Abstract:In order to solve the defects as low positioning accuracy of weighted least squares method, sensitive to initial position and fixed noise covariance of kalman filter, a pseudo-range single-point positioning method combining weighted least squares and improved Kalman filtering is proposed. This method first uses the weighted least squares to calculate the initial receiver position, then uses this position as the initial value of the improved adaptive Kalman filter, and finally establishes a dynamic model for filtering. Experimental results show that compared with traditional Kalman filtering, adaptive Kalman filtering based on moving window covariance estimation can improve the accuracy of single-point positioning by 50% and the convergence speed by 90%. The algorithm can be used in civil navigation and positioning with little high accuracy requirements.

Abstract:Taking the quad flat no-lead (QFN) package chip as the experimental object, proposes a fast image correction method for the semiconductor chip to solve the problem of the chip's image tilt during visual detection of packaging defects. Firstly, an improved Harris corner detection algorithm is proposed to extract the corner vertices of the target contour in combination with the polygon approximation method; secondly, the line fitting on the vertex of the longest edge is conducted by the least square method; finally, the image correction is carried out according to the line fitting results. The experimental results show as follows: compared with the traditional Hough transform, the minimum enclosing rectangle, and the Fourier transform correction method, the tilt angle obtained by the proposed method is more accurate, and the running time is less than one fifth of the above-mentioned traditional methods. Namely, it owns faster speed and higher efficiency.

Abstract:In the existing resonant accelerometer, it cannot be used in high-precision guidance and air attitude fine-tuning because of its small resonance frequency and low sensitivity. For this reason, a resonant accelerometer based on nano piezoelectric beam is designed, which uses the upper and lower tuning fork resonators (the resonant beam uses the zinc oxide with a diameter of 500 nm) to distribute symmetrically with the central mass block and the left and right support beams, thus realizing low cross coupling and high sensitivity output. The mathematical model of the accelerometer structure is analyzed and established in ANSYS. It is analyzed under the simulation platform of workbench: the resonance frequencies of the upper and lower resonators are 2. 987 93 and 2. 987 29 MHz respectively, and the displacement in the X direction under the resonance frequency is two orders of magnitude higher than that in the other Y and Z directions; the maximum stress of the accelerometer under the action of 2 000g acceleration load is 241. 46 MPa; in the design range of ±10g, the sensitivity of the structure is 1. 133 11 kHz/ g. Based on SOI technology, the process flow of nano piezoelectric beam resonant accelerometer is designed to verify its correctness.

Abstract:When RSSI is used to locate unknown node of wireless sensor network, the RSSI values are easily affected by environment will cause location error. Thus, quadrilateral weighted centroid localization algorithm based on range correction of RSSI ( QWCRC) is proposed. Firstly, the optimized RSSI value is obtained by Kalman filtering of the received RSSI values, which makes the ranging as close as possible to the real distance. Secondly, location of unknown node is determined by quadrilateral weighted centroid localization algorithm, at the same time, the method of least squares is used for auxiliary positioning. A new solution is provided by the algorithm to the case where the adjacent anchor node circle does not intersect. Finally, the experimental results show that compared with the quadrilateral weighted centroid algorithm ( QWC) and the triangle weighted algorithm modified by RSSI ranging ( TWCRC), the positioning accuracy of the improved algorithm can be improved by 87. 14% and 35. 51% respectively when the number of anchor nodes is 5×5 and the noise intensity is 0 dbm.

Abstract:In view of the current image restoration algorithms mainly use the R, G, B information of the image to obtain the optimal matching block, which ignoring the structural features of the image and resulting in the problem of texture discontinuity and block phenomenon in the repaired image, this paper designs an image restoration algorithm with variance adjustment strategy coupling structural features. Firstly, the gradient modulus of image is used to construct the structure measurement factor to measure the structure characteristics of image. The priority function is constructed by combining the confidence term, structure measure factor and data term to find the priority repair block. Then, the variance feature of the image is used to establish the variance adjustment strategy, considering the dynamic changes of the image texture, to find the most consistent sample block size with the current texture situation. Finally, the structure measure factor is introduced into the search process of the optimal matching block to make up for the lack of the ignored image structure features when searching the optimal matching block through R, G, B information, accurately obtain the optimal matching block, and achieve the repair of the damaged area. The experimental results show that the algorithm has better texture continuity and visual effect than the current algorithm, and the performance is better.

Abstract:Aiming at the problem that the permanent magnet of the six-phase permanent magnet synchronous motor (SP-PMSM) is prone to demagnetize under complex operating conditions, a reconstruction method of permanent demagnetization fault is proposed based on the super-twisting algorithm. Firstly, the mathematical model of the demagnetization fault for SP-PMSM is constructed based on the vector space decomposition (VSD) theory through order reduction and decoupling. Secondly, taking the stator current as the state variables, the sliding-mode observer is designed using the super-twisting algorithm. The real-time reconstruction of rotor flux is realized according to the principle of sliding mode equivalence. A kind of strong quadratic Lyapunov function is used to ensure the stability of STA-SMO. Finally, the simulation and experimental results demonstrate the effectiveness of the proposed method. Compared with the traditional sliding mode observer (SMO), the STA-SMO can reconstruct the demagnetization fault accurately, reduce chattering effectively and has a good robustness.

Abstract:Aiming at the problems of insufficient data reliability and resource waste in the decision of redundant data of UAVs, a compensation algorithm for UAV IMU multi-sensor redundancy based on BP neural networks is proposed. The low-precision IMU sensor data is input to the BP neural network, and the non-linear fitting capability of the BP neural network is used to compensate for errors in low-precision IMU data, then use data arbitration algorithm based on confidence to arbitrate multiple higher-precision data and output the sensor data after data fusion. This process can also judge and locate sensor faults. The singularity problem can be solved by changing the installation method of similar sensors. The experimental results prove that after neural network error compensation, the error is reduced by 55. 2%. Furthermore, the error after neural network error competition is 53. 9% smaller than the error after using the kalman filter algorithm for error compensation. The algorithm takes full advantage of redundant sensor design, improves the reliability of the sensor system.

Abstract:In order to solve the problems of low data transmission quality and low network transmission bandwidth in the process of data transmission of mobile Internet of things, a stable scheme of data transmission of mobile Internet of things based on node link evaluation model is proposed. Considering the characteristics of node movement, a new node link evaluation model was designed to realize multidimensional evaluation of data transmission process. Then, according to the node link evaluation model, three indexes of data transmission order, relay link control and transmission energy control were proposed. By matching the transmission power of nodes, a link stability method based on the order degree of data transmission was designed, and the poisson distribution model was used to build a node stability method based on the control degree of relay link, and by predicting the limited performance of nodes, a node failure state transmission energy saving scheme based on the control degree of transmission energy was proposed, so that the data transmission performance of the mobile Internet of things was optimized from node, link and energy, which can effectively improves the stability of the network transmission. The simulation results show that compared with the data control synchronous transmission algorithm based on the likelihood estimation compensation mechanism and LTE-5G data transmission algorithm based on the pre transmit precision improvement mechanism, this algorithm has lower congestion level and node limited probability, and higher network transmission bandwidth.

Abstract:In order to improve the accuracy of driving deviation measurement and intervention, the strategy of two-parameter deviation measurement and intervention is proposed. VBAI was used for image processing, and the optimal binarization segmentation threshold under different environments was automatically solved by using the maximum inter-class variance method. The gray acquisition Lines ware designed to obtain the lane edge points, and the Fit Line algorithm was used to fit the edge points to complete the lane line recognition. According to the prediction model, two parameters of the relative deviation angle and the ratio of pixel distance ware analyzed, and the different intervention results of prompting or alarming ware displayed on the interface. Measured by the road, the alarm accuracy was over 97. 7%, and the processing speed was higher than 1 / 42 s/ frames. The system has practical application value to improve driving safety and reduce traffic accidents caused by deviation.

Abstract:In order to solve the problems of slow dynamic response and weak robustness of the integer order PI controller in the double closed-loop vector control system of permanent magnet synchronous motors, an off-line parameter tuning method of outer speed loop and inner current loop of the system is proposed by using dragonfly algorithm and fractional order PI control. The parameter to be optimized is regarded as the spatial position of the best individual that the dragonfly searches for the food source in the search space, and the error performance index ITAE is used as its target fitness function. Simulation and experimental comparisons of motor speed regulation performance of traditional engineering experience tuning integer order PI, dragonfly algorithm for integer order PI, dragonfly algorithm for fractional order PI, and particle swarm algorithm for fractional order PI, respectively. The results show that the dragonfly algorithm optimized fractional-order PI controller has the advantages of improving the system′s dynamic response performance, reducing the amount of overshoot, and enhancing the robustness, proving the superiority of the optimization strategy.

Abstract:To solve the problem that the daily pulse signal is susceptible to movement during the detection process, which causes the heart rate cannot be accurately monitored. A method to improve the accuracy of heart rate monitoring based on the fusion analysis of pulse signals behind the ears and the fingers is proposed. Use the difference between time and degree of pulse information of different parts affected by exercise to achieve information complementation. The pulse signals behind the ears and the fingers are detected simultaneously, and it is judged whether they are usable segments through quality assessment. When the signal quality is reduced due to motion interference, a better-quality pulse signal is selected for analysis by the information fusion method to ensure the accurate detection of the heart rate. The experimental results show that compared to using only a single-site pulse signal, the accuracy of peak detection after fusion analysis of the pulse signals behind the ear and the finger has been improved from 90. 2% to 98. 8%. The proposed method based on pulse information fusion has better ability to resist exercise interference and provides an effective way for the accurate detection of heart rate in daily life.

Abstract:Dynamic compressive sensing is an extension of traditional static compressive sensing to dynamic signals, which has a wide application in MRI, video compressive sensing and target tracking. Since dynamic signals are usually sparse in some transformed matrices and change slowly with time varying, an underdetermined measurement matrix can be used to compress the signals. The research of dynamic compressive sensing mainly focuses on three parts: Sparse representation of dynamic signals, dynamic compressive measurement, and reconstruction of dynamic signals. A comprehensive survey about dynamic compressive sensing is given in this article. At first, the basic concept of dynamic compressed sensing is introduced, which includes several mathematic models of dynamic signals, sparse dictionary learning algorithms and methods of adaptive measurement. Secondly, we classify the reconstruction algorithms into two main parts: Least square based algorithms and Bayesian algorithms, and we also introduce some representative algorithms in detail from each part. Finally, several applications of dynamic compressed sensing are introduced, and we provide a reference for further investigation on reconstruction algorithms.

Abstract:In order to distinguish the functional connectivity on different brain areas between two mental states of lying and telling-truth and to research this functional connectivity change in time-frequency domains,forty healthy right-handed subjects with an average age of 21 were randomly divided into two groups (20 each): Lying and telling-truth. Through standard three stimuli paradigm, we recorded the 12 channels electroencephalogram ( EEG) signalsin two states. Then, used wavelet coherence method to calculate the coherence coefficient on66 pairs of channelsof the following time-frequency bands:θ(0. 5 ~ 4Hz)、δ( 4 ~ 8Hz)、α( 8 ~ 13 Hz)、β( 13 ~ 30 Hz)、γ (30~ 100 Hz) and the time range of 250 ~ 1 300 ms after the stimuli ( typical occurrence time of P300). Analyzed the functional connectivity on different channels pairs in different time-frequency areas. Finally, Wilcoxon test was used to compare the difference of wavelet coherence on the same time-frequency domain between the two groups of the subjects. The experimental resultshows that in the time-frequency domain corresponding to frequency bands θ and δ, there were statistical differences in the coherence values of OZ-P4,OZP3 and P3-P4of the two groups of subjects. The finding indicates that when lying with physicalevidence, the associative visual cortex(P3 and OZ) and inferiorparietallobule ( P4) may be activated during utilitarian and nonutilitarian moral judgments, yielding significant statistical differences in the functional connectivity between different brain regions.

Abstract:Automatic modulation recognition (AMR) can automatically estimate the modulation type under the condition that the signal is unknown at all. A deep learning based AMR is proposed. The proposed method can update the filter taps through waveform optimization, which can filter the signal samples in order to overcome the unfavorable effects of the transmission channels. In the proposed method, a feedback path exists between the recognition network and the inverse-channel filter. According to the experiments from an open-source dataset, the proposed feedback-structured method can increase the recognition rate compared with the traditional deep learning methods. Specially, compared with the CNN based method, the recognition rate has increased by about 7%.

Abstract:Insulated gate bipolar transistor ( IGBT) is a key component of power electronic devices. Its high reliability is an important guarantee for long-term stable operation of the system. Fault monitoring of IGBT modules is one of the effective ways to improve system reliability. Presents a new health-sensitive parameter, gate-emitter pre-on voltage VGE (pre-on) , for monitoring IGBT chip failures in highvoltage multi-chip parallel IGBT modules. First compare existing fault monitoring methods, then build a pre-threshold voltage reliability model, and then detect the IGBT chip failure in the multi-chip high-voltage IGBT module by monitoring VGE (pre-on) during the gate-emitter voltage turn-on transient. In order to verify the feasibility of this method, a 16-chip DIM800NSM33-F IGBT module was simulated. The results show that under different external conditions, the average deviation of the pre-on voltage VGE (pre-on) generated by each parallel IGBT chip failure is about 900 mV, and has high sensitivity and anti-interference ability, which can effectively monitor the IGBT module chip failure.

Abstract:In view of the existing electromagnetic flowmeter conductivity measurement methods, there are problems such as low measurement accuracy, time-sharing of additional excitation and magnetic excitation, which affect the real-time flow measurement. This paper proposes a parallel excitation electromagnetic flowmeter conductivity measurement method. The single pulse square wave capacitive coupling manner is adopted in the excitation part, which can be with magnetic field excitation in parallel. The flowmeter signal and the electrical conductivity can be measured in the same time. The band-pass filter and the method based on energy attenuation are given, which improve the precision and stability of conductivity measurement. The flow velocity measurement shows that the conductivity and the flow can be measured simultaneously with parallel excitation, and the conductivity measurement is accurate to 3% within the range of 100 μS / cm~ 3 mS / cm, the flow calibration precision is better than 0. 5 level.

Abstract:In order to solve the problem as the target information is not prominent in the fusion results of many visible and infrared image fusion methods. The Laplace decomposition mechanism is introduced to fuse the visible and infrared images with the brightness characteristics of the image. With the help of Laplace decomposition method, the input image is layered to obtain different layer information. Using the mean value of the image, the brightness information of the image is calculated, and the fusion weight of the lowfrequency layer is adjusted adaptively to obtain a low-frequency layer with high integrity of the target information. Based on the spatial frequency characteristics of the image, the detail richness of the high frequency layer is evaluated to obtain a fusion high frequency layer with rich details. And the low and high frequency layers are fused by using the Laplace inverse decomposition method. Experimental data show that the fusion results of the proposed algorithm can highlight the target information and have more detailed features than the existing fusion algorithms.

Abstract:The demands for computing efficiency of digital image correlation (DIC) is increasing, a GPU-accelerated path-dependent DIC method is proposed. This algorithm uses FFT-CC algorithm to calculate the initial seed points and IC-GN method to carry out subpixel registration. Then, through the initial value transfer scheme, multiple seed points are generated to participate in the parallel calculation of sub-pixel registration, so that the calculation points can be rapidly spread until the region of interest is completed. Experiment results show that, the proposed method has a good accuracy and can obtain a clear surface deformation nephogram. It achieves a computing speed of 6. 5×10 5 points per second with a subset size of 17×17 pixels. Compared with typical high-speed DIC algorithm, the speed enhancement is over 50%. It provides reference for the design of high speed DIC algorithm.

Abstract:In order to deal with the problem that sensors in industrial production systems are prone to failure due to long-term exposure to harsh environments, a new method of sensor performance evaluation and fault diagnosis based on standard uncertainty is proposed. Taking the production process of square-tube as the research object, firstly, the magnetic scale and encoder are chose as the distance measuring sensor through the hardware test experiment, and the automatic adjustment system of the roll position is studied and designed, the roll positioning accuracy is improved within ±1 mm. Then, wavelet filtering is used to denoise the output sequence of the sensor, and the neural network is combined to predict the output sequence. Finally, the standard uncertainty is calculated by the residual of the predicted value and the actual measured value, the fault diagnosis of the ranging sensor and the evaluation of the real-time performance of the ranging sensor are realized by judging the size of the uncertainty. Simulation experiments on the faulty sensor with accuracy decrease showed that the fault detection accuracy of this method can reach more than 90%, and the proposed method is effective.

Abstract:The super-resolution reconstruction algorithm is applied to synthetic aperture radar (SAR) target recognition. The parameters of SAR images are estimated based on the attributed scattering center model for high-resolution reconstruction, thus generating the multiresolution representations of the original images. The joint sparse representation is employed to represent the original image together with its high-resolution representations, thus considering their inner correlations. Finally, the target label is determined based on the total reconstruction errors. The experimental results on the MSTAR dataset confirm the validity of the proposed method.

Abstract:This paper proposes a synthetic aperture radar (SAR) target recognition method based on image blocking and matching. The test SAR image is blocked into four patches, which are analyzed and matched separately. For each SAR image patch, the monogenic signal is employed to describe its time-frequency distribution and local details thus to construct the feature vector. The sparse representation-based classification (SRC) is used to classify the four monogenic feature vectors and produce the reconstruction error vectors. For the four reconstruction error vectors, a rich set of random weight vectors are used to fuse them and all the results are analyzed in a statistical way. Finally, the decision value is designed to determine the target label. The proposed method is tested on the MSTAR dataset. The results confirm the validity of the proposed method.

Abstract:For the problem of depression of the high sidelobes in synthetic aperture radar (SAR) images, a new way based on attributed scattering center model and spatially variant apodization ( SVA) is proposed. SVA is one of the classical super-resolution image processing technologies, which could keep the resolution of mainlobe while depressing the sidelobes. Attributed scattering center can properly depict the scattering properties of targets at high frequency region, which is an important tool to analyze the SAR image. Attributed scattering center is employed to do the parameter estimation and then the parameters are used to do the SVA filter. Experimental show that the effectiveness of the proposed method as for depressing the sidelobes.

Abstract:Foreign object debris is one of the important factors that threaten the flight safety of the aircraft. To carry out automatic detection research of foreign object debris is an inevitable requirement to effectively ensure aircraft flight safety. Focus on detection of FOD, firstly an image difference method based on DGPS was proposed, this method used RTK to obtain high-precision position information of the image of runway and aligned two images in the same place to meet the conditions of image difference. Histogram for specification was also developed to complete brightness correction. Then FOD could be detected by image difference. Finally, experimental verification was carried out. Experimental results show that the method can effectively detect foreign objects debris on airport runways with no less than 2 cm×2 cm.

Abstract:In order to solve the problem of low accuracy of traditional algorithm for train number identification of railway freight trains, Faster R-CNN neural network for train number location of railway freight trains is proposed. The detailed features of the final convolution feature map are enhanced by adjusting the relevant size parameters and connection mode of the feature extraction network. The k-means ++ clustering algorithm is used to calculate the length width ratio of the train number area. The improved anchor size design makes the target detection frame more suitable for the actual train number area. In the experiment, data augmentation and dropout are used to improve the robustness of the network. The results show that the improved Faster R-CNN network has achieved 93. 15% accuracy in the location of railway freight train number, 90. 76% recall rate and 91. 94% comprehensive F1 index. It also shows that this method can accurately locate the railway freight train number and provide reliable data support for the identification process.

Abstract:In order to effectively monitor whether the elderly fall, an external contour features extraction combining background subtraction based on human body boundary was proposed, which identify human movement behavior. Firstly, the background subtraction method was used to extract the moving objects from the video, and then the pictures of extracted moving objects were preprocessed. Secondly, the detection method of the minimum external rectangle and the center of gravity was used to extract the moving objects’ features, so as to obtain the overall external contour and the center of gravity position of the elderly. Finally, according to the different positions of human body, the movement model was established to effectively identify the movement of the monitored object, such as walking, falling, etc. The experimental results show that the algorithm proposed in this paper can effectively process the actual video, and the accuracy of human behavior recognition reaches 94. 3%.

Abstract:Being lack of the pre-earthquake reference information, a new method of damaged building detection of high-resolution remote sensing image based on optimized visual dictionary is proposed. Firstly, wavelet-JSEG (WJSEG) segmentation and a set of non-building screening rules are applied to extract the potential building set. Secondly, a visual dictionary model of earthquake damage is constructed by introducing spectral, texture and geometric morphological features to across the semantic gap between pixels and earthquake damage features. On this basis, a visual dictionary optimization strategy based on intra-class and inter-class penalty indexes is designed to further reduce redundant information and evidence conflict. Finally, the buildings are further classified into intact buildings, partially damaged buildings and ruins by random forest classifier. In two experiments, the overall accuracy of the proposed method reached more than 85%, which can provide key decision support information for post-earthquake emergency response and reconstruction.