Beam pumping units are main oil producing equipments in land-based oilfield. Oil producing energy consumption caused by beam pumping unit accounts for one third of the total energy consumption in oilfields. However， due to the structure of the system and operating conditions， the average load rate of pumping unit motor is very low， which is generally below 30%. In order to improve the utilization efficiency of oil recovery energy， many new technologies and methods have been put forward and tried by scientists and technicians， which have got good results； meanwhile， limitations and technical risks exist. Firstly， main sources of energy consumption are analyzed， and it is pointed out through comparison that the motors have the most potential in energy-saving. Then， the researches on improving electric efficiency of motor recently reported in public are classified and summarized. Relevant technical researches can be divided into three categories： （1） Improvement of mechanical transmission structure； （2） improvement of motor and its control technology； （3） adding energy-saving devices. Based on the background of offshore oil exploitation and shale oil and gas exploitation， through summarizing and analyzing various technical schemes， it is proposed that adding energy-saving devices to traditional pumping equipment will be the most effective technical direction for beam pumping units in the future. Among them， hydraulic hybrid energy-saving technology has more obvious advantages.

A survey of the mainstream multi object tracking methods based on tracking by detection in recent years is carried out. Then， the concept of detection based multi object tracking is introduced. The multi object tracking methods are summarized in object detection， feature extraction and data association & tracking. The performance of some multi object tracking（MOT） methods are compared and evaluated on the MOTChallenge series datasets. The future development direction of multi object tracking is discussed.

In order to explore the dynamic characteristics of particle collision in diesel engine exhaust phase， a simulation model of particle collision was established by coupling discrete element dynamics software EDEM with Fluent based on similarity theory. The collision process of diesel engine with different exhaust pressure difference and flow velocity were simulated. The effects of collision dynamics parameters such as angular velocity， rotational turbulent kinetic energy， torque and collision scraping force on the diesel engine during collision were analyzed. The results show that the normal force and tangential force increase by 1.5 and 1.7 times respectively when the velocity of flow is constant， the particle size is the same and the exhaust pressure difference increases from 0.188 MPa to 0.268 MPa， and the average rotational turbulent kinetic energy increases from 2.26×10^-9 J to 3.52×10^-9 J. When the pressure difference is constant， the particle size is the same and the gas flow rate increases from 5.65 m/s to 6.78 m/s， the normal force and tangential force increase by 50.5% and 45.5% respectively， and the average angular velocity of particles increases from 7.87×10⁵ rad/s to 10.85×10⁵ rad/s. The research results can provide a basis for improving diesel particulate filter （DPF） capture efficiency.

Aiming at the problems of diesel engine combustion process complexity， mechanism modeling difficulty， dynamic coupling and constraints， a data-driven modeling and rolling optimization control method for diesel engine combustion process was proposed to reduce fuel consumption and torque tracking， and to limit NOx emission and control volume. First， a 2L diesel engine model in GT-suite is chose as a plant. The influences of VGT and EGR control valve opening， injection angle and injection amount on fuel consumption rate， NOx emission and crankshaft output torque are analyzed. Second， using input and output data， a control-oriented data-driven prediction model is obtained by using subspace identification method. Third， to reduce the fuel consumption rate and torque tracking error， and the NOx emission and control input are considered as constraint， the optimal control input is obtained by solving a multi-objective optimal control problem. Finally， the co-simulation based on MATLAB and GT-suite is presented to verify the effectiveness and superiority of the data-driven receding optimization control method.

According to the functional characteristics of Multi-wheel distributed hybrid-driven unmanned frame vehicle （MHUFV）， the vehicle control system is developed. This system includes high-voltage power-on and power-off management based on the coordination of two power sources， fault diagnosis and fault-tolerant control of complex systems with multiple control nodes， and the algorithm design like driver input command determination with dual input mode， anti-skid control， mode switching and energy distribution control. On this basis， the MHUFV software architecture is constructed， and the vehicle controller is developed by using the Rapid Control Prototype （RCP） technology platform. The basic performance of the vehicle is studied by real road test. The results show that the developed vehicle controller can meet the design requirements of the real vehicle， the test results of each key state quantity are reasonable when the full load limit speed is 15 km/h， and the fuel saving effect is 38% compared with the traditional diesel vehicle under the same operating condition， which can meet the development requirements of the current frame car industry.

A new type of electric vehicle powertrain configuration was proposed in order to make use of the flexibility of open-winding permanent magnet synchronous motor in power distribution between dual energy sources， and realized the process from power distribution instruction formation to instruction following under this configuration. First， an expected power calculation method based on fuzzy control is proposed to make the main energy source work in high efficiency zone. Then， at the level of motor control， based on the voltage vector model of dual inverters， two distribution methods， namely precise power following and linear distribution， are proposed to realize the expected power following of primary and secondary energy sources. The simulation results show that under the voltage vector distribution method， the motor speed and torque follow well， and the torque fluctuation is limited within 3 N·m.

Under variable working conditions， the repeated oil liquid sloshing often results in sloshing noise， the interruption of oil liquid supply under the fast change of acceleration， the increase of evaporative emissions， and the structural damage of fuel tank and its inner parts. To optimize the suppression structure of oil liquid sloshing， the 3D finite element models of fuel tank filled with oil liquid are established， and the rationality of simulation parameters are verified. The accuracy of simulation results are verified by test. The influences of baffles with no holes， few holes and a large number of holes， high and low baffles and six different shapes of baffles on oil liquid sloshing are simulated by VOF method. The concept of time-area value is proposed on the basis of free oil liquid surface area. Comparing the parameters of oil liquid sloshing， it can be seen that the best structure is baffles with few holes and circular holes， which effectively reduces the pressure of oil liquid sloshing and time-area value. The corrugated and straight baffles have positive influence on suppressing the sloshing pressure of oil liquid at the center point and on the center line， respectively. The sloshing pressure of oil liquid is the greatest in fuel tank with eccentric arc baffles. The maximum and minimum time-area values of oil liquid are in fuel tank with multi-arc and corrugated baffles， respectively. The baffles obviously attenuate the change amplitude of free oil liquid surface area. The combined action of inertia force and gravity is beneficial to the decrease in time-area values.

In order to study the seismic isolation performance of waste Scrap Tire Pads （STP） under aging-loading coupling， 6-layer STP with a geometry of 180 mm×180 mm×69 mm were selected for aging test at a temperature of 100 °C. The hot air accelerated aging test was carried out for 77h， 154h， 231h， and 308 h under the load of 5MPa. The STP with accelerated degradation were subjected to the pseudo-static test under the wall. The effects of aging time and coupling load on the hysteresis， skeleton curve， horizontal equivalent stiffness， equivalent damping ratio， residual displacement and stiffness degradation of the layered STP under the wall are analyzed. Combined with internal/external damage behaviors such as STP deformation， degumming， wire mesh failure and crack propagation in the aging test， the influence law of the effect of thermal oxygen erosion on the seismic isolation performance of STP and the failure mechanism of isolation performance are expounded comprehensively. The experimental results show that the new STP structure system has good effects of isolation， limit and reset， and stable mechanical performance in the whole life cycle of building structure， and reliable isolation performance. On the basis of considering STP self-reset， it is recommended that its safe service life is 50 years. The research results are intended to provide a theoretical basis for the popularization and application of the new type of under-wall waste laminated tire isolation structure in high-intensity villages and towns.

In order to explore the influence of spoke features on vehicle aerodynamic drag coefficient （C_d）， an improved MIRA model was studied in Fluent to investigate the relationship of vehicle aerodynamic drag coefficient with different spoke offset distances and curvatures. Meanwhile， the reasons of drag variance in different cases were analyzed. Then combination cases were explored on the influence of various conditions on the vehicle C_d. The simulation results indicate that smaller spoke offset distance is beneficial to the C_d reduction， and when the wheel spokes offset distance is 10 mm， the vehicle C_d is minimum， which is 0.2514. Compared with the base case， the rate of C_d reduction is 5.56%. The increase in spoke curvatures will increase the vehicle aerodynamic drag coefficient.

Dual-material auxetic double arrowhead structure with negative Poisson’s ratio was proposed. Under the assumption that all geometric dimensions were the same， the stiff material fraction ψ was introduced. Based on explicit dynamic finite element simulation by LS-DYNA， the in-plane dynamic responses of dual-material auxetic double arrowhead structure with negative Poisson’s ratio were numerically studied. The effects of the stiff material fraction and the impact velocity on the in-plane impact resistance and energy absorption capacities were discussed in detail. In order to obtain better comprehensive performance， a functionally layered honeycomb model was established and the effects of the gradient forms on the peak crush force， specific energy absorption and plat-form stress were investigated. The results show that the platform stress has a great improvement and increases with the fraction of the stiff material， and the influence mechanism on the peak impact force and the specific energy absorption are significantly different from the others. Under the presupposition that the peak impact force was lower， the energy absorption capacity can be improved by reasonable designing gradient forms to realize the control of the energy absorption process.

To solve the problem that manufacturing enterprises generally can't make full use of the data resources of metal cutting scattered in various application systems， an approach of building metal cutting knowledge graph is proposed to realize cutting data integration and enhance data value density. Firstly， metal cutting knowledge is classified into factual type and procedural type. A complete ontology model is established by using OWL language and top-down method. Then， a multi-source data integration framework is constructed， and a data fusion algorithm suitable for identifying the equivalent entity of cutting data is determined as well. Finally， a visualization system of knowledge graph is developed， which has been applied to an aero-engine maintenance company in China. The research results of this paper provide strong support for the construction of data-driven based and intelligent metal cutting.

Zero speed detection is a key technology in vehicle integrated navigation and is the basis for zero speed error correction and initial quasi-static alignment. Zero-speed detectors that rely solely on IMU （Inertial Measurement Unit） have zero-speed misjudgment in dynamics conditions， which leads to the problems such as erroneous updates. In order to solve the mentioned problems in vehicle integrated navigation， a novel zero-speed detection method based on generalized likelihood ratio test and IMU/wheel speed fusion is proposed， which is then verified by experiments. The test results show that compared to the zero-speed detector relying only on IMU or only on the wheel speed， the zero-speed detector based on the IMU/wheel speed sensor fusion could reduce the misjudgment in dynamic conditions， and improves the accuracy of zero-speed detection.

The aging behavior and mechanism of adhesively bonded joints of CFRP- aluminum alloy exposured to hygrothermal environment were studied in this paper. First， the joints were aged at 80℃ with 95% relative humidity （RH） for different durations and then the quasi-static tensile and shear tests were performed. Second， the residual strength and failure mode of the joint were analyzed. Third， combined with the Fourier transform infrared spectroscopy （FTIR） results of adhesive， the aging mechanism of adhesive joints was revealed. The results indicate that the strength of joint decreases significantly due to hygrothermal aging， which approximates the exponential function. The failure modes of unaged and aged joints all presented cohesive failure primarily. However， with the increase in aging duration the adhesive-substrate interface was damaged， and the surfaces of aged joints became relatively rough due to the plasticizing of the adhesive. The hydrolysis reaction occurred in the adhesive due to hygrothermal aging， resulting in chain scission. The post curing of adhesive was also observed at the initial stage of aging. Adhesive degradation is the primary cause of the decrease in the strength of the joint. Finally a residual strength prediction method was established， which can be used to predict residual strength of hygrothermally aged joints by testing the chemical characteristics of adhesive.

Polyimide （PI） and Poly-tetra-fluoroethylene （PTFE） have been widely used in the field of fluid boundary and mixed lubrication because of their excellent self-lubrication properties. The tribological properties of the PI based coatings and PTFE based coatings sliding against on 38CrMoAlA in 70℃ aviation kerosene were studied by altering p and v in this paper. The relationship among the wear factor k^*， the linear wear rate Wt， the limiting pv and the loaded pv was discussed， and the worn surfaces of the coatings was analyzed by SEM. The results show that the wear factor k^* and the limit pv are not constant values， but closely related to the wear mechanism of the coatings. The PTFE based coating against 38CrMoAlA has more excellent friction and wear properties， which make it more suitable for the tribopair material of aviation kerosene piston pump.

The active suspension structure of the Mars Rover was designed based on Rocker-Bogie Mechanism， and the creeping control strategy was proposed during a large sinkage. Four Mars Rover creeping tests were carried out. The test results show that the Mars Rover finally escaped from the large sinkage during all of four creeping tests. The average power of the Mars Rover is 13.5-18.94 W， and the maximum power is 24.8-52.4 W when the Mars Rover suspension is creeping. The average current and maximum current of the six drive motors are increased correspondingly compared with normal driving. The average current increases by 9.7%-74.8%， and the maximum current increases by 9.6%-34.5%. The results show that the active suspension combined with the creeping escape strategy can solve the sinkage risk and improve traffic ability of Mars Rover of China.

Aiming at the thinning defect which is easy to occur in the processing of aluminum profiles， the thinning rate of multi-point flexible forming parts was studied on the self-developed flexible three-dimensional bending equipment for aluminum profiles. The variations of the thinning rates of the outer web and the inner web along the length of the profile after the three-dimensional multi-point stretch-bending deformation of the rectangular section profile are studied. The results show that the thinning rates of the outer and inner webs of the profile gradually increase from the symmetry center of the profile along the direction of the profile during the forming process， and the thinning rate of the outer web is larger than that of the inner web. The change of die number makes the maximum thinning rate of the outer web up to 1.682% and the minimum is 1.317%. The thinning amplitude is 0.365%. The maximum thinning rate of the inner web is 1.657%， and the minimum is 1.198%. The thinning amplitude is 0.459%， and the number of dies is the main processing parameter that causes the variation of the thinning rate of the parts.

The influence of tensile-unloading pre-strains on the mechanical behavior of shape memory alloy-metal glass composites and its internal mechanism were studied by finite element simulations. Simulation results show that both the yield strength and the limit of the elastic strain of the composites increase with the pretreatment strain， but the tensile ductility decreases sharply. With increasing pretreatment strain， the subsequent deformation of shape memory alloy particles shifts from a classic martensite phase transformation to a continuous martensite transformation， and finally to a simple martensite elastic deformation. The weakening or even disappearance of the phase transformation effect leads directly to the deterioration in the tensile performance of the pretreated shape memory alloy-metallic glass matrix composite.

In this study， first， basalt fiber surface was modified via silicon carbon black （SiCB） with dopamine （DOPA） solution to improve the interfacial bonding between basalt fiber and polyamide 6 （PA6）. Then， the effects of different SiCB contents on the properties of basalt fiber/polyamide 6 composites were investigated. The microscopic morphology and structure of the modified BF was characterized by Fourier transform infrared spectroscopy， differential scanning calorimetry （DSC） and scanning electron microscopy （SEM）. The fracture behavior of the composite was evaluated by static testing. When the content of SiCB is 2.0 g/L， the tensile strength and impact strength of the composites are increased by 109.4% and 89.3%. SEM proves that the silica carbon black on the surface of BF is homogeneous. According to the DCS results， the silica carbon black effectively reduces the crystallization enthalpy of PA6 and improves the crystallinity of PA6.

Aiming at the springback problem during manufacturing process of aluminum alloy thin-walled parts， a springback prediction method for double-curvature stiffened panel in milling process was proposed. The method estimates the springback value in the two directions of the panel according to the redistribution of the residual stress after the material is removed layer by layer. The finite element simulation results show that the stress distribution and radius after springback of the panel during the milling process were close to the calculated results. The maximum stress variation in both directions appears when the milling depth reaches the initial neutral surface. When the remaining thickness decreases， the errors between calculation results and FEM results increase gradually. In addition， the simulation results of different single-layer milling thickness show that the larger the thickness of single-layer milling， the larger the error of this method， and the maximum error is 9.07%. The forming-milling experiments were performed to verify the proposed method. The results show that the proposed method could predict the springback of the double-curvature stiffened panel.

In order to obtain the influence of crowding degree on the choice between two modes， the ground public transit and subway， this paper proposes a method to calculate the maximum crowding degree in the operating time. Passenger flow survey is conducted for 71 bus lines， metro line 1 and 3 by manual sampling survey in Harbin City. The maximum passenger number is observed and the crowding degree is calculated. Combining RP investigation with SP investigation， the corresponding relationship between the measured data of congestion degree and the congestion degree of passenger's subjective perception is established. The travel time， crowding degree and travel cost are graded in detail， and the situation of public transport travel choice is designed. The discrete Logit model is established. The results show that the crowding degree has the greatest influence on the public transit choice， the travel cost is the second， and the stroke time has the least impact.

To overcome the shortcomings of existing braking performance testing methods， a non-contact dynamic testing method for vehicle braking time sequence was proposed. Based on the relationship between slip rate and adhesion coefficient， a measurement target of braking time sequence based on slip rate was put forward. Based on the measurement principle of binocular stereo vision， a measurement model for wheel slip rate based on vision was established. With the help of LM （Levenberg-Marquardt） algorithm， the calibration parameters were optimized nonlinearly. Image processing techniques such as pseudo-median bilateral filtering， Canny edge detection， redundant boundary clearance and Hough transform were used to denoise， extract edges， simplify and extract features respectively， by which the central coordinates of circular markers were obtained. To verify the feasibility of the proposed method， a real-time test was carried out and the standard uncertainty evaluation results of measurement errors were given. The results show that the maximum relative error of the proposed method was 2.74% and the maximum repeatability error was 3.88% under the conditions of U=2.52 and k=2.

In order to meet the requirements of unstressed length and elongation control for stay cable during its installation， the analytical method for calculating the elongation of the stay cable with catenary configuration is discussed. First， the geometric equation of the stay cable with catenary configuration is obtained by applying the force equilibrium condition to the stay cable， and the analytical expressions of the cable length and its elongation under a certain given cable tension are acquired by direct integration. Moreover， the analytical expressions of the cable elongation under varying cable tension are obtained by using the deformation coordination condition. Then， the equivalent elastic modulus for stay cable with catenary configuration is discussed based on strain equivalence. It is verified that the equivalent elastic modulus for the stay cable with catenary configuration can be simplified to the equivalent elastic modulus for the stay cable with parabolic configuration. Taking the real parameters of the stay cables of a certain cable-stayed bridge with composite/hybrid girder as the case study， the analytical expressions of stay cable length and its corresponding elongation under a given cable tension are verified by numerical integration and simplified formulas respectively. Furthermore， the analytical expressions for the elongation of stay cable with varying cable tension are verified by the Nlabs finite element and equivalent elastic modulus method. The results show that the proposed analytical expressions can simply and directly figure out the exact elongation of stay cable， which provide the theory and straight forward method for the unstressed length of stay cable and the elongation control of stay cable during its installation.

The free vibration characteristics of cracked box girder bridges were analyzed in this paper. Firstly， the additional flexibility matrices of box girder in the cases of deep crack and shallow crack were derived based on the basic principle of local flexibility formation， and the stiffness matrix of cracked beam element was obtained， and then the free vibration characteristic analysis method of cracked box girder bridge based on finite element method was formed. Next， the dynamic characteristics of the cracked cantilever steel beam were tested， and the experimental results were compared with the natural frequency calculation results to verify the accuracy and reliability of the proposed method. Finally， the numerical analysis of continuous reinforced concrete box girder was carried out， and the influence of crack location， depth and number on the natural frequency of box girder bridge and the influence of crack on vibration mode were analyzed.

The traditional methods for assessing the fire damaged concrete structures require large amount of experimental efforts， and some other new technologies are expensive and their applicability still needs further investigation. In view of this， a colorimetric method based on the optical analysis is proposed to evaluate the heating damage of the concrete in terms of the translation of the color coordinates in the chromaticity diagram. Fire damage experiment was first conducted on the C30 concrete cubes. And then photos were taken under the conditions of daylight vs. fluorescent and manual white balance vs. auto white balance. The chromaticity diagram is constructed to represent the relationship between the coordinates of the concrete colors and the maximum temperature that the concrete was subject to. The maximum temperature that the concrete experienced can thus be determined based on concrete color. The conditions under the daylight and the manual white balance are recommended to take photos for constructing chromaticity diagram. The results of this study will advance the technology for assessing fire-damaged concrete.

In order to improve the accuracy of settlement calculation of tunnel shield construction， a method for measuring and calculating ground settlement of tunnel shield based on NNBR model is proposed. Taking the Wangjing tunnel shield section to the airport Sanyuanqiao~T3 section as the research area， the general situation of the research object was analyzed， the engineering data are investigated， the relevant actual data are obtained and the corresponding analysis is made. Based on the measured data， combined with the NNBR model（Nearest Neighbor Bootstrapping Regressive Model）， the basic principle of generating similar results based on similar reasons is that the most similar one is detected in the historical sample set， or several samples are used as calculation results to obtain the final tunnel shield construction surface. It is verified that the calculated method has small relative error and is feasible， the talent method improves the accuracy of engineering survey and provides a theoretical basis for future research and development.

Desulfurization gypsum is a kind of industrial wastes. When mixed with fly ash and slag， it can enhance the reaction efficiency， increase the cohesiveness of cementation systems and reduce the amount of cement. Furthermore， gypsum cement gel material has a slight expansion function， which contributes to improving the compactness of materials， delaying the shrinkage cracking. Currently， gypsum is just used as roadbed filler， which can not fully play its performance. Based on this background， in this paper， the gypsum cement stabilized aggregate formation mechanism and its engineering application were investigated to identify the mechanism of “excitation”， “enhancement” and “reduction” of shrinkage. Referring to the cement performance test， the effects of gypsum on the retarding effect， strength characteristics and stability of the composite gel system were analyzed. Scanning electron microscope images and X-ray diffraction patterns were applied to reveal the mechanism of the effect of gypsum on the chemical reaction of composite gel system. The results reveal the volumetric micro-swelling and the three-stage development of the composite gel system. The feature of volume swell is derived from the crystallization of gypsum and the formation of ettringite. The three-stage strength development characteristics are derived from the excitation of desulfurization gypsum to active ingredients in the slag， crystallization pressure generated by the ettringite formation， and gypsum filling. For medium-particle composite stabilized aggregates， the proportion of slag should not be lower than the proportion of cement incorporation and the proportion of gypsum should not be less than 3.0%， not be more than 6.0%.

In order to study the shear behavior of reinforced concrete beams based on the elastoplastic stress field theory （EPSF）， ICONC， a finite element program based on the EPSF theory， was used to model the concrete members. The influences of finite element mesh size， shape and iterative step number on the simulation results were investigated. Twelve test beams were selected. The simulation results based on EPSF theory， experimental results and ABAQUS simulation results were compared to verify the accuracy of the results based on EPSF theory. The shear capacities of 70 beams subject to shear failure were calculated by the formulas in the Chinese and American code， and also calculated by EPSF theory， then those calculated values were compared with the experimental values. The results show that the mesh size， shape and iteration steps have little influence on the simulation results. ICONC， an applicable program based on EPSF theory， can accurately simulate the failure phenomena of reinforced concrete beams， the yield of reinforcement and the crack distributions of concrete. Compared to ABAQUS， EPSF theory can permit an accurate， convenient and rapid prediction of the ultimate shear capacity for reinforced concrete beams. The changes in concrete strength， shear span ratio and stirrup ratio have little impact on ICONC program simulation results. So the proposed method has certain accuracy and stability.

In order to study the seismic performance of two different forms of structures， i.e. "functional integration" and "functional separation" in the Reinforce Concrete （RC） frame-seismic wall dual structure system， quasi-static tests of two 1/3 scale frame-seismic wall specimens with two single-story and two-bay were carried out. The damage development， hysteretic behavior， stiffness degradation， energy dissipation capacity， ductility and other seismic behaviors of the seismic wall were investigated as the wall mainly bears the lateral resistance but bears a small part of the vertical load， and as the wall bears both the lateral resistance and the vertical load. The performance of the specimens was also studied by ABAQUS finite element software simulation， and the results are consistent with the test results．The results of this study show that the structure has better ductility， energy dissipation capacity and damage development process when the load-bearing function and the anti-lateral function of the seismic wall are separated.

In event-based social network， users and activities construct heterogeneous network naturally. We propose an end-to-end recommendation model using high-order connectivity in heterogeneous network and non-linear matching relations to increase hit ratio. First， we extract the effect of high-order connectivity of each node， and get the embedding of nodes by graph neural network. Then， the embedding of user-activity pair is input of the multilayer perceptron to gain the current matching probability. Finally， by repeating this process the final matching probability is obtained. Experiments show that， the recommendation method is stable. Comparing with existing algorithms， our algorithm increases hit ratio more than 10%， and normalized discounted cumulative gain about 10%. Comparing with using multilayer perceptron only， both hit ratio and normalized discounted cumulative gain rise when considering high-order connectivity information.

In order to make pencil drawings more artistic and retain the structure of the original face images， a method of transforming real face images into pencil-style images by Generating Antagonistic Networks is proposed. Generation Network generates face pencil drawing images by convolution， and Antagonistic Network is used to learn the distribution of artist's real face pencil drawing paintings. The distance deviation is optimized and the loss of detail control is added. The experimental results show that the face pencil drawing images generated by this method is more flexible and artistic.

To solve the problem that current image multi-label annotation methods can only annotate image content information （ontology）， but can not simultaneously annotate image implied information （implicit）， this paper proposes a double-layer multi-label annotation model based on multi-task learning （MTL-DMAM）. Firstly， the image ontology annotation and implicit annotation are regarded as two related tasks， and ResNeXt-50 is used as the backbone network of shared features. Then， in order to realize image double-level annotation， attention mechanism is used to construct a branch structure for each task. In order to eliminate the influence of different object sizes on labeling results in images， the ELASTIC structure is added to the model to improve the performance of the model. The comparative experiment results show that， on single task MS-COCO data set， the proposed model is superior to most advanced models in the indicators of C-R， C-F1， O-R， and mAP， and on multi-task traditional costume data set， the proposed model is superior to all other models in 10 indicators. Finally， we use the Grad-cam method to visualize the image region that MTL-DMAM focuses on when labeling， and the experimental results show that the proposed model can effectively learn the salient features of the image corresponding to labels.

Micro-expression is a kind of facial feature that can reflect the most real emotional state hidden in human heart. Existing micro-expression recognition technology often extracts a lot of feature information unrelated to micro-expression in feature extraction. In order to solve the above problem， this paper proposes a feature extraction method of specific region of face micro-expression based on deep learning. Firstly， we preprocess the original micro-expression sequence to locate the key regions of human eyes， mask the eye regions， and reduce the blinking action to extract micro-expression features. Then， Euler video magnification algorithm is used to magnify the key areas of facial micro-expression change， which makes the part of micro-expression change more obvious. Finally， the continuous micro-expression sequence is trained and recognized by 3D ResNet network. The proposed method is tested on CASME II and SMIC datasets， and he experimental results show that the accuracies are 77.3% and 72.4% respectively. Compared with the latest methods DSSN， CNN+LSTM， the accuracy is improved at least 5%.

To solve the security problem caused by OpenStack Swift cloud storage system directly and completely storing user files， this paper proposes a method to enhance its security function. First of all， the user files are divided into several fragments by using data dispersal technology， which makes it difficult for attackers to obtain complete user files； Second， the segmented files are stored in ciphertext form through data encryption technology to prevent attackers from using the logical relationship between segmented file contents to recover complete user files； In addition， this paper presents a hierarchical management approach to protect various keys in different layers， and combines user password with secret sharing to prevent cryptographic materials leakage. The experimental results indicate that it will increase the difficulty for attackers to obtain user files after using this method， so as to ensure the security of data storage in OpenStack Swift system， and the performance cost is acceptable to users.

In order to improve the image quality of night roads， to promote the visibility and contrast of the target in the image， and to reduce the difficulty of target detection in night road image， a night road image enhancement method based on the optimized Multi-scale Retinex （Multi scale retinex，MSR） algorithm is proposed in this paper. First， the Red， Green and Blue （RGB） image is converted to the YUV color space. Secondly， the optimized MSR algorithm is constructed by using the reciprocal of the Just Noticeable Distortion （Just noticeable distortion，JND） as the coefficient before the incident image in the MSR algorithm， which is used to adjust the brightness of the channel Y adaptively. Meanwhile， the channels of U and V are proportionally adjusted to obtain a new image， which is then combined with the original image in a 1∶1 ratio to preserve the image details. Finally， the Contrast-limited Adaptive Histogram Equalization method （CLAHE） is used to improve image contrast. Experiments on datasets containing 770 images of night road show that the proposed method exhibits the ability to adaptive adjustment of night image brightness， alleviate uneven brightness of the image， enhance the sharpness of nighttime images and the detail information of image. By using the support vector machine algorithm to detect vehicles in image， the missed detection rate and false detection rate are decreased by 2.61% and 4.35%， respectively..

In order to solve the problem of energy consumption control and system outage of D2D resource allocation algorithm， a dynamic exhaustive D2D resource allocation algorithm with simulated annealing is proposed. The algorithm uses a dynamic interval exhaustive search algorithm to initially determine the user transmit power， and creates a two-dimensional multiplex table containing user QoS information. Then the algorithm combines the QoS of the cellular user with the D2D user to determine the multiplexing combination， adds an adjustment power module in the power dimension and introduces a simulated annealing algorithm in the combined dimension to reduce the outage probability. The simulation results show that， compared with the traditional algorithms， the dynamic exhaustive resource allocation algorithm with simulated annealing decreases the average power by 78.3% during the power allocation phase， increases the probability of connectivity in the channel allocation phase by 10.2% on average， and reduces the calculation time by 10.1% on average.

At present， remote sensing satellite staring imaging attitude planning usually only considers the roll axis and the pitch axis， resulting in small angle rotation of the simulated image， reducing the video imaging quality. Also most of the simulation process does not consider the realtime simulation， that the simulation time is too long to meet the engineering requirements. To overcome the above problrms， in this paper， the yaw compensation for image rotation is corrected， and a fast simulation method for staring imaging for flat terrain is proposed. First， the formula for calculating the three-axis attitude of the spacecraft during ground staring imaging is deduced. Then， the geometric model of staring imaging is established based on the central projection， and the two-dimensional Lagrangian interpolation method is used to reconstruct the local surface of the earth. A simulation experiment for staring imaging of flat terrain was carried out. The simulation parameters were set to a track height of 526 km， a camera focal length of 1.45 m， and a pixel size of 5.5 μm. The simulation results show that the attitude planning error under this simulation condition is less than 0.004 m. When the elevation of the target area is simulated without considering the elevation change， the pixel error is less than 0.031. Compared with the direct method of point-by-point calculation， the simulation time is reduced by 87%.

This paper considers the adaptive cruise control problem of vehicles in the vehicular ad hoc network and proposes a delay-feedback predictive cruise control scheme of autonomous and connected vehicles for the case of bounded long time-delay in the information transmission of front vehicles. Firstly， the incremental control and time delay model is used to formulate the adaptive cruise systems of autonomous and connected vehicles. Then the delay-feedback predictive cruise controller is computed by the receding horizon optimization principle in terms of the driving safety， ride comfort and economy performance of the vehicle. Finally， by the Lyapunov’s stability theorem and linear matrix inequality technique， the delay-dependent sufficient conditions are established to guarantee closed-loop stability of the predictive cruise control system of the autonomous and connected vehicle. Comparison with PID control simulation is used to illustrate the effectiveness of the proposed method.

In order to overcome the issue of large amount of calculation and long computing time of the traditional multi-threshold image segmentation methods， a multi-threshold image segmentation method based on improved cuckoo search algorithm is proposed. Firstly， the teaching-learning search strategy is introduced into the cuckoo algorithm to improve local search ability of the algorithm. Secondly， the elite solution with better fitness value in the current population is selected to construct the elite database， and the elite solution is randomly selected to guide the search direction， so as to strengthen the advantage of experience learning. Finally， the simulated annealing mechanism is introduced to select the bird's nest location， which can effectively avoid the individuals falling into the local optimum in the process of optimization. Several different types of complex multi-target images are selected for segmentation experiments in comparison with those of cuckoo search algorithm （CS）， shuffled frog leaping algorithm （SFL）， teaching-learning-based optimization （TLBO） and hybrid PSOGSA with generalized oppositing-based learning （GOPSOGSA）. Experimental results show that the proposed method is superior to the contrasted algorithms in segmentation accuracy， running time and convergence. It can quickly and effectively solve the multi-threshold segmentation problem of complex multi-target images.

To dispose of the influence of time-varying configurations and modeling accuracy on robot structural optimization， a Real-time Modal Analysis （RMA） method for collaborative robots was propose， which combines the finite element method and analytical method. This method could real-time obtain the natural frequencies and mode shapes of robots in any pose with high efficiency and precision. To reduce the calculating amount and realize global structure optimization， based on orthogonal design， a global structural optimization design method of collaborative robots was proposed with the structure parameters as optimization variables and the ratio of robot mass to the global first natural frequency index （M/GF） as optimization objective. The M/GF indexes of SHIR5 robot before and after optimization are calculated. The index after optimization is 9.90% higher than that before optimization， and the global first natural frequency index （GFNFI） is increased by 0.91Hz compared with that before optimization.

In the conventional sampling ways， the cost in data transmission and storage is very high. To reduce the transmission and storage cost of the temperature， a novel compressed sensing system based on semi-tensor product （STP） is proposed. First， a low-dimensional random matrix is generated to globally sample the original data. Then a grouping reconstruction method is proposed to obtain the solution with the iteratively re-weighted least-square （IRLS） algorithm. Numerical results show that the proposed system outperforms conventional way in speed of reconstruction and its comparable quality of reconstruction， which is important for real-time applications.

To enable the weeding component to avoid the seedling on the operation path and reduce the seedling injury rate of machinery weeding， an automatic seedling avoidance control system based on machine vision and hydraulic servo control is proposed in this paper. First， by the way of overshoot， the centroid image coordinate position of the seedling canopy is extracted. Second， through pinhole imaging principle conversion， the seedling ground coordinate position and the distance between the weeding component and the center line of rice row are obtained. Third， a hydraulic control systems model of parallel four connecting rods of organizations is built， and the mapping relations between deviation control and hydraulic stem is acquired. Finally， based on PID algorithm， a proportional valve hydraulic system mathematical model is built. The Matlab/Simulink simulation shows that the model has a steady-state response time of 0.28 s and a static error of 0.07 mm. The results of bench test indicate that the error of the center line of the seedling belt extracted by the image processing system does not exceed 3 cm. The Orthogonal experiment indicates that the factors affecting the seedling injury rate in the avoiding plants control system are in the order of speed of travel， time of transplanting and spacing of plants. When the traveling speed is 0.8m/s， the transplanting time is 10 days， and the intra-row distance of rice seedling is 12 cm， the seedling injury rate is 2.48%.