Abstract:With the development of deep learning technology and the increasing demand for image scene understanding, the application of semantic segmentation networks based on FPGA to provide low-latency and high-energy-efficiency edge-end intelligent services for all users has become a research hotspot. The computing and storage of the semantic segmentation network structure have the intensive feature. To address this issue, the construction of a customized FPGA-based computing structure is a key research issue. In view of this, this paper summarizes the basic principles of semantic segmentation networks and analyzes the characteristics of its internal calculation structure, then elaborates FPGA-based semantic segmentation network computing acceleration methods from two perspectives: model compression methods with hardware resource constraints and custom hardware architecture design. Furthermore, this paper focuses on a summary and analysis of typical methods of computing structure design and memory access optimization in hardware architecture design. Finally, this paper looks forward to the future development trend of FPGA-accelerated computing methods for semantic segmentation networks, in order to provide design references for researchers in semantic segmentation, edge computing, customized energy-efficient computing and other related fields.

Abstract:The ability of unmanned platforms to achieve autonomous positioning and navigation in a wide range of environments is increasingly demanding, in which Lidar-based Simultaneous Localization and Mapping (SLAM) is the mainstream of research schemes. In this work, this paper systematically outlines the framework and key modules of 3D Lidar SLAM algorithm, analyses and describes recent research hotspot problems and future development trends, collates the evaluation criteria for the performance of 3D Lidar SLAM algorithm, based on these, selects six representative mature open source 3D Lidar SLAM algorithms, which are then tested and evaluated on the robot operating system (ROS), based on the KITTI benchmark data set, the parallel comparison is carried out from three aspects: KITTI official precision standard, SLAM algorithm precision index, algorithm time consuming and processing frame rate. The results show that the LIO-SAM algorithm has the best performance among the six algorithms. The RMSE data of ATE and RPE in the 00 sequence data set test are 1. 303 and 0. 028, respectively, and the processing frame rate of the algorithm is 28. 6. Finally, the application trend of 3D laser SLAM technology is discussed based on CiteSpace analysis.

Abstract:The accuracy of target recognition and location is directly related to the picking efficiency, quality and speed of the picking robot. In this article, the research works on target recognition and three-dimensional location of the picking robot in recent years are systematically summarized and analyzed. Several main methods of fruit and vegetable recognition and location are summarized. For target recognition, the methods include digital image processing technology, machine learning image segmentation and classifier and algorithm based on deep learning. For three-dimensional location, the methods consist of monocular color camera, stereo vision matching, depth camera, laser rangefinder and optical 3D camera based on flight time. The main factors that affect the accuracy of fruit and vegetable recognition and positioning are analyzed, which include illumination change, complex natural environment, occlusion and imprecision imaging under dynamic environment interference. Finally, the future development of target recognition and location of picking robot is prospected.

Abstract:To improve the perception ability of typical small water targets for unmanned surface vehicle (USV), this paper proposes an improved SSD object detection algorithm based on multi-scale convolution layer fusion and spatial attention enhancement architecture. Firstly, a multi-scale fusion method is utilized to improve the semantic representation of SSD shallow layer for small targets. Secondly, the spatial attention architecture is designed for each convolutional feature extraction layer to improve feature retention of small targets with weak texture. Finally, the proposed algorithm is evaluated on VOC and self-constructed surface target dataset. The real sea target detection and identification verification based on USV are carried out. Experimental results show that the proposed method can reach high operating efficiency with 15 fps on the USV Nvidia platform. The targets, such as buoys, bridge piers, fishing boats, speed boats and cargo ships, can be identified accurately. Compared with the original SSD algorithm, the proposed method could achieve a higher detection rate for small targets in the typical sea scene, which is increased by nearly 20. 2% when the false alarm rate is 5% . The average effective detection rate can reach 79. 3% .

Abstract:The precise feedback of the end position of the micro-manipulation actuator is of great significance in the micro-automation operation, and the existing research cannot overcome the problem of accurate tracking of the end of the actuator in a complex interference environment. Aiming at the above mentioned problem, a method for detecting and tracking the end position of the actuator based on the semantic segmentation model is proposed. Firstly, an end-to-end semantic segmentation model of the actuator image is built. Secondly, the contour inflection point detection algorithm is used to track the end position of the actuator in the segmented mask image. In order to further improve the tracking accuracy and robustness of the algorithm in a complex environment, 2D Kalman filter algorithm is used to process the occlusion situation, and the position tracking is realized when the end of the actuator is occluded. The experiment results show that the semantic segmentation model can achieve the segmentation accuracy of 62. 4% for the actuator, and the maximum average error of tracking the end position of the actuator in a complex environment is 1. 51 pixels, which provides a basis for improving the manipulation accuracy of the micro end effector.

Abstract:In close range industrial photogrammetry, according to the requirements of precise positioning and accurate recognition of circular coded point, a robust recognition algorithm is proposed. After preprocessing the target image, the algorithm firstly locates the coded point initially according to the edge filtering criterion and segment the feature region containing coding band. Then, the least square method is used to perform affine transformation on the feature region and the degenerated ellipse in perspective projection is mapped into a regular circle. Finally, the average pixel method is used to obtain the decoding value of the circular coded points, which further improves the anti-noise performance of the circular coded points. A large number of experiment results show that this method can not only achieve sub-pixel level in the center location of circular coded points, but also improve the robustness to the recognition angle of the algorithm. When the recognition angle is 70°, the correct recognition rate can still reach 97. 9% , which has good practical application value in actual complex scene.

Abstract:To achieve multi-target tracking tasks in complex water environment, a multi-target tracking method based on slope constraint and backtracking search strategy is proposed. Firstly, the threshold detection method is used to extract targets based on the bearing-time recording (BTR) data and the underwater kinematics analysis. Then, a novel hypothesis generation rule is proposed under the frame of traditional multi-hypothesis tracking (MHT) algorithm, which could be termed as slope constraint and measurements sharing. When the trajectory is interrupted, the retrospective search method is used to determine the interruption starting track point of targets. In addition, the cubature Kalman filter (CKF) is utilized to predict and compensate the interrupted trajectory. The hypothesis generation result is reduced to optimize the space complexity of algorithm. Experimental results show that this strategy can complete tasks such as multitarget automatic association tracking, interrupted track automatic prediction, and automatic track termination. The the average root mean square error of target tracking is 0. 594 4°, and the average running time of the algorithm is 0. 826 5 s.

Abstract:Aiming at the problems of low efficiency and low detection accuracy of parts defects in forging manufacturers, an improved EfficientNet model (EfficientNet-F) is proposed to detect the fluorescent magnetic particle flaw detection images of two kinds of forgings. A deep learning model with EfficientNet as the backbone feature extraction network is formulated, and the feature pyramid network is introduced as the feature fusion layer to improve the multi-scale feature fusion ability of the model. Complete intersection over union and attention mechanism are utilized to improve the robustness and detection efficiency of the model. Meanwhile, the fluorescent magnetic particle flaw detection image acquisition platform and the defective sample data set are both established. Experimental results show that the mean average precision precision of the optimal model of EfficientNet-F on the test set reaches 95. 03% . The F1 score is 0. 96 and the floating point operations is 1. 86 B. Compared with other deep learning models, the proposed method improvec the detection accuracy and efficiency. It can meet the needs of relevant production enterprises.

Abstract:This article presents a sensing method of the reflective absolute nanometer time-grating displacement sensor. The reflective single-row sensor is used as the precision measurement part of the reflective absolute sensor, which is marked as sensor A. To achieve absolute displacement measurement, a reflective single-row sensor with a period of difference from sensor A is designed, which is marked as sensor B. The phase difference between sensor A and sensor B is utilized to achieve absolute displacement measurement. The sensor prototype is made by using standard printed circuit board technology. An experimental platform is established, and evaluation experiments are carried out. Results show that the lead method of the excitation electrode brings interference to the receiving electrode, which causes the first-order harmonic error. To suppress error, a cross-reflective structure and a time-division method are proposed. The cross-reflective structure connects the sensing electrode with the reflective electrode lead at the other end to increase the distance between the excitation electrode and the receiving electrode. The time-division method applies an excitation signal to sensor A and sensor B at different time. In addition, it connects the inoperative electrode to the ground. Experimental results show that the structure and method cooperate with each other to effectively suppress the interference. Finally, within the range of 400 mm, the measurement accuracy of ±300 nm is achieved after compensation.

Abstract:Aiming at the problem that when a single excitation AC electromagnetic probe is used to detect cracks in arbitrary direction, missing detection happens easily, a novel dual excitation sensor for detecting the oblique cracks on the surface of Austenitic stainless steels is designed. A simulation model for the detection of surface cracks of Austenitic stainless steel with alternating current field measurement is established. A metal plate surface crack detection system based on a novel double-excitation sensor and high-resolution TMR sensors was developed. Based on the theoretical model and experiment system, the influence rule of crack orientation and crack width on the magnetic field component amplitude picked up by the sensor are studied. Simulation and experiment results show that using the distortion characteristics of magnetic field components Bx, By, the tiny cracks with surface dimension of 15 mm×0. 2 mm and depth greater than 3 mm can be detected with the same sensitivity, and the determination of the crack direction can be achieved. After the width compensation parameter is introduced the determination errors of transverse and longitudinal cracks are reduced, and the maximum determination error is 3. 9°.

Abstract:To realize the accurate measurement of mT magnitude magnetic gradient tensor in small space for traveling wave main magnetic flux leakage of high-speed maglev track stator, the Prewitt gradient operator of image edge detection is instantiated. A small volume and low-cost Hall magnetic gradient tensor measurement structure including 8 Hall magnetic sensors is designed. It overcomes the characteristics of large zero drift and poor sensitivity consistency of the Hall magnetic sensor by spatial redundancy, which provides a new way for sensors spatial layout design of the magnetic gradient tensor system. Compared with the traditional cross and square structure, experimental results show that the Prewitt structure can reduce the measurement error caused by the inconsistency of the above two parameters of multi-sensors to - 10. 4% and 58. 1% . The requirements of the measurement system for the consistency of sensor parameters is effectively reduced, and the complex and high-cost calibration of the three-axis sensor and error correction are avoided.

Abstract:The thin-film pressure sensor in prosthetic tactile measurement has problems of low accuracy, poor linearity and high hysteresis. To address these issues, a tactile sensor based on the fiber Bragg grating (FBG) is studied. It can be easily installed on the fingertip. Firstly, a sensing unit is designed to convert the external force into an axial one needed by the FBG. Secondly, the finite element simulation is utilized to optimize the structure of the sensing unit. In this way, the sensitivity of pressure measurement is improved. Finally, the FBG tactile sensor is prepared and stuck to the prosthetic fingertip. Experiments on performance calibration, comparative analysis, and grasping application are carried out. Results show, for the FBG tactile sensor, the sensitivity is 0. 103 8 nm/ N, the linearity R 2 is 0. 998, the repeatability error is 1. 32% , and the hysteresis error is 2. 19% . Compared with the thin-film pressure sensor, the designed FBG sensor has advantages of linearity, stable repeatability, and hysteresis.

Abstract:The sensor nodes of existing wireless sensor systems require actively generated radio frequency electromagnetic waves for data transmission, which leads to high power consumption. Passive wireless sensor systems that use backscattering modulation technology require high-cost dedicated readers, which results in limited applications. To solve these problems, this article proposes an ultra-lowpower wireless sensor system that manipulates the Wi-Fi channels. The designed system modulates the sensing data into the signals transmitted by the Wi-Fi device using backscattering modulation, which affects the parameters of the Wi-Fi channels. The system achieves ultra-low-power wireless sensing without a dedicated reader by using a Wi-Fi receiving device to obtain channel state information and training a machine learning detector capable of demodulating sensing data through the support vector machine. The wireless temperature sensing system is established using a router, a wireless network card, and a wireless sensor node. The maximum power consumption of the wireless sensor node is less than 350 μW. The Wi-Fi receiving device that employs a wireless network card can read the sensing data transmitted at 200 bit / s when the wireless sensor node is within 5 m of the router. The system provides an effective solution and satisfies the demand for ultra-low-power wireless sensors in smart homes, wearable devices, and other applications.

Abstract:Aiming at the problem of serious vibration interference effect in checkweigher measurement process, a new anti-vibration method of checkweigher using dynamic neural network system identification based on nonlinear auto regressive with exogenous inputs (NARX) model is proposed. The vibration characteristics of the checkweigher system is estimated through the redundant distribution of acceleration sensors. Then, combined with the weighing sensor error caused by the vibration interference in the case of no-load transmission, the dynamic neural network is used to automatically identify the vibration interference signal, and the vibration signal analysis model is established, which is used to match and eliminate the vibration interference in the dynamic weight detection signal. In resonance state, the proposed method is compared with traditional anti-vibration methods such as moving-average filter and adaptive notch filter methods in simulation and experimental. The result proves that the dynamic weighing anti-vibration performance of the multiacceleration sensors is superior. Finally, the operation speeds of 2 m/ s and maximum weighing of 200. 0 g are achieved, which meets the requirements of establishing the national standard “GB/ T 27739- 2011 Automatic Catchweighing Instruments” category XIII scale.

Abstract:Traditional acoustic black hole (ABH) changes the impedance by power law cutting of the material to achieve elastic wave focusing or control. But, at the same time, the material structure is destroyed, the rigidity of the material is reduced, and its application is limited in the field of elastic wave detection as a sensibilization structure. In this article, an upside-down adhesive ABH structure is proposed. In a certain range, its thickness change conforms to the power attenuation law, and it is small and easy to attach to the surface of the object to be measured without damaging the object. Finite element simulation results show that the inverted ABH has about 7 times of elastic wave amplification effect near the center of the circle under medium and high frequency excitation. Scanning test with the laser Doppler vibrometer shows that the inverted ABH has good elastic wave focusing effect and can amplify the elastic wave about 2-4 times. fiber Bragg grating(FBG) is bonded on the surface of the inverted ABH and the intensity demodulation is utilized to form an elastic wave sensor with focusing and amplification effect, which has important application value in the domain of nondestructive testing.

Abstract:On-line accurate measurement for the polished rod load of the beam pumping unit is a key technology for well pump indicator drawing and operation analysis. The precision of indirect measurement methods, such as vibration of rod liquid column and motor parameters, is low. The main problems include poor engineering application and poor stability of the load sensor direct measurement method affected by aging fatigue of material. To address these issues, a new method for soft-sensing of the polished rod load using input electrical parameters is proposed. This method firstly formulates a new mathematical model of the pumping unit′s torque factor by directly replacing the crank angle with the beam inclination angle. Secondly, it establishes the correlation among the motor input electrical parameters, the parameters of the pumping unit′s four-link mechanism and the pumping unit polished rod load. Finally, the collected motor input electrical parameters are substituted for calculation, the singular mutation values are eliminated by the Chauvenet Criterion, and the mean filter method is used to further improve calculation. After modification, the on-line measurement of the polished rod load is obtained. Engineering experiments and application results show that the method has an average relative error of 3. 87% , which shows the features of high accuracy, good stability and strong engineering practicability.

Abstract:Aiming at the problems of time-consuming, low accuracy and high energy consumption existing in the detection of the state of health (SOH) of the decommissioned power battery at current stage, this paper proposes a method to quickly predict the state of charge (SOC) and SOH based on electrochemical impedance spectroscopy (EIS). Through testing and analyzing the EIS of decommissioned lithium iron phosphate power batteries in different SOH at different SOC and different temperatures, an EIS equivalent circuit model is constructed. Then, the relationship between the constant phase component parameters and the SOC and SOH of the decommissioned power battery is used to establish a mathematical model to realize the rapid estimation of the SOC and SOH of the decommissioned power battery. Verification experiments show that using this method can greatly reduce to less than 20 min, save energy source and achieve rapid estimation of the battery with unknown SOC and SOH, and the prediction error is within 4% .

Abstract:The traditional multi-surface wavelength tuning phase-shifting interferometry can only achieve the measurement of transparent objects in fixed positions, and the effectiveness and solution accuracy of the algorithms are not intensively discussed and analyzed when the phase-shifting value or cavity length is changed. To achieve the simultaneous measurement of various surfaces of the transparent objects at arbitrary cavity lengths, a weighted multi-step sampling algorithm that can achieve adaptive phase shift matching is proposed in this paper. Firstly, based on the characteristic polynomial, a weighted multi-step phase demodulation technique with phase-shifting error and coupling error suppression capability is introduced, which can measure three surfaces of a transparent object simultaneously. Then, based on the Zernike polynomial, the algorithm solution error is iteratively calculated and analyzed for the arbitrary combinations of the phase-shifting reference coefficient and cavity length coefficient to provide a reference for the optimal parameter matching. The simulation results show that the residual error of the proposed algorithm solution is less than 3×10 -4 λ0 . Besides, the 3D height parameters in the ISO-25178 series are introduced to verify the algorithm performance. In the experiments, a Fizeau interferometer was used to measure the transparent flat plates with thicknesses of 10 and 5 mm, respectively at different cavity lengths. The simulation and experiment results show that the proposed algorithm can achieve multi-surface measurements at arbitrary cavity lengths and thicknesses, and its reliability and practicality are verified.

Abstract:Magnetic particle imaging is a new type of tracer imaging technology that uses the nonlinear magnetization characteristics of magnetic particles in magnetic field-free to image the measured objects. The spatial resolution is determined by the fineness of magnetic field-free, and the fineness is determined by the spatial magnetic field gradient. To improve spatial resolution, the static magnetic field structure which can produce large magnetic field gradient is combined with the driving structure to form a magnetic field-free line system. Firstly, this article designs static magnetic field structure based on ring magnet array. The large gradient static magnetic field is used to construct fine magnetic field-free line. Secondly, the driving structure is designed, which is based on Helmholtz coil and its driving method. The relationship between magnetic field-free line scanning range and driving current is determined. Finally, the spatial resolution of the magnetic field-free line system is calculated and its regularity with the magnetic field gradient and particle sizes is studied. Experimental results show the magnetic field gradient generated by the static magnetic field structure based on ring magnet array is 4. 804 T/ m, and the spatial resolution is 0. 540 mm when the 30 nm magnetic particle is used. The magnetic field-free line can scan within 30 mm range. It proves the feasibility of magnetic field-free line system based on ring magnet array for high-resolution magnetic particle imaging.

Abstract:There are various system errors in the inertial measurement unit ( IMU). It needs calibration to enhance the accuracy of attitude calculation. The correction matrix is usually determined by the measurement values and their theoretical values at some specially designed positions where the measurement values of sensors also contain various system errors. The error caused by the correction matrix is deduced theoretically, and its influence on the error of attitude parameters is studied. Experimental data are utilized to verify those influences. According to the point design of balance correction, the influence models of balance correction and its structure topology on attitude accuracy are researched in detail. The balance correction is again implemented on the original experiment data. The results at some classic attitude positions show that the balance correction generally reduces the absolute error. The average and peak values of absolute attitude error are decreased to 30. 8% at least with the average balance correction. Moreover, the attitude accuracy of small inclination angle with the average balance correction occupies the similar attitude accuracy under large and medium inclination angles with the traditional correction method.

Abstract:Travel error is an important parameter to evaluate the accuracy of planetary roller screw pair, but there are a few studies on the influencing factors and modeling of travel error. Therefore, firstly, the influencing factors of travel error of planetary roller screw pair are analyzed, including machining error, installation error and deformation error. Then, based on the error conversion, the travel error model is established and the errors are measured. Finally, the travel errors of four planetary roller screw pairs and the travel errors of planetary roller screw pairs under six different load conditions were measured with the travel error test bench. The test results show that the machining processing level and installation accuracy have a great influence on the travel error index, and the travel error of the screw pair increases with the increase of the load, which indicates that the deformation error caused by the increase of the load has a great influence on the travel error. The relative error between the test value of the screw travel error and the calculated value of the model is 1. 62% ~ 4. 37% , and the relative error of the screw pair is 1. 81% ~ 6. 53% . The effectiveness of the model is verified, the study can provide a reference for the transmission accuracy analysis of the feed system.

Abstract:In chip manufacturing, intelligent manufacturing, aerospace and other fields, precise plane positioning urgently needs synchronous independent precise measurement of two-dimensional displacement. In this article, a planar displacement sensor based on the principle of electromagnetic induction is proposed, which consists of a moving front and a fixed front. The fixed array includes m×n planar spiral coil arrays in series. The planar standing wave magnetic field array is generated on the measuring plane when 4 kHz AC current is applied. The moving front is arranged by four spiral coils in the form of 2×2 matrix, and four-channel modulation signals whose amplitude changes with the displacement of x axis and y axis are induced. The Cordic algorithm is used to solve the displacement of two dimensions. This article first introduces the structure and working principle of the sensor. The finite element analysis is implemented to the electromagnetic model, and numerical simulation is applied to the dislocation algorithm. According to simulation results, the measurement error is analyzed and traced, and the sensor structure is optimized. The sensor prototype is made and the experimental verification is carried out, which verifies the feasibility of the decoupling method of sensor structure and position shift. The maximum error of the sensor in the counter pole is 48. 7 μm. And the sensor resolution is 0. 317 μm. Experimental results show that the linearity of the sensor reaches 0. 15% within the range of 147 mm×147 mm, which provides theoretical support and experimental guidance for the further development of the high-precision two-dimensional time-gate displacement sensor.

Abstract:Temperature is the important function that affects the normal function of human body. Blood temperature change monitoring is particularly important for the assessment of human health. Based on the proposed extended HANAI theory of the dense dispersion system, the influence of temperature on the impedance characteristics of blood has been quantitatively evaluated by bioelectrical impedance spectroscopy measurement, which provides a feasible quantitative characterization basis for blood temperature and related blood disease monitoring. Experimental results show that, under the condition that the temperature T increases, the impedance Z ∗ decreases continuously. It founds that the resistance Rc and reactance Xc have great linear relationship with the temperature T, which could be described as Rc = -0. 3T+24. 14 and Xc = 0. 08T-5. 78, respectively. The tranditional HANAI equation for dense solution is extended. The numerical analysis and simulation results show great agreement with experimental results which confirm the fact that the conductivities σp of plasma and the conductivities σc of cytoplasm are responsible for the influence of temperature on blood impedance characteristics. This study indicates a potential non-invasive temperature measurement method for blood or other solutions, which has certain scientific research and application value.

Abstract:The development of the latest generation of electronic industry has requirements of wearable, portability, comfort, etc. In recent years, the design and fabrication based on flexible actuators have become the state of the art in the field of commercial consumer electronics and scientific research. Compared with rigid actuators, flexible actuators are more portable, flexible, and easier to be conformal to human skin. With these characteristics, flexible actuators show the great potentials in human-machine interfaces, such as creating a variety of haptic feedbacks which mimic the sense of real touch (knead, press, and pull). In addition, the whole devices can be made with portable and cost-effective process. The research and development of flexible actuators used for haptic force feedback are analyzed from the functional materials and their correspondingly working principle in this review. Different working mechanisms are discussed based on their advantages and disadvantages of flexible actuators. In addition, the paper concludes the potential applications of flexible actuators in augmented reality, virtual reality, education, and medical assistance, etc. The existing difficulties of flexible actuators used to achieve haptic force feedback are also summarized. Flexible actuators with safe, comfortable, portable, beautiful, and quiet characteristics are key features to be merged with future human-machine interfaces.

Abstract:The surface electromyography signal can reflect the user′s action intention. Therefore, it becomes the main control signal for human-computer interaction. However, the individual variability makes the user model universally un-applicable, which is not conducive to the development of the universal EMG equipment. In this paper, from the perspective of neural synergy control, muscle synergy is extracted by the non-negative matrix factorization algorithm. Then, the pre-experimental data of new user are combined with least squares to obtain training synergy as a feature quantity, which is similar to pre-experimental synergy. For application consideration in low-frequency wearable scenarios, three simple and easily portable classifiers ( i. e. , support vector machine, error back propagation network, and K-nearest neighbor algorithm) are trained and tested, respectively. Four sets of gesture recognition experiments are implemented in DB1 ( 100 Hz) and DB5 ( 200 Hz) of the Ninapro database. The average recognition accuracy rates are 81. 12% , 78. 19% , 74. 07% , 60. 11% ( DB1 ) and 85. 75% , 83. 25% , 79. 07% , 66. 10% ( DB5 ), which are higher than the existing low-frequency online recognition algorithms by more than 10% . The proposed algorithm is simple and easy to train the classifier using existing user data and a small amount of pre-experimental data from new users. Meanwhile, the action intention can be judged from the perspective of neural coordination, which is more conducive to the development of a control method that conforms to the natural movement of the human body. It provides a feasible solution for the popularization of wearable electromyography equipment.

Abstract:The “black flying” incidents of unmanned aerial vehicles (UAVs) in airports have frequently happened across the country. As the main interference for the detection and identification of UAVs, the characteristics analysis for the echoes of flying birds is of great significance. A fined modeling method of bird echoes is proposed, which considers the actual electromagnetic scattering characteristics. The multi-layer fast multipole method is used to calculate the radar cross section (RCS) of birds in three typical flapping wing attitudes. The flapping wing attitude is calculated and combined with the signal propagation model for the array antenna. Then, the fined modeling of the bird echoes for each channel at any observation point is realized. Experimental results show that the RCS of birds varies greatly with different observation angles. The RCS value corresponding to the side irradiation is higher than that corresponding to the front and back irradiation. The fine modeling of bird echoes can demonstrate the fine description for the bird targets, which can provide the reference to study the echoes of birds.

Abstract:The low frequency response of oil paper insulation is very important for the diagnosis of damp and aging of insulation. In view of the lack of analysis methods for low-frequency response data and the imperfection of the corresponding equivalent circuit model, the Kramers-Kronig transform is used to separate the conductance, infinite frequency capacitance and relaxation process in the frequencydomain dielectric spectrum of oil paper insulation. The Dissado-Hill model based on multi-body theory is introduced to fit the lowfrequency relaxation process. A new oil paper insulation model is formulated, which is the equivalent circuit model of paper insulation for low frequency response. Finally, the low frequency response data of oil paper insulation samples with different moisture content are modeled and analyzed. Experimental results show that the Kramers-Kronig transform can effectively decouple the relaxation process, conductivity process and infinite frequency capacitance of oil paper insulation. The goodness of fit of reconstructed spectral lines of the low-frequency response model for samples with different moisture content is more than 0. 98, which evaluateds the accuracy of the model. There is an exponential function relationship between the amplitude of low-frequency diffuse polarizability χ LFD ( 0) and micro water content. The goodness of fit is as high as 0. 99, which can be utilized as a new characteristic quantity for quantitative evaluation of moisture content of oil paper insulation. It provides a new idea for diagnosis of moisture affected state of oil paper insulation.

Abstract:Electrical impedance tomography ( EIT) provides an effective method for monitoring the spatial features of human lungs because of its non-invasiveness and visualization natures. However, the inverse problem of EIT has serious non-linearity, ill-posedness and indeterminate feature, which makes the reconstructed images contain serious artifacts. Aiming at the above problems, a deep network imaging algorithm of V-ResNet composed of pre-mapping module, feature extraction module, deep reconstruction module and residual denoising module is proposed in this paper, which achieves the reconstruction of the spatial position and conductivity parameter distributions of the field. This algorithm can effectively increase the feedforward information by multiple transmissions and solve the phenomenon of gradient disappearance in deep networks. Meanwhile, the residual denoising module is utilized to effectively smooth the image boundary. The relative error (RE) and structural similarity (SSIM) are used to evaluate the imaging quality, and the experiments show that the average RE is 0. 14 and the average SSIM is 0. 96. The results of the simulations and experiments illustrate that compared with traditional imaging algorithms, the imaging algorithm based on V-ResNet achieves clearer boundaries and higher resolution in imaging result.

Abstract:The accuracy of prior information is a key element to ensure accuracy and reliability of the collaborative navigation system. The unknown and time-varying noise will be generated by external disturbances in a complex environment. To address this issue, a variational adaptive cooperative navigation method based on belief propagation is proposed. Firstly, based on the basic model of sigma point belief propagation (SPBP) cooperative navigation, the forward filtering process of cooperative navigation based on the confidence propagation mechanism is completed. The process noise and measurement noise are treated as the prior information of Bayesian estimation by IW (Inverse-Wishart). Then, the forward filtering value is used to establish a sliding window to smooth the noise variable to solve the filtering accuracy decline caused by the time variation of noise. Compared with that of the SPBP algorithm without smoothing operation, simulation results show that the position error of the slide window variational adaptive sigma point-belief propagation (SWSP) algorithm with smoothing operation is reduced by 90% due to the noise time-varying. The accuracy is much close to that of the opt SPBP algorithm.