Aiming at the problem that the road recognition method requires a large number of training sets or high computational power support is not conducive to the realization of ride sensation improvement. In this paper， an improved least squares estimation method is proposed， which does not require a training set and directly collects vehicle responses to identify road excitation and road grade changes. On the basis of the variable parameter model of road grade coefficient and vehicle speed， the sampling processing rules of road excitation data are discussed， and the real-time road roughness coefficient is obtained by decoupling the influence of driving speed. The simulation results show that the comprehensive estimation accuracy of the A-E road grade is above 97%， the response time to the sudden change of road surface grade is less than 0.15 s， and the following performance to the road surface input is good. The dynamic parameters of the real vehicle at different speeds of different road sections are collected for identification. The test results show that the accuracy of the estimated value under this working condition is 98.2%， which is consistent with the actual road surface grade obtained by the three-meter-foot detection method. The feasibility and accuracy of this vehicle speed decoupling road identification method are verified.

In order to solve the problem of insufficient gas diffusion in air-cooled heat dissipation system， this paper proposes a solution of installing diversion plates inside the expansion structure. The fluid dynamics software Fluent was used to build the simulation model of the expanded flow structure and analyze the internal flow field of the expanded flow structure. The conclusions are as follows： The local head loss of fluid in different planes of the diffuser structure is 3.24 m and 1.21 m respectively. In the simplified model， the heat dissipation efficiency of fluid with low speed and large area distribution is 8% higher than that of fluid with high speed and small area distribution. The orthogonal test is used to study the influence of the deflector parameters on the heat dissipation efficiency. The results show that for the heat dissipation efficiency of the system， the order of the degree of influence from large to small is： the shortest distance L between the deflector group and the inlet end， the radius R of the surface hole of the deflector， the inclination angle between the deflector and the flow direction \theta， the projection length h of the deflector in the flow direction， and the shortest distance d of the deflector in the direction perpendicular to the flow direction. The experimental results show that the cooling efficiency of the heat dissipation system is increased by 7.67%， 3.62%， 4.26% and 6.93% when the fan is driven by output voltage of 3， 4， 5， 6 V respectively to produce cold fluid.

In order to improve the consistency and stability of automobile transmission， a design method of automobile mechanical automatic transmission system based on torsional damping is proposed. The hardware structure of automatic mechanical transmission system is composed of main control chip， minimum system， input/output circuit， communication interface， motor drive circuit and torsional damper. The dynamic model of the vehicle in the starting state and the sliding state is established. On this basis， the vehicle speed control is realized through fuzzy control and Fuzzy-Bang Bang dual mode control， and the design of the automatic mechanical transmission system of the vehicle is completed. The experimental results show that the shift time of the proposed method is short and the stability of the shift process is high.

Aiming at the lack of research on the parallel row ordering problem（PROP） of logistics interaction points， the PROP considering logistics interaction point and corridor width was proposed by taking workshop layout as the research background. A mixed-integer programming model with the goal of minimizing logistics costs was constructed. The solutions of a small-scale example were verified by Lingo solver. An improved sparrow search algorithm（ISSA） was proposed by combining the problem characteristics. The good point set was introduced for population initialization to make the population more diverse. The number of vigilants were changed dynamically by proposed algorithm， and algorithm performance was improved by combining operations such as PMX crossover operator， continuous 2-opt operator， and insertion operator. Early termination rule was applied to reduce redundant iterations. By comparing the ISSA with sparrow search algorithm， simulated annealing algorithm and genetic algorithm to solved 25-49 cases， it showed that ISSA has certain advantages in solution quality and solution speed. Finally， the proposed ISSA was applied to a production workshop layout in PROP mode， and the workshop layout was optimized. The results showed that the improved layout reduces the logistics cost by 32.40%， indicating the effectiveness of the proposed model and algorithm.

Because the trajectory planning of the grasping manipulator is unreasonable， the movement distance is increased， and a large amount of energy is lost. Therefore， an algorithm for trajectory planning of the grasping manipulator considering energy consumption and load factors is proposed. The kinematics model of the grasping manipulator is established according to the kinematics principle. The space position and attitude of each link are solved through homogeneous transformation matrix. The time interval sequence and cubic non-uniform B-spline interpolation algorithm are used to constrain the angle， velocity and acceleration of each joint， so as to optimize the energy consumption and load of its motion trajectory. The linear regression analysis equation is used to generate a new population， find out the optimal particle of the population and replace it with any particle in the original population， and the optimal movement trajectory of the population was found by repeating the process. Experimental results show that the proposed algorithm has fast convergence speed and minimum energy consumption.

A diagnosis method of planetary gearbox laser sequence pulse based on high-speed photography technology was developed. The serial pulse laser was used with a high-speed camera to collect the serial pulse signal of the vibration laser of the gearbox， and the vibration signal was collected by multi-sensors to calculate the entropy of signal arrangement. To determine whether there are abnormal characteristics in the laser sequence pulse signal, if abnormalities are detected, the sequence pulse signal is converted into a non?separable linear relationship. Then, the radial basis function(RBF) kernel is utilized to calculate the pulse training values. Based on the level of these training values, the extent of faults in the planetary reducer's gears and bearings is assessed. The test results show that the research method is feasible and effective， and the characteristic points of probability density function are distributed in ［0.23，0.34］ by detecting the sequence pulse signals of different fault signals， which can accurately distinguish four different fault types.

Aiming at the mechanical property analysis problem of piezoelectric composite materials with microstructure-based moisture-electro-mechanical multi-physical-field coupling， the moisture-electro-mechanical coupling smoothed finite element method based on asymptotic homogenization was proposed. The theoretical basis of this method includes the basic equation of piezoelectric composite materials， the effective performance parameters predicted by asymptotic homogenization， and the moisture-electro-mechanical coupling effect of piezoelectric composite materials. The dynamic control equation of this method was deduced， and the structural dynamic problems of piezoelectric composite materials were solved by applying the Wilson-θ method. The effects of moisture variation on the structure natural frequency and dynamic response were studied. The results were compared with those of the finite element method to verify the correctness and validity of the method. Therefore， this method has a broad application prospect for analyzing the multi-field coupling mechanical properties of piezoelectric composite components.

In this paper， the influence of surface temperature on the strength of contact surface is investigated based on the influence of displacement on the quality of overmolded products by simulation software. In the process of research， the model is coupled with temperature and displacement， the loading mode is unchanged， and the temperature and loading displacement are changed. Through comparison， it is found that the simulation results of different displacements have similar characteristics under the same temperature difference. The decrease of temperature will accelerate the damage and failure of adhesive layer， and the value of failure load will increase； Increasing temperature will promote the initial damage of adhesive layer， but delay the failure of adhesive layer and reduce the value of failure load. The increase and decrease of temperature will cause the change of stress， and the change trend is gradually increasing from inside to outside. The peeling force is provided by displacement load and stress in the loading process.

The existing lane detection methods have the problem of low detection accuracy due to fuzzy details in a complex environment. Therefore， this paper proposes an accurate lane detection algorithm based on a double feature extraction network in a complex environment. Firstly， a double feature extraction network is built to obtain feature maps of different scales， extract more effective features， and improve the feature extraction ability of the model in complex environments. Besides， a cross-channel joint attention module is constructed to improve the attention of the model to lane details and suppress useless information. Finally， combined with the improved void space pyramid pooling module， the receptive field is enlarged to improve the utilization of context information of the model， to strengthen the detection ability. The experimental results show that the F₁-measure of the proposed algorithm on CULane dataset reaches 72.43%， which is 4.03% higher than that of the mainstream UFSD algorithm. When detecting lane lines in complex scenes， the detection effect of the proposed method is significantly improved， which has been proven to be able to meet the needs of practical applications.

In order to study the safety of pedestrians crossing the street during the second conflict between people and vehicles at signalized intersections， three signalized intersections in Kunming City， Yunnan Province were selected as the research objects， and the traffic conflict analysis software T-Analyst was used to obtain vehicle trajectory data and traffic conflict data. A multi-angle safety evaluation method was adopted， based on the DV framework to judge the initial situation of the person-vehicle conflict， the traffic conflict indicators PET_S and PET_H， the maximum rejection gap and acceptance gap， and the average crossing speed to analyze the safety of pedestrians crossing the street. The results show that compared with the secondary conflict， the initial conflict situation faced by pedestrians in the primary conflict is more dangerous， and the frequency of pedestrians choosing to cross the street at risk is also higher. There is no significant difference in the secondary human-vehicle conflict in the indicators such as vehicle crossing behavior， traffic conflict indicators PET_S and PET_H， maximum rejection gap， acceptance gap， and critical gap. However， the average crossing speed of the secondary conflict was significantly higher than that of the primary conflict， and the potential risk of pedestrian crossing was higher in the secondary conflict.

This paper using the Geographically Weighted Regression model to fit the relationship between built environment variables and subway passenger flow， and analyzes the influence of significant variables on subway passenger flow. Using the passenger flow data of 63 stations of Xi'an Metro Line 1， 2 and 3 for 5 working days， and ArcGIS to match the passenger flow with the traffic analysis zones. On the basis of the traditional least squares regression model， constructed the GWR model. Considering the impact of GDP per capita， land use mixing degree， parking lot density， intersection density， subway entrance and exit density， etc. on the passenger flow of subway entrances and exits. The following conclusions can be obtained： The GWR model can depict the spatial non-stationarity and influence scale of the interaction between subway commuter passenger flow and built environment variables， and its results are better than the traditional least squares method. Meanwhile， we found that four variables-GDP per capita， land use mixing degree， intersection density， and subway entrance and exit density-have significant effects on subway passenger flow. The attraction of land-use mixing degree to subway commuter traffic is much greater than the density of subway entrances and exits， and more obvious in traffic analysis zones with low degree of land use development and balance.

To improve the driving safety of the superhighway and verify the feasibility of multi-sensor fusion algorithm in the superhighway， starting from the roadside sensing unit in vehicle-road cooperation， a multi-sensor perception fusion system model was established by using PreScan. The advantages and disadvantages of different methods were analyzed by understanding the types and architectures of multi-sensor fusion methods based on the multi-sensor fusion method. The hybrid multi-sensor information data fusion was adopted as the overall architecture， and the back-end fusion was adopted as the way to process and analyze data. This method improved the detection accuracy of target distance and speed of sensors on superhighways. The maximum improvement of distance and speed of Camera can reached 7.2% and 36.8%， respectively. The research results demonstrate that the method improve the detection capability of roadside units， and verify the feasibility and effectiveness of the adaptive weighted fusion algorithm on the superhighway.

This study uses the risk-avoiding disengagement to represent the risk. This study first uses the LSTM method to identify the risk-avoiding disengagement of autonomous vehicles in the road-testing phase based on a large scale of realistic road-testing data， then use the Tobit model to analyze the impacts of factors on the risk-avoiding disengagement rate， and the risk analysis model of autonomous vehicle road testing was established. This method can avoid the accidents of autonomous vehicle road testing in advance， and promote the autonomous vehicle road testing in a safe and regulated way.

In order to study the seismic behavior of polypropylene fiber lithium slag concrete （PLiC） columns in saline soil environment， low-period cyclic load tests of six columns were conducted. On account of the test results， the four-polyline skeleton curve was obtained. Based on the research of many scholars， considering the role of additives， obtaining the calculation formula of correction coefficient of each characteristic point by regression analysis. Introducing stiffness degradation index μ and damage factor D to combine the effects of erosion time， using the hysteretic rule applicable to the attenuation of bearing capacity and stiffness degradation of PLiC column to establish the force-displacement restoring force model of PLiC column under saline soil environment. The results showed that polypropylene fiber and lithium slag could improve durability and the mechanical properties such as ultimate bearing capacity， ductility of concrete columns. The skeleton curve and hysteresis curve obtained from the established restoring force model are in good agreement with the test results， which can provide a reference for the nonlinear analysis of seismic performance of polypropylene fiber lithium slag concrete columns in saline soil environment.

In the long-term service of steel-concrete composite beams， due to the carbonization of concrete and the intrusion of chloride ions， the passive film on the surface of the stud connectors is destroyed， and the studs will rust. The corrosion expansion of the stud connector will cause the cracking of the concrete slab， which will seriously affect the durability and safety of the steel-concrete composite structure. In order to study the cracking behavior of concrete slab in steel-concrete composite beams caused by the corrosion expansion of stud connector in corrosive environment， the plastic damage model of concrete slabs were established by using ABAQUS finite element analysis software in this paper. The uniform radial displacement of the stud hole is used to simulate the corrosion expansion of the stud， and the cohesion zone model based on the traction-separation law is used to simulate the concrete cracking. Considering 3 different constraint boundaries， 4 different stud spacing and 4 different stud diameters， the corrosion expansion stress distribution and cracking behavior of concrete slabs were investigated. The research results show that， with the increase of boundary constraints， concrete slab will be cracked due to the larger corrosion depth of the studs. The transverse spacing is increased from 50 cm to 80 cm， and the corrosion depth required for concrete edge cracking is increased to 1.59 times. When the diameter of stud increases from 16 mm to 25 mm， the corrosion depth required for concrete edge cracking decreases to 0.87 times.

To prevent and reduce the moisture damage of asphalt pavement， based on the surface energy theory and the monthly average temperature of major cities in six regions of China， two evaluation methods， including surface energy test and macroscopic moisture stability test， were combined to accurately evaluate the water stability of asphalt mixture under multi-temperature conditions. First， the surface energy of raw materials was tested in the temperature range of -20 to 50 ℃ to determine the value of adhesive evaluation index ER（Energy Ratio） at various temperatures. Four types of asphalt mixtures were selected for five freeze-thaw cycles splitting tests at -20 ℃， immersion Marshall tests at 10 ℃ and 30 ℃， and immersion Hamburg rut tests at 50 ℃. According to the test indexes determined by the above three macroscopic tests and their corresponding technical requirements， the threshold values of ER corresponding to each temperature were determined. The experimental results show that the threshold values of ER of the asphalt mixture without modifiers at -20， 10， 30， 50 ℃ are 0.992， 1.693， 0.994 and 0.512， respectively. The threshold values of ER of modified asphalt mixture are 1.562， 1.944， 1.144 and 0.512， respectively. Furthermore， based on the threshold values of asphalt mixture water stability evaluation index under multi-temperature conditions， the relationship model between threshold value of asphalt mixture without modifiers and temperature was established as y=-0.000 6x²+0.009 7x+1.472 6， and the relationship model between threshold value of modified asphalt mixture and temperature is y=-0.000 6x²+0.000 4x+1.834 1， in which the x indicates temperature and y indicates the corresponding threshold value.

In order to investigate the rheological properties of the modified asphalt compounded with reactive elastomer terpolymer （RET）， polyphosphoric acid （PPA） and SiO₂ aerogel （SG）. Firstly， the three modifier doping amounts were determined by using the three index tests； the high and low temperature properties and permanent deformation resistance of the modified asphalt were analyzed by DSR， BBR and MSCR tests. Fourier transform infrared spectroscopy， fluorescence microscopy and scanning electron microscopy were used to study the modification mechanism and microscopic properties of asphalt. The results showed that the incorporation of RET， PPA and SG improved the high and low temperature performance of the asphalt. Compared with the as-built asphalt， the complex shear modulus of the composite modified asphalt increased by 170%； creep stiffness decreased by 35 MPa at -12 °C and creep rate increased by 21.3%. The aerogel and the two polymers were mainly physically blended， and the modification process generated new absorption peaks. The aerogel improved the compatibility of the asphalt interface， and the modifier particles were uniformly dispersed， which improved the asphalt road performance.

Cracks in pavement layer of asphalt mixture bridge deck will further cause looseness， pit and other diseases under water-load action. In order to reveal the influence of pore water pressure on the expansion of existing cracks under load， this paper established a finite element model of pavement layer of asphalt concrete continuous box girder bridge deck， and verified the correctness of the finite element model by field dynamic load test， taking dynamic deflection and modal of bridge as indexes. Then， longitudinal cracks are preset at the bottom of the upper layer of the deck paving layer， and the propagation effects of existing cracks on the deck paving layer under water-load coupling are studied by analyzing the variation of stress intensity factor K_I and K_II at the crack tip. It is found that the change of stress intensity factor （K_I） reflects that the action of water pressure can promote the expansion of tensile cracks in pavement layer. The presence of water leads to the increase of K_I value by about 50%， and the larger the hydrodynamic pressure is， the faster the expansion rate is. The hydrodynamic pressure has little effect on the expansion of shear cracks， and the stress intensity factor （K_II） has little change. However， due to the presence of water， the response time of stress intensity factor becomes longer， so the water will have a continuous impact on the bridge deck paving layer.

To analyze the variation law and influencing factors of surface displacement during the construction of curved shield tunnel. The mechanical model of curved shield tunnel excavation considering active articulated mechanism is established， and the calculation methods of surface displacement caused by additional stress and ground loss based on mirror method and orifice expansion theory are deduced respectively. Based on the actual project， the corresponding monitoring test and numerical simulation are carried out， the sensitivity index affecting the surface displacement offset of curved shield was researched. The results show that： The surface displacement of the curved section is offset to the inside of the tunnel in the direction of tunnel excavation， and the maximum displacement value and offset value of the curved section obtained by the calculation method are basically consistent with the measured values. It is found that the sensitivity of the imbalance coefficient is the highest， while the turning radius and excavation radius are sensitive in the curve over-digging gap factor. It is recommended that the thrust difference and over-excavation range of the inner and outer cylinders should be reasonably controlled in the construction of the curve section.

In order to improve the accuracy， generalization and robustness of short-term energy consumption prediction for public buildings， a short-term energy consumption prediction model for public buildings based on short-term memory network was proposed. The long-term and short-term memory network is used as the energy consumption feature extractor of public buildings to retain valuable historical energy consumption data in the process of continuous iteration， adjust the output of different time sequences through autonomous learning and self-organization， and introduce the gray system to reduce the number of sample data required and reduce errors. The output weight value is calculated by the minimum multiplication method to obtain the prediction value under the long-term and short-term memory network. The short-term prediction value of building energy consumption is obtained by accumulating the results after the inverse normalization function processing. The experimental results show that the proposed method has excellent energy consumption prediction ability and can be effectively used for public building energy consumption prediction.

The primary objective of the Chang'e-7 mission is to detect water ice in the permanently shadowed regions of the Moon's south pole. This study involves conducting simulation research within an Earth-based environment， preparing test samples of polar hydrous simulated lunar soil， and assessing the mechanical properties of these simulants under ultra-low temperatures. Additionally， it analyzes how various factors—including moisture content， compactness， temperature， loading rate， substrate composition， and compressive strength—interact. The results indicate that the compressive strength of the frozen lunar soil simulants increases with higher moisture content， greater compactness， and lower temperatures， while it is less influenced by variations in loading rate. The compressive strength is affected by substrate composition in the following order：100% plagioclase <90% plagioclase +10% basaltic <70% plagioclase +30% basaltic <100% basaltic lunar soil simulant. Furthermore， this study employs grey correlation analysis to explore the relationships among fundamental physical property indices such as moisture content， compactness， temperature， loading rate， substrate ratio， and compressive strength. The analysis reveals the following order of correlation strength： moisture content>compactness>temperature>substrate ratio>loading rate. This paper not only examines the preparation and mechanical properties of aqueous lunar soil but also proposes a more systematic and standardized implementation scheme. This serves as a foundational basis for understanding the characteristics of polar aqueous lunar soil， providing valuable theoretical insights and references for subsequent lunar soil simulation research..

Most of the existing sentiment analysis methods based on commodity comments texts seldom consider the aspect features， and the related analysis models can′t consider both the long-term context dependence and the local text features， thus affecting the accuracy of sentiment analysis. To solve the problem， a sentiment analysis method based on bidirectional gated recurrent network （BiGRU） and capsule network is proposed. Firstly， the method uses the approach based on the word frequency statistics to extract the aspect features from the comment texts， and integrate them into the word vector representation to effectively improve the expression ability of the word vector. Then， the BiGRU network is used to extract the long-term context features of the texts， and Capsule Network is used to extract local features of the comment texts， thus realizing high-precision text sentiment analysis based on aspect. The experimental results on real datasets show that the proposed method is superior to bidirectional long-short term memory （BiLSTM）， CNN-LSTM， TextCNN and other sentiment classification models in Accuracy， Precision， Recall and F₁ score.

Aiming at the defects of traditional density-based uncertain clustering algorithm， such as parameter sensitivity and poor clustering results for complex manifold uncertain data sets， a new uncertain data density peak clustering algorithm based on JS divergence （UDPC-JS） was proposed. The algorithm first uses the uncertain natural neighborhood density factor defined by uncertain natural neighbors to remove noise points； secondly， the local density of uncertain data objects is calculated by combining uncertain natural neighbors and JS divergence. Then， the initial clustering center of uncertain data sets is found by combining the idea of representative points， and the distance based on JS divergence and graph is defined between the initial clustering centers. Then， the local density calculated based on uncertain natural neighbors and JS divergence and the newly defined distance based on JS divergence and graph between the initial clustering centers are used to construct the decision graph on the initial clustering center， and the final clustering center is selected according to the decision graph. Finally， the unassigned uncertain data objects are assigned to the cluster where their initial clustering centers are located. The experimental results show that the algorithm has better clustering effect and accuracy than the comparison algorithm and has greater advantages in dealing with uncertain data sets of complex manifolds.

In recent years， Vision Transformer （ViT） has shown amazing potential in areas such as image classification， target detection， and image generation. However， the performance improvement of ViT relies on the number of parameters in the network， leading to limited application scenarios. Inspired by neurology， this paper innovatively proposes to apply the recurrent structure between neurons in the human brain to ViT. This paper explains for the first time how the recurrent structure works from the perspective of Riemannian geometry， then presents the recurrent Vision Transformer model based on Token-to-Token Transformer architecture. Experimental results show that the introduction of the recurrent structure can substantially improve the performance of ViT with essentially no change in the number of parameters： we apply the recurrent structure in the Transformer networks， and on the Imagenet classification dataset， the network increase the parameters by only 0.14%， but bring a 9% improvement in classification accuracy. In the target detection task， increasing the parameters by 0.1% brings a 10.7% performance improvement.

To ensure smooth link transmission and stable network operation， a random backoff algorithm for multi link transmission in embedded wireless sensor networks was proposed. According to actual needs， use different frequency channels or orthogonal channels to transfer data， the cluster head rate vector was adjusted according to the principles and attributes of utility function design， and the backoff window size consistent was made. By recording the backoff times of embedded wireless sensor network link transmission， clarifying the busy status of the network， increasing the transmission success rate gain， reducing link transmission competition， and achieving network security and stability through adaptive random backoff correction， the dynamic adjustment of the initial competition window was carried out. The results show that the proposed algorithm has a high system throughput， can reduce network congestion and occurrence of link data collisions.

The image super-resolution reconstruction methods based on convolutional neural network mostly assume that a low resolution （LR） image is obtained by Bicubic downsampling of high resolution （HR） image. However， LR images in real environment contain unknown noise， which inevitably leads to poor network performance. To solve this common problem， A noise-resistant multistep image super resolution network is proposed. First of all， combine information distilling image denoising network and generative adversarial network to train the denoising network， in order to improve image denoising ability of the network； Secondly， the pure feature map in the middle layer of the network and the denoised image are combined with the stepwise image super-resolution reconstruction network， which is combined with the stepwise network training， to reconstruct low-resolution image of the real environment effectively. The proposed network is trained and evaluated on BSD100* and BSD100^# datasets with gaussian noise. Experimental results show that the proposed network achieves good improvement in image quality evaluation and visual comparison compared with existing advanced networks.

The imprint characters on the surface of the workpiece are uneven， rusty， and weathered， which the traditional character recognition methods hard to achieve satisfactory results. This paper regards the characters recognition task as a particular detection problem and designs a two-stage recognition network according to its characteristics： location and classification network. The location newtork uses the anchor-free method to extract the region of interest of characters， which effectively solves the problem of character region extraction. The classification network uses the Feature Decoupled Convolution Block and the Structural Re-parameterization technology， which can significantly improve the classification accuracy without any extra parameter. The transferring learning is used to solve the small sample problem and imbalance problem in the training stage. The experimental results on the self-built bolt dataset and the SynthText dataset show that the algorithm can achieve overall accuracies of 98% and 92%， respectively， which is superior to the compared algorithms.

The traditional unmanned aerial vehicle （UAV） multispectral image matching algorithm has some problems， such as low matching accuracy， low matching accuracy， and poor overall effect. In order to solve the problems in traditional methods， a UAV multispectral image matching algorithm based on affine invariant operator is proposed. Firstly， the UAV multispectral image is enhanced by histogram equalization method， and the affine invariant region of the UAV multispectral image is extracted by affine invariant operator； Secondly， corner detection and feature point extraction of UAV multispectral image are completed by multi-scale product LoG method； Finally， the shape description operator is used to describe the image feature points， and the Gaussian weighted distance algorithm is used to complete the final image matching. The experimental results show that the proposed method has higher accuracy， smaller error and better overall effect for UAV multispectral image matching.

In order to improve the scheduling effect of the continuous production line and improve the processing efficiency of the production line， a deep reinforcement learning optimization scheduling algorithm for the continuous production line is proposed. Combining Monte Carlo algorithm and Bayesian evaluation method to reduce the data complexity of the continuous production line problem； A deep neural network model is used to optimize the pipeline scheduling parameters， evaluate and code them； The iterative greedy algorithm is combined with the deep reinforcement learning method to solve the scheduling data problem and realize the continuous production line scheduling. The experimental results show that the optimal comprehensive evaluation results of the scheduling results of the proposed algorithm are higher than 0.9531， and the process delay is optimized to less than 5 min， which improves the processing efficiency of the production line.

The distributed trajectory tracking problem for multiple UAVs under unknown external disturbance is investigated. A new nonlinear robust trajectory tracking strategy for multiple UAVs is proposed. The dynamic model for the UAVs' formation is illustrated in the Tangent frame. For the trajectory tracking problem， a robust formation tracking control strategy based on robust integral of the signum of error （RISE） is designed to compensate for the effects of unknown external disturbances， which improves the robustness of the UAVs' formation system. Based on the Lyapunov stability analysis， it is proved that the global asymptotic convergence of the coordination errors and the semi-global asymptotic convergence of the UAVs' position tracking errors are achieved. The proposed formation control strategy is validated via real-time experiments that are performed on the self-build UAVs' formation flight control testbed. Flights without wind disturbance and flights with disturbance are all performed. The performance comparison experiment with the conventional sliding mode control algorithm is performed. The experimental results show that the designed control strategy can achieve good coordinated trajectory tracking of multiple UAVs， and has better control performance compared with normal sliding mode control law.

A method of group vehicle recognition and tracking based on roadside millimetre-wave radar was proposed to improve roadway traffic detection accuracy. Based on pre-processed detection data of millimetre-wave radar on multi-lane traffic flow in an urban arterial road， a Gaussian kernel-distance based spatial clustering algorithm with noise density （DBSCAN） was proposed to conduct spatio-temporal clustering of effective radar signals reflected by group vehicles. Then， a fusion algorithm of unscented Kalman filter （UKF） and linear Gaussian mixture probability hypothesis density （GMPHD） was proposed to improve tracking accuracy of group vehicles which move nonlinearly on the road. The algorithms were tested in simulation and onsite environment. Simulation results verified that the UK-GMPHD algorithm can accurately and stably track nonlinear moving vehicles. Results of onsite test showed that the kernel-distance based DBSCAN algorithm can solve the problem of classical algorithm effectively that the parameter tuning of feature vector was difficult to adjust. The UK-GMPHD algorithm reduced the root mean square error of target tracking in term of target distance， velocity and angle by 21.03%， 23.41% and 20.67% in comparison with GMPHD algorithm.

In order to comprehensively improve the accuracy of real-time tracking results for mixed moving targets， a real-time tracking method for mixed moving targets based on GPS-UWB combined positioning technology is proposed. Establish a three-dimensional surface information reconstruction model for mixed motion target images， and use multidimensional pixel reconstruction methods to perform information fusion and motion feature point calibration for mixed motion targets， analyzing the differential feature vectors of different target points. At the same time， the GPS-UWB combined positioning technology is introduced to optimize the real-time positioning of mixed moving target points， ultimately achieving real-time tracking of mixed moving targets. The experimental test results show that the proposed method can obtain real-time tracking results of mixed moving targets with higher stability and smaller errors.