy （the AIC and BIC values are small） compared with the model that does not consider the driver factor. In addition， different intelligent network information （real-time distance， overtaking advice， speed advice） brings different passing maneuvers risks， and when the intelligent network information is speed advice， the overtaking risk of the car is the smallest. Therefore， the non-stationary extreme value model considering the driver factor and binary extreme value model proposed can effectively evaluate the driving risk through the traffic conflict index. Secondly， the experimental data of intelligent and connected vehicles show that the proposed model can accurately assess the overtaking risk level of intelligent and connected vehicles when they provide different intelligent network information.

An improved Faster-RCNN algorithm for detecting small traffic signs was proposed， which addresses the issues of poor recognition performance of distant small targets and high computation cost in real-world traffic scenes affected by weather and lighting conditions. Based on the basic architecture of Faster-RCNN， the algorithm reconstructs the backbone network and improves the region proposal network to make the network framework lightweight. A multi-scale feature fusion network is designed by integrating the scSE attention and GSConv modules， and the Anchors box size was updated to improve the localization and recognition of traffic sign targets. The ROI Align pooling operation with bilinear interpolation for each target subregion was used to preserve the detailed features of the target region and improve the ability to capture details of distant targets. The balanced L₁ loss function was adopted to address the problem of imbalance between samples with large gradient difficulty and those with small gradient easiness， thus improving the training effect. Experiments were conducted on the expanded TT100K dataset. Results show that compared with traditional Faster-RCNN， the model weight is reduced by 200 MB， and detection accuracy is improved by 21.3%. The algorithm achieves a detection accuracy of 85% in low-intensity environments such as cloudy days， which helps improve the traffic sign detection performance in extreme environments.

In order to solve the problem of low accuracy of accidents severity prediction caused by data imbalance， a traffic accidents severity prediction model based on deep learning technology was proposed. Machine learning algorithm was used to determine the key variables affecting the severity of accidents， Numerical accidents variables were converted into image data and applied to MobileViT network that combined the convolutional neural network and the self-attention mechanism. Focal loss function was used to adaptively adjust the loss contribution of injury and severe accidents with small data volume， so that the model paid more attention to unbalanced data， and the prediction performance of the model was evaluated by precision， recall and F₁ score. The results show that the overall predictive performance index of the proposed model is above 0.81， which is better than other baseline models， and recall and F1 score are at least increased by 4% and 2.5%. Compared with WGAN-GP-XGBoost and ResNet18 models， MobileViT model has improved recall and F1 score of injury accidents by 25.9% and 4.5% respectively. Compared with the other two models， the prediction performance for severe accidents is the best， with an increase of 8.9%， 4.2%， and 6.7% in precision， recall rate， and F1 score. Compared to other data balancing methods， the MobileViT model enhanced with focal loss as the loss function demonstrates superior performance in imbalanced data prediction.

To predict the possibility of traffic accidents in advance， a traffic accident risk anticipation model gated recurrent unit spatial-temporal transformer（GST） was established based on the combination of vision transformer（ViT）， gated recurrent unit （GRU）， and MLP-Mixer. By modeling spatial-temporal relational learning through ViT， the frame features of predicted targets were enhanced to improve their distinguishability. On this basis， GRU was used to extract temporal relational， and then GRU and MLP-Mixer were combined to enhance the hidden layer frame features， establishing and optimizing spatiotemporal relational， the confidence score of traffic accidents for each time step was predicted based on the corresponding feature frames to predict the probability of future accidents and effectively distinguish between dangerous driving and accident driving behavior. Finally， the proposed model was validated on the public datasets DAD and A3D， and the results showed that the recognition accuracy of the proposed model was superior to other advanced algorithms. The AP on the two datasets reached 59.9% and 94.6%， respectively， demonstrating good predictive performance and generalization ability. In the DAD dataset， the algorithm proposed was compared to the DSTA model. With similar AP， the proposed algorithm can advance the prediction time of accidents by 2.38 seconds， an increase of about 13%. This indicates a significant advantage and provides assistance for road hazard warnings and safe driving.

Aiming at the limitations of single-dimensional car-following model to describe vehicle car-following behavior in local multi-vehicle environment， the mechanism of vehicles in adjacent lane influencing the subject vehicle car-following behavior is explored， and a vehicle car-following model more suitable for local multi-vehicle environment is attempted to be established. The driving behavior variable reflecting the influence of vehicles in adjacent lane was determined by correlation analysis. Vortisch indicator of similarity（VIS） was used to characterize the influence of vehicles in adjacent lane on car-following behavior. Chi-square independent test and kernel density curve were used to determine the VIS demarcation threshold reflecting whether the influence is significant. Recursive feature elimination was used to screen the variables related to car-following samples significantly affected by vehicles in adjacent lane. The influence mechanism of variables was determined according to the statistical analysis results. Based on the mechanism proposed， the car-following model suitable for local multi-vehicle environment was constructed and its prediction effect was evaluated. Results show that VIS between the speed of subject vehicle and following vehicle in adjacent lane can characterize the influence and the VIS threshold is 0.668. It is concluded that the attention mechanism and memory effect can explainthe influence mechanism of car-following behavior in multi-vehicle environment. Considering the attention mechanism and memoryeffect， RMSE decrease by 65% in full velocity difference model （FVD） model and 62% in intelligent driver model （IDM） model， MAE decrease by 65% and 59% and R² increased by 180% and 288% respectively， which proved the rationality of the attention mechanism and memory effect in explaining the car following behavior of subject vehicle in local multi-vehicle environment.

In order to study the energy evolution law and failure behavior of high-strength concrete under conventional triaxial compression state， five groups of tests of C60 and C70 high-strength concrete under different confining pressures were carried out. Based on the test results and the principle of thermodynamic energy conservation， the variation laws of input energy density， elastic strain energy density and dissipation energy density of high-strength concrete with axial strain and confining pressure were obtained. According to the linear growth relationship between elastic strain energy density corresponding to peak stress of high-strength concrete and confining pressure， the failure criterion of high-strength concrete was established. The results have shown that：① The input energy density and dissipated energy density of high-strength concrete increased with the axial strain， while the elastic strain energy density increased with the axial strain in the pre peak stage and decreased after the peak；② The input energy density and dissipated energy density corresponding to the peak stress of high-strength concrete sample increased with confining pressure， and the input energy density and dissipated energy density corresponding to the peak stress of C70 high-strength concrete were greater than that of C60 high-strength concrete under the same confining pressure；③ The failure criterion of high strength concrete based on elastic strain energy density has high accuracy， few parameters and clear physical meaning. The form of the criterion was similar to Hoek-Brown failure criterion， but it has wider applicability；④ The failure criterion of high-strength concrete was a symmetrical hexagon with equal edges and unequal angles in the π plane. The singular points of the failure curve were rounded according to the "turning angle into arc" method. The research results provide a new idea for studying the deformation and failure behavior of concrete materials from the perspective of energy.

In order to improve the realism and naturalness of spoken facial videos， and more accurately reflect the speaker's speech content and facial expressions， an optimization algorithm for generating speaking facial videos based on dense convolutional generative adversarial networks and keyframes was proposed. Bounding box regression was used to correct the candidate window of the face， comparing facial features based on keyframes， and speaking facial information was obtained. Search for the optimal set of hyperparameters through the validation set， random feature vector with a standard normal distribution was established， the internal parameters of the discriminator were calculated， and a speaking face video was generated. Using the objective function of the discriminative network， correcting prior information， repair results that match reality were selected， and the generation of speech facial videos was optimized. Experimental results have shown that the detection recall of the proposed algorithm is higher than 96%， with SSIM， PNSR， and LMD metrics of 0.99， 34.7 dB， and 2.2 pixel， respectively. The video authenticity is 74.1%， indicating good video generation performance.

AV1 compression technology increases encoding complexity. In order to meet the requirements of real-time communication （RTC） applications， encoders need to run at low latency， low bit rate， and low encoding time on low-end platforms， which makes using AV1 encoders challenging in RTC tasks. In this regard， this article proposes an AV1 fast intra frame encoding algorithm. By pre calculating the directional characteristics of blocks， during the intra frame encoding direction prediction traversal， significantly different directions from the pre calculated direction are skipped， thereby improving the encoding speed of the libaom-AV1 encoder， saving intra frame encoding time， and accelerating the progress of AV1 in low latency video encoding applications. The experimental results show that the algorithm can save 25.89% of running time， and after saving encoding time， the loss of encoding performance is relatively small.

Aiming at the complex semantic rules and potential topic structures in different fields， which result in poor scalability and portability of topic sentence generation， low similarity between Chinese text topic sentences， and high redundancy in topic sentence extraction from Chinese text， a Chinese text topic sentence automatic extraction method based on TextRank algorithm and similarity was proposed. Using the Bi LSTM model for Chinese text segmentation， continuous Chinese text is segmented into independent words. Through mutual information method， Chinese text feature selection is carried out， and feature values are calculated to extract the most relevant and representative features of the task （such as keywords and clue words）. Keywords and clue words are used as important clues and basis for topic sentence extraction. Based on the TextRank algorithm and similarity， various weights and weight coefficients were considered to automatically extract Chinese text topic sentences. The experimental results show that the proposed method has low redundancy in automatic extraction of Chinese text topic sentences， and the completeness， scalability， and portability of the document are all in a good state. Moreover， the results of ROUGE-1， ROUGE-2， and ROUGE-L are all high. Ensure the automatic extraction effect of Chinese text topic sentences， with a high degree of application.

In order to effectively ensure the security and privacy of sensitive attribute data republication， a privacy protection method for sensitive attribute data republication based on k-prototype clustering algorithm and relative entropy is proposed. Based on the type attributes of sensitive attribute data， fast clustering is used to calculate the distance between each data type attribute， and k-prototype clustering algorithm is used to cluster sensitive attribute data. Using relative entropy to calculate the sensitivity of sensitive attribute data after clustering， a weighted multi-dimensional bucket group is constructed based on the sensitivity to unfold sensitive attribute data and republish it anonymously， ultimately achieving the goal of privacy protection. The experimental results show that the proposed method can effectively reduce the concealment rate， information loss， and information disclosure of private data， improve the degree of privacy protection for sensitive attribute data republishing， and ensure the security of data.

To address the issue of large parameter quantities in high-performance image super-resolution networks， a lightweight model was proposed. First， three kinds of large kernel attention were integrated， namely dual-path large kernel attention， large kernel pixel attention and large kernel channel attention， aiming to expand the model perceptual field and establish the long-range dependence of pixels. Second， an adaptive attention fusion mechanism was introduced to enhance the characterization of features and improve the model performance. Experiments demonstrate that the model performs well on visual perception and quantitative tests. On the Urban100 dataset， the mean value of peak signal-to-noise ratio of ×4 reconstruction results was improved by 0.25 dB compared with the currently popular ARRFN algorithm. The reconstructed images have more realistic visual effects， clearer and more natural texture information， which fully demonstrates the proposed algorithm effectiveness.

A low illumination image enhancement algorithm based on dual branch channel priors and Retinex is proposed to address the issues of local dimming， detail loss， and over enhancement in existing algorithms. Firstly， on the basis of Retinex， a dual branch bright dark prior feature guidance method is proposed， and a bright channel prior feature module and a dark channel prior feature module are designed to guide the network to suppress reflection component noise and improve lighting component brightness； Secondly， a pixel mixed attention mechanism module is designed to learn targeted features from three dimensions： channel， space， and pixel； Thirdly， a dark channel refractive index estimation module is designed to amplify image detail features. Finally， a mixed loss function is used to adjust brightness， contrast， noise， and color constraints. The experimental results on public datasets show that compared with 10 advanced algorithms， this algorithm improves image brightness while reducing color distortion and detail loss， achieving the best visual effects and quality indicators.

The existing visual object tracking methods only use the target area of the first frame as a template， which makes it easy to fail in rapidly changing and complex backgrounds. To address this issue， this article proposes a Transformer based object tracking algorithm that focuses on the target focus information within the template and dynamically updates the template features. In order to reduce the interference of background information on attention， this algorithm uses a sparse Transformer module to achieve feature information interaction； A template focus attention module was also proposed to focus on the target focus of dynamic template features and initial template features， in order to preserve highly reliable feature information in the initial template. The experimental results show that the success rate and accuracy of our algorithm in OTB100 benchmark testing reach 70.9% and 91.6%， respectively， which are 4.11% and 3.27% higher than similar template update algorithms STARK. This algorithm effectively addresses the limitations of existing visual object tracking methods and improves the accuracy and robustness of tracking.

For the cross-domain authentication problem of mobile users in mobile edge computing， we proposed a dynamic authentication framework consisting of three secure authentication models， which provides security and efficient authentication. Built on the authentication framework， a lightweight authentication protocol based on elliptic curve encryption， an edge-centric security authentication protocol， is designed. The protocol ensures the anonymity and untraceability of mobile users and their mutual authentication with edge servers. Security analysis and performance evaluation show that the proposed protocol has better performance advantages in both computational and communication cost and is more suitable for mobile scenarios.

To solve the problem of ignoring the interrelation between key points in the existing human pose estimation models， a new algorithm based on correlation matrix to establish implicit key point interconnection was proposed. The algorithm adopts a two-stage network structure. In the first stage， the existing model was used to obtain the accuracy of key points. In the second stage， the correlation coefficient matrix was calculated to build a key point corrector to achieve implicit key point interconnection. By selecting some key points with good effect as guidance， the other key points with poor effect were optimized. The pseudo-thermal map information was reorganized into two matrices， the guide region and the region to be indexed. The guide matrix was divided into three regions according to the accuracy rate， and was multiplied by the transposed matrix to be indexed to obtain the correlation matrix， and different influence factors were added to different correlation matrices. The total correlation matrix was obtained by formula， and the index region was obtained by multiplying the matrix to be indexed. Finally， the reverse information was reconstructed to the original dimension， and the pseudo-thermal map was added pixel by pixel to obtain the final predicted thermal map， which automatically establishes the relationship between key points and corrects the relationship between key points ignored by the existing model. The experimental results on COCO2017 data set show that the key point recognition accuracy of the human pose estimation correction algorithm is 70.5%， an average increase of 0.4% compared with the existing model SimpleBaseLine. In some parts that are easily obscured， such as ankles and wrists， they are small in size and at the end of limbs， and the probability of occlusion is high. The number of adjacent key points is less， and the surface information features are less. The human posture estimation corrector strengthens the connection between the key points at the extremities and other key points in the whole body， and establishes more feature connections.The experiments demonstrate that the recognition accuracy of ankles and wrists has improved by 1.5%， with a significant increase compared to other existing models， proving the effectiveness of the implicit keypoint-interconnected human pose estimation correction algorithm.

To address the problems of low programming efficiency， high development difficulty and high Vivado HLS resource utilization in the hardware development process， a pipelined direct parallel finite impulse response filter was designed and implemented with a 127-order Hamming window using Vivado and Vitis HLS platforms. And the filtering performance of HDL FIR IP， HLS FIR IP， and XILINX FIR IP in the Vivado Simulator environment was compared. Detailed analysis was conducted on the differences in resource utilization， timing， power consumption and execution time among different implementation methods. The experimental results show that under the same conditions， the resource utilization of HLS FIR IP is reduced by 1% compared with HDL FIR IP and XILINX FIR IP， the execution time is reduced by 24.5% and 808.2% respectively， and the code size is saved by 98.5%. Based on the proposed experimental method， a comparison was made with previous work to objectively analyze the efficiency differences of different development platforms and methods under certain conditions. The results show that our design method can significantly reduce the usage of logic units and storage resources and improve the development efficiency.

In view of the insufficient accuracy of classical time-frequency domain harmonic detection methods and the limited detection accuracy and noise immunity of weak harmonics in single chaotic systems， this paper first compares the Duffing system with the Van der pol system in terms of detection sensitivity， determines the critical driving force based on the bifurcation diagram， and judges the phase transition of the system by the phase diagram and Lyapunov index； Further utilizing the damping and restoring forces of the displacement coupling of the Duffing system and Van der pol system， a closed-loop feedback controlled coupling system is formed. By utilizing its high sensitivity to the amplitude of specific frequency periodic signals and noise resistance， the amplitude of each harmonic frequency component of the mixed harmonic signal is accurately detected； Based on the single coupling system， the phase response interval of the system is determined by analyzing the phase response relationship， and the coupling system array covering all phase response intervals is designed. Simulation experiments were conducted on harmonic signals with multiple integer harmonics and inter-harmonics mixed under harsh random noise backgrounds. The results show that the method proposed in this article can achieve accurate amplitude detection of various harmonic components in harsh noise environments， and eliminate the impact of phase on system detection performance.

Aiming at the problem of poor estimation performance of the compression-aware DOA estimation method based on compression in the context of Alpha stable distribution noise under the conditions of low signal-to-noise ratio and small number of fast beats， a DOA estimation algorithm based on approximate l₀ norm sparse reconstruction is proposed. The fractional low-order statistic combined with KR subspace method is used to reshape the fractional low-order moment matrix to suppress the Alpha stable distribution noises and construct the sparse measurement direction model. The smoothness and steepness of the function are analyzed by the exponential family distribution function， and a optimal smoothing function approximating the l₀ norm is constructed to solve the sparse measured direction model. For the algorithm error caused by the off-grid effect， the offset is introduced into the in-grid sparse vectorization model， and the first-order Taylor series expansion is performed on the guide vector matrix to establish the off-grid sparse vectorization model. The algorithm uses the alternating iteration method to calculate the signal components and offsets to obtain the off-grid target DOA estimates. The effectiveness and superiority of the proposed algorithm for direction angle estimation in the background of Alpha stable distribution noise is verified by simulation experiments.

For the compound jamming scenario where noise suppression jamming and interrupted sampling repeater jamming simultaneously exist， a method of staged suppression of them using distributed arrays was proposed. Firstly， non-uniform samples are selected using the generalized inner product. By iteratively applying this method in a staged selection process， the statistical feature estimates of noise suppression interference and slice-repeater interference were obtained respectively. Then the robust suppression of the compounded jamming was completed using the distributed array spatial domain adaptive processing. Finally，with typical configuration of parameters， the effectiveness and performance of the proposed method was validated and analyzed through numerical simulation experiments.

In order to solve the problems of large differences between hole size and theoretical size， as well as low hole stability caused by the uneven soil surface and soil entry and exit methods of the current hole-forming machine of rapeseed transplanting machine. a flat surface and hole-forming machine for rapeseed bowl seedling transplanting machine was designed based on the agronomic requirements of rapeseed bowl seedling planting and the external dimensions of the bowl seedling， including flat surface device and hole-forming machine. The motion trajectory of the hole forming machine was analyzed by using Matlab software， and the parameter equation of the motion trajectory was obtained in this paper， verifying the theory of "zero speed time period" of the hole-forming machine during the hole forming operation process. The field test was designed to verify the performance of the complete machine. It was found that the flatness of the chamber surface was 1.93 mm， the coefficient of variation of the hole taper was 2.7%， the qualified hole rate was 98.1%， and the hole formation frequency was 107 holes per min， when tractor forwarded speed at 1.62~1.81 km/h. All indicators met the agronomic requirements of rapeseed pot seedling transplantation. This study can provide a reference for the design and optimization of flat chamber hole-forming machine for rapeseed bowl seedling transplanting machine.