To actively control the development and evaporation processes of the wall-impinging sprays， the metal surfaces with different microstructures and wettability were prepared with various methods. The morphological development processes of diesel and n-butanol sprays after hitting the surfaces under different surface wettability， wall temperatures and ambient back pressures were investigated based on a high-speed photography platform. The results show that increasing the roughness of the oleophilic surfaces can enhance the oleophilicity of the surfaces. The oleophilicity of the surfaces can be significantly reduced with the method of surface free energy modification. As the oleophilicity of the surfaces decreases， the spreading radius and the entrainment height of the fuel sprays after hitting the surfaces increase. Compared with diesel droplets， n-butanol droplets have a stronger wetting ability to surfaces， resulting in a smaller spreading range of n-butanol sprays after hitting different surfaces. Compared to n-butanol， the surface temperature has less influence on the spreading radius of the diesel sprays. The spreading radii of the diesel sprays are higher than 40 mm at the later stage after hitting the walls with different temperatures. The increasing ambient back pressure has an inhibitory effect on the diffusion of the sprays.

In this paper，the feedforward-feedback hybrid FxLMS（Filtered-x Least Mean Square） algorithm is employed to control the active hydraulic mount with oscillating coil actuator for the narrowband and broadband vibration issues at the autobody end of the mount. Based on the preliminary work of mechanical model， mathematical model and all parameters identification of the active mount， MATLAB/Simulink is used to perform offline simulation of the active mount control system. The result shows that the hybrid algorithm can suppress vibrations at the autobody end more effectively compared to the feedforward FxLMS algorithm. A controller prototype is developed and full object bench experiment is carried out to test the vibration control performance of the hybrid algorithm. The result shows that the hybrid algorithm can reduce the overall vibrations at the autobody end by 11 dB to 23 dB. By combining the advantages of both feedforward and feedback FxLMS algorithm， the hybrid algorithm effectively reduces the adverse impacts of vibrations.

In order to study the excitation effect of gait displacement caused by rough terrain environment on passive walking robot， the sinusoidal function model is used to describe the concave and convex characteristics of the road surface， and the random variation of amplitude and frequency is introduced to simulate the unevenness in the real road environment. By using bifurcation diagram and Lyapunov exponent， the influence of road unevenness on the gait stability of robot is compared and analyzed， and the boundary crisis event in global bifurcation is deeply studied. In addition， a robot prototype is built and ADAMS walking simulation is performed to verify the walking capability of the model. It is shown that， on an uneven sinusoidal road surface， the robot's gait exhibits quasi-periodic motion and transitions into chaos with the change of system parameter， occurring via the torus multiplication bifurcation. The excitation effect of the robot gait caused by the rough terrain makes the robot deviate from the limit cycle trajectory， and this results in system degradation in terms of stability. In the walking environments above， the double boundary crisis event triggered by the unstable orbit generated by the saddle-node bifurcation is the main reason for the disappearance of the gait attractor.

To improve the classification accuracy of fault detection for wind-turbine blades， a machine-learning-based fault detection and classification method is proposed. First， a Ridge-regression-enhanced Brain Storm Optimization （R-BSO） feature-selection algorithm is developed to identify an optimal feature subset. The best feature combination extracted by R-BSO is then fed into a Stacking-based classifier to produce the final prediction， completing the RBVS blade-fault detection framework. Finally， a convolutional neural network equipped with a Convolutional Block Attention Module （CBAM）， termed CBCNN， is introduced for blade-fault classification. Experimental results demonstrate that the proposed algorithms achieve superior performance in both detection and classification of wind-turbine blade faults..

To address the current problems of structural redundancy and poor human-machine compatibility in lower-limb exoskeletons， a multi-degree-of-freedom， size-adjustable lower-limb power-assist exoskeleton robot was designed with mechanical limit devices installed at each joint. The force， deformation， stress and resonance of the exoskeleton structure during walking were simulated by the finite-element method. A plantar-pressure-sensor system was used to collect and analyze the force distribution and its variation on the human sole. Lagrange’s equations were employed to establish the dynamic model of the exoskeleton leg， and the loads at every joint were solved. The experimental and analytical results were then taken as parameters of the joint drive functions to perform a dynamic simulation of the exoskeleton， yielding spline data of the exoskeleton motion. The obtained results provide a basis for the subsequent drive control of the robot and simultaneously verify the human–machine coordination and rationality of the proposed lower-limb power-assist exoskeleton.

A proposed analytical process aims to reduce vibration and noise by adjusting the clearance of the taper bearings for the systematic issue of whine in the drive axle. This process is based on a simulation model of the dynamics of the mid-axle assembly， which includes the gear， axle， bearings， and housing. The accuracy of the model is verified by the transmission error test of the bevel gears. By considering the clearances of five taper bearings that support the input shaft， pinion shaft， and differential as design variables， the response surface method was employed to establish a surrogate model for the vibration response of the axle case and the fatigue life of the pinion shaft ball bearings. Through multi-objective optimization， an improvement plan for addressing the drive axle whistling issue was obtained， considering the bearing life. The results show that this process yields better outcomes and demonstrates high adaptability.

Aiming at the low computational efficiency and absence of safety guarantees in existing centralized cooperative control schemes， a swarm-coordination algorithm based on reinforcement learning is first proposed. Single-agent proximal policy optimization is extended to multi-agent interactive environments so that complex cooperation in multi-agent systems can be addressed. Secondly， the cooperative control of vehicles at unsignalized intersections is formulated as a multi-agent reinforcement-learning problem， and a safety-augmented centralized cooperative control method—behavior-restricted proximal policy optimization—is developed. Formal safety verification and behavior restrictions are integrated into the swarm-coordination algorithm， whereby the policy is guided to be iteratively optimized in a safe manner and unsafe driving behaviors are avoided， so that traffic safety in unknown scenarios can be further guaranteed. Finally， simulation experiments are conducted with the Carla platform. It is shown that the incorporation of behavior restrictions causes an 8.06 % loss in traffic efficiency， yet a 100 % safety improvement is achieved. Compared with the representative model predictive control approach， the proposed method reduces the computation time to 1/326 of the original， increases traffic efficiency by 67.0 %， lowers the collision rate from 63.5 % to 0， and improves ride comfort by 26.5 %.

In this paper， fast hot-pressing sintering process assisted by an electric field was employed to fabricate Ni-doped 3 mol% yttria-stabilized zirconia（3YSZ） solid-state electrolytes. The effects of Ni doping content ranging from 3% to 8% on the phase composition， microstructure， densification， and ionic conductivity of the sintered products were investigated. The research findings indicate that when the Ni doping content is 5%， under process parameters of constant pressure at 20 MPa， heating rate 60 ℃/min to 1 200 ℃， hold for 10 min， the Ni/3YSZ grains are tightly packed with a relative density of up to 98%. At a test temperature of 300 ℃， the ionic conductivity reaches 6.24 μS/cm. By utilizing the multiphysics coupling software COMSOL， a three-dimensional multi-channel cross-flow steady-state single cell model was established， taking the multi-channel planar anode-supported solid oxide fuel cell（SOFC） as the research subject. The electrical conductivity of the prepared 3YSZ/Ni solid-state electrolytes with different Ni contents was evaluated， and the power density achieved could reach 22 W/m2. This provides a new approach to enhancing the low-temperature performance of SOFC electrolytes.

Aiming at the problems of low data efficiency and poor scene adaptability of current reinforcement-learning methods in autonomous-driving applications， an environment-representation-based reinforcement-learning strategy for self-driving is proposed. First， a driving-environment representation model is devised： multi-head attention， convolutional neural networks and long short-term memory networks are combined to extract spatio-temporal features from consecutive visual inputs， while a variational auto-encoder is employed to reduce the dimensionality of bird’s-eye-view inputs. Second， measurement information is fused to form a comprehensive representation of the driving environment. Finally， the representation model is integrated with several classical reinforcement-learning algorithms and evaluated in CARLA simulation. Results show that the proposed representation model markedly improves the learning efficiency of driving policies， accomplishes diverse dynamic and static driving tasks， and enhances both the accuracy of agent decisions and adaptability to different scenarios.

To facilitate safe and efficient interactions between Autonomous Vehicles（AVs） and pedestrians， this study employs the Multi-Agent Deep Deterministic Policy Gradient（MADDPG） algorithm to establish a pedestrian-vehicle interaction model in a mixed traffic context that includes both autonomous and human-driving vehicles. This model formulates interaction strategies enabling AVs to avert accidents without the necessity of direct inter-vehicle communication. In comparison with several benchmark algorithms， the proposed algorithm demonstrates substantial improvements in terms of training efficacy， collision frequency reduction， and traffic capacity. Additionally， the robustness of the proposed model is assessed across varied risk scenarios. Findings reveal that as the intensity of pedestrian behavioral randomness， or behavioral noise rises， the duration of interaction delays of both vehicle categories increases. Remarkably， the collision rate of AVs initially increases before declining， indicating an adaptive learning phase. Under conditions of elevated noise， AVs exhibit a superior capability for collision avoidance compared to human-driving vehicles， highlighting their enhanced resilience in chaotic urban traffic conditions. These outcomes underscore the potential of MADDPG-based frameworks to significantly contribute to safer， more efficient AV integration in mixed traffic scenarios.

In order to quantitatively assess how the design of sound barriers affects the driver's behavior and experience， the study proposes various acoustic barrier design schemes for the Sichuan Tianfu New Area-Qionglai Expressway， with different levels of complexity and style. To analyze the driving behavior of subjects under different design schemes， 25 participants were recruited to conduct driving simulation tests in five different scenarios. The study selected representative vehicle operation indexes and physiological and psychological indexes of the subjects. A comprehensive evaluation model was constructed based on the combination of game theory and the Topsis evaluation model. The results of the study are summarized in the following table. The research results indicate that the Topsis comprehensive evaluation model has a significant impact on the driving behavior， physiology， and psychology of the driver. Specifically， the continuity of the sound barrier pattern and the degree of information loading are important indicators. To improve driving stability and safety， it is recommended to reduce the degree of information loading and improve the degree of pattern continuity. These changes can effectively reduce the lateral acceleration standard deviation and speed standard deviation. Additionally， the driver's driving comfort is also improved.

In order to investigate the road network cascading failures mechanism after the introduction of connected automated vehicles （CAV）， a road network cascade failure model considering the critical density of a new type of road with mixed traffic flow is developed. The method considers the way of traffic redistribution after failures and captures the effect of mixed traffic flow on the road's ability to resist congestion. In this paper， the urban road network of Xi'an City is taken as an example to study the regular characteristics of the cascading failures of mixed traffic flow. The results show that when the CAV penetration rate reaches 0.6（critical value）， the outermost nodes start to resist failures， and the propagation of cascading failures slows down significantly； after the penetration rate exceeds 0.6， the outermost nodes successfully resist failures， and the total size of failures decreases by about 91%. Failure propagation is fastest and largest when attacking the largest node of the traffic flow， and controlling the largest node of the traffic flow is the focus of cascade failure.

To improve the efficiency and safety of lane-changing merging of connected autonomous Vehicles， the game interaction process of vehicle lane-changing behavior is portrayed. The German ExiD high-precision trajectory dataset is used to deeply analyze the dynamic interactive game characteristics of vehicle lane changing and merging， and the lane-changing cut-out behavior of mainline vehicles in trajectory data is analyzed and defined from the perspectives of game decision-making and cost. When facing the ramp vehicles with clear intention to merge， part of the mainline vehicles choose to accelerate to the inside to change lanes to cut out and at the same time to provide a gap for the ramp vehicles to merge to reduce the cost of driving efficiency. When the vehicle speed is higher， the mainline vehicle tends to change lane and cut out to reduce the loss. The vehicle game cut-out and merge model based on trajectory data can portray the vehicle game decision-making process and effectively shorten the distance of changing lane and merging， the average reduction of the distance of changing lane and merging is 11.51 m， and the average improvement of vehicle collision time is 6.77 s.

In order to ensure that the structure has sufficient bearing capacity under horizontal forces such as wind loads and earthquakes， the shear performance of cold-formed thin-walled steel composite walls with different cross-sections is studied. Determine the basic parameters of materials for preparing cold-formed thin-walled steel composite walls， prepare three types of cold-formed thin-walled steel composite wall test samples： rectangular section， U-shaped section， and Z-section. Test the displacement and crack changes of each test sample under the loaded shear load， monitor the ultimate load and failure load of the test sample， verify the degradation changes of the bearing capacity of each test sample， and analyze the influence of screw spacing， steel thickness， and spacing on the shear resistance of the sample. The experimental results show that compared to the other two types of sections， the Z-shaped section cold-formed thin-walled steel composite wall has the smallest displacement change under load and shorter crack occurrence. When the screw spacing is 150 mm， the steel thickness is 50 mm， and the spacing is 60 mm， the shear performance of the composite wall is optimal.

This study builds on existing evaluation criteria by introducing additional indicators， including cone penetration ratio， compression elastic recovery rate， and low-temperature tensile strength， which are assessed through short-term and long-term aging tests. An evaluation model tailored to the performance of sealants in seasonally frozen asphalt pavements is developed using the entropy weight-TOPSIS method. The experimental results indicate the following： After short-term aging， the control standards for cone penetration ratio and compression elastic recovery rate are 90% and 30%， respectively， whereas after long-term aging， these standards decrease to 50% and 20%. The evaluation criteria for sealant viscosity are 1-3 Pa·s at 190 °C， with a grouting depth of 5 cm. The low-temperature tensile stress evaluation criteria for sealants after immersion are 0.06 MPa at -30 °C and 0.03 MPa at -20 °C， while the adhesion strength after freeze-thaw cycles is 0.03 MPa at -30 °C and 0.025 MPa at -20 °C. The entropy weight-TOPSIS method not only determines whether the sample meets qualification standards but also provides a comprehensive assessment of its quality across multiple indicators. These findings contribute to improving the application efficiency and durability evaluation of sealant materials， offering supplementary data for performance assessment and specification updates in cold regions.

In this paper， a novel type of super-long-span continuous rigid-frame bridge was presented， ultra-high-performance steel-shelled concrete（UHP-SSC） continuous rigid-frame bridge with box girder， with steel shell filled with ultra-high-performance-concrete（UHPC） and perfobond leiste stiffeners（PBL） as fundamental components unit. UHP-SSC box girder employed UHPC to bear pressure， and steel shells bear tension， which may be used as the UHPC pouring non-removable formwork， allowing the light self-weight and high strength properties of steel and UHPC to be completely exploited. Conceptual design of the 400-meter-long continuous rigid-frame bridge with UHP-SSC box girder was studied， the results indicated that the thickness of top plate， bottom plate， and web plate of the UHP-SSC box girder can be greatly lowered. The suitable side-to-mid span ratio new continuous righid-frame bridge is 0.5~0.65， and the reasonable beam height ratio of the pier and midspan is 2.59~3.59， and the height of box girder at the pier accounts for 1/26~1/32 of the main span. The strength and stiffness of the new box girder meet the mechanical requirements of the code for fully prestressed concrete structure. The superstructure weight per square meter of UHP-SSC continuous rigid-frame bridge is reduced by approximately 132.8%， 76.8%， 324.4% and 338.3% respectively， when compared to mixed beam continuous rigid-frame bridge， arch bridge， cable-stayed bridge， and suspension bridge with similar span， and the carbon emissions are also reduced by about 28.5%， 57.1%， 14.1% and 55.7%. The superstructure cost of the new UHP-SSC continuous rigid frame bridge is increased by about 3.1% and 1.3% respectively， when compared to the mixed beam continuous rigid-frame bridge and arch bridge， and decreased by about 26% and 46.1% respectively， when compared to the cable-stayed bridge and suspension bridge. The new UHP-SSC continuous rigid-frame bridge has evident structural light self-weight properties， better economic and environmental performance， and has strong competitiveness in long-span beam bridges， as well as excellent technical popularization and application possibilities.

To address the problem of life assessment for thin-walled cracked steel circular tube under tension-torsion composite fatigue loading， this paper conducts a study on the fatigue crack propagation mechanism. Firstly， based on the maximum circumferential stress criterion and the Paris equation， a analysis model for crack propagation of pre-cracked thin-walled steel circular tube subjected to tension-torsion biaxial proportional in-phase cyclic loading is established. Secondly， based on existing fatigue tests， the interactive techniques of finite element software ANSYS and FRANC3D is jointly used to simulate the crack propagation behavior of steel circular tube， validate this paper's proposed composite crack propagation analysis model. Finally， based on the finite element model， parameter analyses are conducted on factors such as shear-to-tensile stress ratio， diameter-to-thickness ratio， and initial damage size. The results demonstrate that this paper's proposed analysis model accurately predicts crack propagation behavior of thin-walled steel circular tube （with diameter-to-thickness ratios exceeding 10） under tension-torsion composite fatigue loading. Under tension-torsion in-phase proportional fatigue loading conditions， the composite cracks in thin-walled steel tubes quickly evolve into opening-mode （Mode I） cracks， with the crack propagation angle determined by the maximum nominal shear-to-tensile stress ratio. The shear-to-tension stress ratio and initial damage length were identified as the primary factors on the composite crack propagation behavior of the steel circular tubes under small-scale yield condition with the identical maximum principal tensile stress and stress ratio.

In order to obtain the shear deformation law and failure characteristics of precast utility tunnel joints under different foundations， this paper comprehensively considers the influence of the structural characteristics of utility tunnel joints and the foundation， and reveals the influence of foundation conditions on the shear performance of utility tunnel joints through model tests. The results show that the socket joints of the precast utility tunnel with connectors have good ductility and can meet the requirements of use when the utility tunnel is subjected to shear load under gravelly sand， silty clay and clay foundations. When the utility tunnel bears the same load， with the increase of the foundation stiffness， the dislocation of the utility tunnel will decrease， but the damage range of the joint will expand and the damage will be more serious， the final joint damage area accounts for about 15 % of the whole utility tunnel. The bearing end of the joint is more seriously damaged than the socket end， and the cracks in the chamfer position， near the bolt hole and at the top and the bottom are the most concentrated. It is recommended to strengthen the above positions locally. With the increase of dislocation， the longitudinal connector has gradually become the main force component of the joint to resist shear deformation， which plays an important role in improving the shear capacity of the joint and the integrity of the joint.

For patients， the traditional means of physical rehabilitation training have some problems such as high repeatability， lack of individuation and invisible training data. This paper provides the design， implementation and effect evaluation of a physical rehabilitation training evaluation system based on virtual reality technology. The system is designed on the principles of improved patient engagement， personalized rehabilitation， real-time feedback and monitoring， safety and data analysis. This paper conducted experiments on 10 patients. Through virtual simulation environment， patients can participate in more attractive and interactive rehabilitation training， and the system can dynamically adjust rehabilitation programs according to patients' conditions. In addition， the system can also collect a large amount of rehabilitation data， providing detailed information for medical staff to support the optimization and personalized adjustment of the treatment process.

The high degree of freedom of human limbs often constitutes complex poses in which the key points are prone to occluded， and locating the occluded key points is one of the difficulties in human pose estimation. To this end， this paper proposed a method with a guided graph structure and enhanced key points location information. The method incorporates a location information enhancement module in the HRNet， which can improve the representation of the spatial location information of visible key points. A visual graph neural module is integrated into backbone network to extract relevant features containing key points and exploit the local and global topological connectivity relationships between key points in pixel coordinate space to infer the location information of the occluded key points. Finally， a heatmap aggregation unit and a semantic graph convolutional network are employed to update the affinity weights between key points in the semantic space， which can represent the topological dependencies between key points under the constraints of the skeleton structure and further optimize the estimation of the occluded key points. The proposed model achieves an average accuracy of 78.1% on the COCO2017 test set， and can accurately estimate the occluded key points prone to occlusion in complex poses.

Aiming at the limitations of some current defogging algorithms caused by using foggy and non-foggy image pairs and the cost consumption caused by supervised learning， this paper proposes an image defogging algorithm based on comparative learning and recurrent consistent generative adversarial network. By training recurrent generative adversarial network with unpaired foggy and clear images， the value of image defogging algorithm in real scenes is improved， and the domain shift problem of defogging algorithm is alleviated； meanwhile， we design the contrast-guided branch to learn the potential feature distribution of the image， implicitly constrain the embedding of different samples in the depth feature space， deeply mine the similar features of foggy and clear images， pull the similar characteristics of the images closer together， retain the mutual information between the two types of images， maintain the consistency of image content， and improve the performance of network defogging； introduce the frequency loss， constrain the output of the generator， reduce the loss of information in the frequency domain， further retain the content and structural information of the image， reduce the blurring and distortion of the defogged image， and improve the quality and clarity of the generated image. Experimental results show that the model proposed in this paper is an effective image defogging algorithm with improved information entropy and average gradient and richer detail information compared to the current mainstream deep learning-based and traditional defogging algorithms.

In the face of complex weather conditions such as cloudy days， rainy days， and nights， vehicle target detection is affected by factors such as lighting， rain， snow， and dust. As a result， problems like false detections and missed detections occur. To address these issues， a YOLOv10-vehicle target detection algorithm is proposed. Firstly， a new attention mechanism module named WT-PSA is designed to improve the model's attention to vehicle targets under complex weather. Secondly， the SPPF module is improved by introducing the average pooling operation to address the problem of insufficient feature information extraction caused by the max pooling operation. Then， an improved C2f-OD module is put forward to enhance the ability of the backbone network to extract image feature information. Finally， the model's loss function is replaced with Focal EIoU to accelerate the convergence speed and reduce the loss value. Comparative experiments are conducted on the vehicle dataset UA-DETRAC. The mean average precision （mAP@0.5） of the improved algorithm is increased by 5.1% compared with that of the original algorithm， demonstrating the superiority of the YOLOv10-vehicle algorithm in vehicle detection under complex and severe weather conditions. Meanwhile， experiments are also carried out on the VOC public dataset. The detection accuracy of the YOLOv10-vehicle algorithm in detecting vehicle targets is improved by 2.8%， which verifies the generalization ability of the improved algorithm in this paper.

To address the limitation that existing learning-based image denoising algorithms struggle to preserve edges and textures， we propose an adaptive edge-aware denoising model built upon a multi-directional gradient network that can capture distinct image information separately. First， multi-directional gradient operators are applied to the clean target image to generate noise-free gradient maps， which then guide the network in learning gradient representations free from corruption. Second， an adaptive gradient-fusion module is introduced to fuse gradient cues with the noisy image adaptively， increasing the network’s attention to edge and texture details. Experimental results demonstrate that the proposed model achieves competitive PSNR and SSIM values. Moreover， the denoised images consistently exhibit superior visual quality， underscoring the model’s potential for practical image-denoising applications.

In the real image， when the target object is occluded by other objects or environmental elements， the noise near the occluded area will confuse the contour features of the background and the target object. It is particularly difficult to infer the contour of the occluded area with continuity and smoothness by analyzing only the contour features of the visible part. This results in a deviation between the extracted foreground target contour and the original target boundary. Therefore， a full-contour extraction algorithm for local occluding object in real image is proposed under the influence of background noise. The background model of the real image is constructed by the mixed Gaussian background difference method， which is used to distinguish the background image and the foreground target image in the real image， remove the background information， and obtain the foreground target image. Multi-resolution method is combined with ACM and GVF field is introduced to extract the contour of the foreground target image. The double-arc interpolation algorithm is used to obtain the full contour of the foreground target continuously and smoothly through smooth contour repair and corner contour repair， aiming at the blocked area in the foreground target contour. The repaired contour is visually consistent with the original contour， and the difference between the boundary and the original target is minimized. The experimental results show that the misjudgment rate for non-background elements is always below 0.1%； The contour curves generated in the process of contour extraction are highly consistent with the target boundary， and the contour lines are smooth and continuous. The contour restoration results are extremely natural and realistic， with an error value of only 0.02%.

To address the key issue in federated learning the resource heterogeneity and data heterogeneity of each client， this paper proposes a resource-efficient client selection method for clustered collaborative federated learning. Firstly， where each client is grouped according to its computational power， the average accuracy of each group of clients is used as an indirect metric to select clients in the same group in each round of training. Secondly， the clients are clustered according to the model similarity of each client within each group， and the clients in different clusters within each group are selected. Evaluate the performance of the method proposed in this paper on real datasets， the experimental results show that this method can reduce the global training time， obtain faster and smoother convergence， and achieve a good balance between training efficiency and global model accuracy.

In the rapid recognition of visually significant targets in images， the presence of a large amount of noise in the image can interfere with the detection of significant targets due to factors such as changes in lighting conditions and background complexity， resulting in poor recognition robustness. To this end， this article introduces machine vision technology and uses Fourier transform filtering technology to enhance the original image， improve its robustness to factors such as lighting changes and background complexity， enhance its resistance to noise， and improve the robustness of target recognition. According to the Fourier transform filtering technique in machine vision， the original image is unfolded and processed to generate a gradient map， completing the enhancement of the original image. By using a linear model of multiple adjacent pixels to calculate the trend of slope difference， the optimal threshold is determined based on the measured values of slope difference distribution. The morphological iterative erosion method is introduced to effectively distinguish the target area from the noise area， achieving high-definition segmentation of the image. Adopting a multi-scale analysis strategy to divide the image into multiple superpixel regions of varying numbers， calculating the color mean of pixels within each superpixel， and achieving abstract representation of the image. Based on the characteristics of salient features， the mean saliency of superpixels at various scales is calculated and fused to obtain the visual salient object recognition results of the image. The results show that the CM of the proposed algorithm can reach 0.597 7， UM can reach 0.913 2， and the target recognition recall rate can reach 99% under different types of noise. The proposed algorithm has good consistency， indicating that the proposed method can effectively improve the robustness of recognition results.

Due to the small size and low contrast of subcortical brain structures（such as the striatum）in medical images， their segmentation is challenging， making their application in automated medical diagnosis difficult，this article proposes a medical image segmentation network based on deep learning methods to segment the three parts of the striatum， namely the globus pallidus， caudate nucleus， and putamen， in magnetic resonance imaging. The network model proposed in this article has the ability to capture global and local features and establish the correlation between global and local information， and effectively fuses deep semantic features and shallow detail features at different scales without degrading the depth， achieving accurate segmentation of the striatum. The model is validated on publicly available brain datasets and compared with other state-of-the-art methods. The model achieved dice similarity coefficient， average intersection ratio， and 95% Hausdorff distance are 94.26%， 90.94%， and 3.82 respectively， which are better than several other methods and have reached the advanced level. This shows that the model proposed in this article can improve the segmentation accuracy of the striatum and provide a basis for research on related diseases.

To address the issues of diverse land cover sizes and spatial combinations， as well as significant interclass similarity and intraclass variability in remote sensing image classification tasks， a lightweight frequency and spatial feature fused multi-scale remote sensing scene classification network（FS-LMFFNet） is proposed， based on the purpose of effective feature extraction and full integration of multi-scale features. Firstly， to combine the advantages of CNN and Transformer， and achieve an adequate extraction of local and global features， a Frequency and Spatial MLP module（FS-MLP） is proposed， which complements traditional spatial operations in extracting global high-frequency texture features by introducing frequency domain analysis. Secondly， to resolve the multi-scale characteristics of remote sensing scene images， a Lightweight Multi-layer Feature Fusion（LMFF） module is proposed， in which lightweight convolutional blocks are employed to efficiently fuse the multi-scale features in the first three stages. Finally， FS-LMFFNet has been extensively experimented on three publicly available datasets UC_Merced， RSSCN7 and AID datasets and yielded remarkable accuracies of 99.10%， 96.60% and 95.48%， respectively. Experimental results demonstrate the superior multi-scale feature extraction and fusion capability of FS-LMFFNet， which achieves better performance than other state-of-the-art models.

Aiming at the problems of difficult identification of occluded objects and insufficient robustness to noise and viewing angle changes in existing instance segmentation， this paper proposes a method of multi-source spatio-temporal information based fine-grained bird's-eye view generation（MSTFB）. The method is based on a rasterized scene bird's eye view， the self-attention mechanism is utilized to fuse temporal bird's eye view features to obtain the scene fine-graine bird's eye view， and the spatiotemporal cross-domain convolutional network is employed to capture the relative position information between instances and fuse the multi-scale features. On this basis， a bird's-eye view instance segmentation prediction method of encoding and sample fusion （ESF-BISP） is proposed. ConvGRU is used to encode the time series semantics of the historical frame to obtain the time series features， and CVAE is adopted to model the state feature distribution of the current frame fine-grained bird's eye view and sample the bird's eye view sample features， GMM is used to fuse the time series features and sample features of the bird's eye view， and then decode the fine-grained aerial view of the future frame scene. The experimental results on the public dataset nuScenes show that compared with the benchmark algorithm LSS， the vehicle segmentation IoU index of MSTFB method is improved by 7.09%， which can effectively segment remote vehicles and occluded vehicles. ESF-BISP can better capture the changes of dynamic instances in the scene， whether for instance segmentation or for future instance segmentation prediction， the performance is significantly better than the benchmark algorithm.

Aiming at the problems of target misdetection and omission， inaccurate association， and re-identification error in multi-target tracking in dense pedestrian scenarios， this study proposes a multi-pedestrian tracking network based on Transformer. The algorithm consists of three modules： detection， data association and tracking， in which the detection module adopts the selective query recollection method to enhance the decoder's collection of key features， improve the model's ability to characterize the target， and effectively reduce the problem of target misdetection and omission； the data association module adopts the fusion strategy of bilinear LSTM and quadratic data association， to solve the inaccurate association of dense pedestrians due to the similarity of the appearance of the target； Finally， the attention pyramid is embedded into the pyramid spatio-temporal aggregation module on the tracking module to capture the spatio-temporal information of the feature map at different scales， which improves the accuracy of target re-identification.The performance of the proposed network is tested on the publicly available datasets MOT16， MOT17， and the experimental results show that the method in this study is able to achieve more accurate multi-pedestrian tracking compared to other methods.

Aiming at the situation where the state of wireless communication links changes over time， in order to achieve real-time updating and adjustment of link reliability prediction，in this paper a wireless communication link reliability confidence interval prediction algorithm based on GA-Elman is proposed. Based on the analysis of the quality characteristics of wireless communication links， a logarithmic distance path loss model is constructed to transform the study of wireless communication link reliability into the study of wireless communication link signal-to-noise ratio. The wireless communication link signal-to-noise ratio signal is decomposed into stationary sequences and noisy sequences through wavelet decomposition， and respectively input into Elman neural networks for prediction. Iteratively optimize the weights and bias parameters of the Elman neural network using genetic algorithm （GA） to improve the accuracy of the prediction model. Combine the predicted values with confidence levels to achieve confidence interval prediction of wireless communication link reliability. Through experimental verification found that the optimal Elman neural network parameters can be obtained when the genetic algorithm iterates 60 times， effectively improving prediction accuracy. This algorithm has higher accuracy and reliability in the field of wireless communication links， and can provide more accurate prediction information， providing strong support for the reliability evaluation and optimization of wireless communication links.

To address the issues of the increase in the number of communication devices and the co-channel interference existing in unmanned aerial vehicle （UAV） cooperative communication process， in this paper a multi-UAVs system with multi-channel co-channel interference is constructed. In this system， there is a direct link between the source node and the destination node， the destination employs either selection combining （SC） or maximal ratio combining （MRC） technology to integrate the signals transmitted from the source node and the optimal UAV relay. Considering the correlation between signal to interference plus noise ratio （SINR） caused by co-channel interferences， the precise expressions for the system outage probabilities with the SC strategy and MRC strategy separately under Nakagami-m fading channels are derived. To further discuss the impact of co-channel interference on the performance of the UAV communication system， the asymptotic outage probability of the system under high signal to noise ratio （SNR） is derived and analyzed in detail. The results show that due to the influence of co-channel interference， the system outage probability exhibits a saturation value in the high signal to noise ratio region； as the Nakagami-m fading parameter m increases， the system outage performance is significantly enhanced； with the increase in the number of UAVs， the system outage probability gains more diversity benefits before reaching the saturation value.

In order to accurately carry out target recognition and tracking， a non equilibrium airborne laser synthetic aperture radar image target recognition method is proposed. Firstly， based on the principle of Kalman filtering， an image correction model is established to correct the distortion phenomenon of images in non-equilibrium states； Secondly， the color compensation rate was calculated to compensate for the color channels of SAR images and image clarity was improved； Finally， the processed SAR image is input into the extended convolutional capsule network to achieve image target recognition through multi-scale feature fusion and feature learning. The experimental results show that the proposed method has good image processing performance and high target recognition accuracy.

Aiming at the security risk of the unauthorized rejected UAV in the restricted flying area， and the problem that it is difficult to counter， this paper proposes an algorithm to use a UAV to intercept the unlicensed and uninformed UAV at high speed. Our algorithm uses a navigation and control architecture to solve this problem. In the aspect of navigation， low-cost omnidirectional perception was realized through the combination of camera and laser radar. Through the fusion of high precision and low frequency lidar data and high frequency and low precision vision data， accurate target position estimation and tracking are obtained. In the control aspect， the dynamic authority allocation control method based on trajectory prediction is used to realize the effective countermeasures against the large maneuvering UAV. Finally， the effectiveness of the proposed algorithm is verified by experiments and simulations. It is verified that the proposed algorithm can lock the denied UAV with a success rate of more than 90% and intercept the UAV with a success rate of more than 85%.

In this paper， four non-metallic materials， polytetrafluoroethylene， polyurethane， polypropylene， polymethyl methacrylate， were selected as adhesion substrates to study the effect of different non-metallic substrate materials on shear ice adhesion strength. Firstly， the static contact angle of the non-metallic substrate surfaces and the surface hardness of different substrates were measured； secondly the shear ice adhesion strength of 5 mL of water frozen at -20 °C for 1 h was measured by a self-made experiment device. The test results showed that the shear ice adhesion strengths on the four non-metallic substrate surfaces were 82.89， 45.26， 99.53， 151.27 kPa， respectively. The shear ice adhesion strength on the non-metallic substrate surface was not linearly to surface wettability， but positively linearly related to the hardness. It was found that the water attached to the substrate surface could undergo phase transformation during the freezing process； in low-temperature environments， low-hardness materials could be prone to shrinkage when exposed to cold. In anice-solid adhesion system， the micro-deformation would occur on the substrate surface in contact with the accreted ice. This deformation interferes with the stability of the adhesion interface between the ice and the substrate. Hence， the shear ice adhesion strength on the non-metallic substrate surface would be reduced. This paper could provide theoretical support for the development and optimization of the anti/de-icing method in the engineering field， especially for the selection of non-metallic materials in the components to be anti/de-iced.