Abstract:Aiming at the problem of low accuracy of the traditional load calibration equation for calculating wing skin load, a novel method of wing skin load calculation based on deep learning was proposed. Considering that the force of the real wing skin was complicated, this paper established a simplified wing structure model. Firstly, the finite element analysis on the wing was carried out by using Ansys software to obtain the strain and force simulation data, then the simulation data was cleaned and preprocessed. Secondly, a deep neural network model was constructed, its input and output were the strain and load values, respectively. The Adam optimization algorithm was used to optimize the model for load calculation. Finally, the test set was used to predict the load value, and the average relative error and absolute error were used as evaluation metrics. Experimental results show that the calculation method based on deep learning obtains the average absolute error of 0. 081 N for small load data and average relative error of 0. 063 8% for normal load data, respectively. The load accuracy of new method is obviously better than that of the traditional method comparing with traditional load calibration method.

Abstract:The principle of wavelet denoising and the rule of optimizing threshold function are studied, aiming at the problem that local oscillation and edge blur of signal after wavelet soft and hard threshold function denoising lead to poor denoising effect. An adjustable threshold function with continuity, flexibility and small constant deviation is designed. A wavelet denoising algorithm based on improved threshold function is proposed, which is applied to denoising signals containing Gaussian white noise. Experimental results show that compared with traditional methods, the proposed method has flexibility and applicability to simulated signals and ECG signals, and the signal-to-noise ratio of the signal after denoising is improved by 16. 21%, and the Pearson correlation coefficient is increased by 1. 62%. Therefore, the algorithm is feasible, which can effectively retain feature information, and the denoising effect is more ideal.

Abstract:In order to improve the accuracy of wind speed prediction, this paper starts from the predictability of mining wind speed data and optimizes the performance of the prediction model, and proposes a prediction model that combines adaptive noise complete empirical mode decomposition ( CEEMDAN) and long short-term memory neural network ( LSTM). First, CEEMDAN is used to reduce the instability of the wind speed sequence and improve its predictability. Secondly, a LSTM prediction model is established for each subsequence obtained by decomposition, and an improved cuckoo search algorithm (ICS) is used to optimize the key parameters of LSTM and improve its regression performance. Then use the optimal parameter LSTM prediction model for each sub-sequence to model and predict, and superimpose to obtain the wind speed prediction result. Verified by actual measurement data, the average absolute error and average relative error of the model proposed in this paper are only 0. 82 and 0. 95. The comparative study shows that the proposed prediction model is scientific and advanced.

Abstract:In view of the traditional detection methods that cannot effectively detect small signals from the strong chaotic background noise, this paper studies the small target detection principle in the strong clutter background, and proposes a chaotic small signal detection method based on SSA-SVM. The sparrow search algorithm is used to optimize the penalty parameter C and kernel function parameter σ of SVM to improve the accuracy of prediction, thus reducing the detection threshold and increasing the detection rate. Adding target signals to Lorenz chaotic system for simulation, the results show that the proposed method can effectively detect small signals from strong chaotic background noise, and the root mean square error of prediction of transient small signals is 0. 000 434 3 (signal-to-noise ratio is -137. 707 3 dB), which is two orders of magnitude lower than the root mean square error of prediction signals of traditional SVM algorithm of 0. 049 (signal-to-noise ratio is -54. 60 dB). The proposed method is verified experimentally by using the sea clutter data measured by IPIX radar, which further demonstrates the effectiveness of the proposed method.

Abstract:The observation of transmission busbar contact status of airport baggage transfer device is of great significance to reduce the unplanned stoppage of airport baggage transfer device and ensure the normal operation of airport. The temperature change can visually reflect the status of the transmission busbar contact, which is often accompanied by the rise of temperature when the transmission busbar contact failure occurs. Therefore, Encoder-Decoder LSTM can be used to predict the temperature of transmission bus contacts. First, an encoder composed of a bi-directional long and short-term memory network (Bi-LSTM) is used to encode the historical temperature data of the busbar contacts, then a decoder composed of a long and short-term memory network (LSTM) is used to predict the temperature value of the transmission busbar contacts for a future period. One month of temperature observation data of a domestic airport baggage conveyor is tested. The experimental results show that the time series prediction method using Encoder-Decoder LSTM outperforms the traditional time series prediction model as well as other existing deep learning prediction models.

Abstract:In the fields of remote sensing mapping, explosion field testing and logistics security, it is often necessary to measure twodimensional temperature field information with high precision. Due to the large measurement area or limited monitoring equipment, temperature field measurement accuracy and resolution are often low. Therefore, this paper proposes a two-dimensional temperature field compression and reconstruction method combining compressive sensing and piecewise Hermite interpolation theory. Firstly, the temperature field is undersampled randomly. Secondly, the sampling results are interpolated by Hermite interpolation to improve the sampling rate. Finally, the interpolation results are used to reconstruct by OMP with high probability. This method can effectively reduce the number of measuring points in two-dimensional temperature field measurement. The experimental results indicate that the reconstruction error of two-dimensional temperature field is no more than 6. 5% when the compression rate is 75%.

Abstract:In order to study the distribution of vibration and noise of transformer core, the sound field distribution of core vibration of a 200 kVA oil immersed transformer is analyzed based on the multi physical field coupling calculation of COMSOL. Firstly, the electromagnetic force multi physical field model of core vibration of 200 kVA oil immersed transformer is constructed by using COMSOL finite element calculation software. The magnetic field distribution of transformer core, the variation law of stress distribution and displacement distribution of core are calculated. The stress time domain data of typical positions of core are analyzed by spectrum, and the stress concentration of core is 100, 200 and 300 Hz. Then, the frequency domain data of the iron core surface acceleration calculated by the above multiple physical fields after FFT transformation is used as the excitation source of the sound field, the harmonic response of the sound field is analyzed, and the sound field distribution of the transformer iron core vibration is calculated. Finally, the correctness of the simulation data is verified by the no-load experimental data of a 200 kVA transformer. After comparison, it is found that it has a good agreement effect. The consistent law between simulation and experiment is that the vibration and noise frequency of transformer core is concentrated below 500 Hz, and it is found that the side sound pressure of transformer oil tank > upper sound pressure > front sound pressure; The sound pressure frequency content on the side and above the transformer oil tank is 100 Hz > 200 Hz>300 Hz, while the sound pressure frequency content on the front of the oil tank is 100 Hz>300 Hz>200 Hz.

Abstract:Aiming at the low fault diagnosis accuracy of IGBTs’ open circuit fault in dual active bridge ( DAB) converter, a fault diagnosis method based on the Levy sparrow search algorithm (LSSA) to optimize the deep belief network (DBN) is proposed. First, the Levy flight strategy improves the convergence speed and global optimization capability of the SSA. Then, the mean square error function of the DBN is taken as the fitness function. The LSSA finds the optimal number of hidden layer units of DBN. According to the optimal number of hidden layers, we construct a DBN open-circuit fault diagnosis model. Through building the hardware-in-the-loop simulation system of DAB converter in RT-LAB, the method uses the transformer leakage current as the diagnostic signal. The comparative analysis is conducted on the convergence speed, fitness value index and diagnosis accuracy. The experiment results show that the diagnosis model can diagnose the open-circuit fault of the DAB converter effectively, and the fault diagnosis accuracy achieves 99%.

Abstract:Aiming at the problems that the fluctuation of field road affects header height of a combine harvester, it further results in the difficulty of measuring the header height and the fluctuation of feed quantity during the operation of a combine harvester, a real-time method for measuring the header height based on double inertial sensors and a method for controlling the header’s height based on linear quadratic regulator (LQR) were proposed. The header height was obtained in real time by measuring the inclination angles of the combine body and the inclined conveyor with two inertial sensors. Mathematical model of the header system was established based on the kinematics and the structural analysis of the header. The optimal solution of the header height control was obtained by selecting the performance function to solve the linear quadratic optimal control problem. According to the obtained optimal solution, the hydraulic cylinder was controlled to adjust the header height, so that the header height was stable in the preset range. The simulation results showed that the root mean square error of traditional PID controller was 0. 226° when tracking the step signal with random noise, while the root mean square error of the LQR controller was 0. 133° when tracking the step signal with random noise. Therefore, the dynamic performance of the LQR controller is better than that of the traditional PID, confirming that the proposed method can improve the control quality of header height of a combine harvester.

Abstract:Based on the fact that linear active disturbance rejection controller (LADRC) cannot cope with q-axis current mutation and parameter setting defects in the compound control of speed and current of permanent magnet synchronous motor (PMSM), an improved ADRC is proposed. Firstly, the design method of conventional LADRC in the compound control of speed and current is analyzed. Secondly, based on the conventional LADRC, the fastest control synthesis function (fhan) is used to replace the proportion of differential (PD) to design the control law, improve the system control performance and optimize the parameter configuration mode. At the same time, based on the function characteristics, the current outer loop controller is designed, and the current deviation feedback algorithm is used to limit the q-axis current, so as to avoid excessive current impact damaging the hardware. Through the establishment of the experimental platform, the experimental results show that the improved LADRC can effectively deal with the q-axis current impact, which has the same anti-interference ability as the traditional LADRC, and reduces the transient response time of disturbance by 20 ms. It shows that the improved LADRC has higher safety performance and good anti-interference ability.

Abstract:Aiming at the nonlinearity enhancement of the GNSS / SINS integrated navigation system in strong maneuvering vehicular environment and the accumulation of rounding errors as the iteration times increase, the covariance matrix is no longer non-negative, resulting in filtering accuracy decrease even filtering divergence, motion constraint aided integrated navigation method based on SVDCKF was proposed. To restrain the divergence of forward velocity errors and the decrease of position accuracy when sideslip occurs, the centripetal acceleration constraint is introduced based on the traditional motion constraint. In order to verify the effectiveness of the algorithm, a car experiment was carried out. Compared with the standard SVD-CKF, the results show that the longitude and latitude error of the proposed motion constraint aided SVD-CKF algorithm reduced by about 10. 54% and 44. 64% on average at the curve, the east and north velocity errors are reduced by about 50. 87% and 62. 61% on average. This algorithm not only ensures the position accuracy of the carrier in vehicular environment, but also improves the stability and robustness of the integrated navigation system.

Abstract:The remote programmable constant current source is an important part of the servo control, and its dynamic model of whole control process is of great significance for the strategy generation and control optimization of the servo control. Aiming at the dynamic performance analysis and remote control of programmable constant current source, this paper proposes a dynamic modeling method based on edge-cloud collaboration. The proposed method takes into account the communication delay, data packet loss and other related situations in the cloud control process. It can simultaneously identify the structure and parameters of the model. Through the transformation of transfer functions of discrete domain and continuous domain, the continuous dynamic model of the constant current source is obtained to support the efficient control of the servo system, it overcomes the inconvenience of re-establishing the discrete model for different control time intervals in traditional side-end control. Finally, simulation verification and application simulation verification show that the method can accurately identify the dynamic model of the constant current source in the continuous domain and can apply to the actual edge-cloud cooperative program control system.

Abstract:A two-step time-to-digital converter (TDC) is designed based on time amplification technology, which can be applied to the field of high-precision flight measurement. This design adopts SMIC 55 nm CMOS process, uses the ring delay TDC as the coarse quantization circuit and uses the Vernier TDC as the fine quantization circuit. The accuracy of the Vernier TDC is limited by the mismatch of delay cells, which makes it difficult to break through the higher accuracy requirements in the design. The time amplifier amplifies the time margin generated by the coarse quantization and continues with the second fine quantization, which reduces the design difficulty of the fine quantization circuit. Aiming at the disadvantages of the limited input range of traditional time amplifiers and insufficient amplification accuracy, a new time amplifier structure is proposed, which has the ability to accurately amplify a wide range of input time intervals. The simulation results show that the achievable resolution of the TDC using this kind of time amplifier is 3. 7 ps, the measurement range is 80 ns, the differential nonlinearity (DNL) is 0. 73 LSB, and the integral nonlinearity (INL) is 0. 95 LSB. This design can better balance the resolution and measurement range of TDC under high linearity.

Abstract:A new denoising algorithm is proposed aiming to decrease the random error of MEMS gyroscope. Firstly, the original data is decomposed into multiple intrinsic mode functions ( IMFs) using complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN). Then the IMFs are divided into noise IMF, mixed IMF, and signal IMF according to multi-scale permutation entropy with Mahalanobis distance. Next, the noise IMF is denoised by wavelet packet (WP) and the mixed IMF is denoised by Savitzky-Golay filter (SG). Finally, the denoised signal is obtained via reconstructing the processed IMF and the signal IMF. The bumps signal is increased from 6 dB to 17 dB, and the mean square error is reduced by 71. 9% after denoising through the proposed method. The angular random walk of the denoised signal is reduced by 31. 5% in the experimental analysis of the measured gyroscope static data, which illustrates that the proposed method can predominantly improve the accuracy of MEMS gyroscope accuracy.

Abstract:To enable fruit picking robots that accurately detect target in complex conditions such as leaf covering and variances of fruit sizes, etc. , improved YOLO( you only look once) model and NMS ( non-maximum suppression) algorithm are proposed. First, the traditional YOLO deep convolutional neural network architecture is upgraded. A more fine-grained SPP5(spatial pyramid pooling)feature fusion network module is generated to enhance the integration of multiple sensory field information in feature maps, based on which a YOLOv4-SPP2-5 model is proposed. The SPP layer is added and improved in the standard YOLOv4 network across layers and redistributed the pooling kernel size and to enlarge perceptual field range, thus decreasing the false detection rate. Moreover, an improved Greedy-Confluence NMS algorithm is proposed. Through direct suppression of high-proximity detection boxes and comprehensive consideration of Distance-Intersection over Union (DIOU) and weighted proximity (WP) for overlapping detection boxes, the computational consumption of NMS was balanced and the error suppression of detection boxes was reduced, so as to improve the detection accuracy of occlusion and overlapping objects. Finally, performance tests are conducted to verify the feasibility of the method, followed by format converting and annotation labelling with fruit training datasets. The training datasets are expanded via data augmentation techniques and the K-means ++ clustering approach is utilized to obtain a priori anchor frames, and the fruit detection experiments are carried out on a computer. The results demonstrate that the improved YOLO network and NMS algorithm-based approach significantly increase the accuracy rate of fruit detection. The mean average precision (MAP) reaches 96. 65% at YOLOv4, which is 1. 70% higher than the previous network. Real-time performance is also guaranteed, hitting 39. 26 frames per second on the test device.

Abstract:Color constancy is an important research direction in computer vision, but most algorithms focus on uniform distribution of single illuminant, and the problem of non-uniform distribution of illuminant has not been well solved. In order to solve this problem, a direct correction method based on generative adversarial network is proposed to transform color constancy into a many-to-one mapping task under the condition of non-uniform distribution of single illuminant. According to the characteristic of color constancy, the image is divided into content code and illuminant code, and the image under the target illuminant is reconstructed by changing the illuminant code to target illuminant code. At the same time, in order to make non-standard illuminant more diversified, the illuminant sampling module is added to help the network learn more abundant illuminant information and realize many-to-one mapping. In order to guide images to be mapped to different illuminants when different illuminant codes are input, the illuminant supervision module is added to distinguish images with different illuminants, so as to help the illuminant conversion module better combine content coding with specific illuminant coding to generate target images and achieve color constancy. At the same time, aiming at the task of this paper, the non-uniformly distributed illuminant is rendered on the existing dataset, and the dataset with non-uniformly distributed single illuminant is constructed. The experimental results show that the proposed method solves the problem of non-uniform distribution of illuminant well, surpasses other algorithms in non-uniform dataset, and the final image is closer to the image under standard illuminant.

Abstract:At present, most of the non-invasive monitoring equipment for cerebral blood oxygen in clinical use is dual-wavelength. Because the absorption of melanin will cause deviations in the measurement results, the measurement results need to be corrected. For this reason, based on the basic principle of non-invasive monitoring of cerebral blood oxygen by near infrared spectroscopy, in view of the shortcomings of existing non-invasive monitoring equipment for cerebral blood oxygen, considering the optical characteristics of the prefrontal lobe, a four-wavelength detection light source (700, 760, 805, 850 nm) and a dual-channel photoelectric detector for noninvasive monitoring of cerebral blood oxygen, and a mathematical model of spectral absorbance is established. On this basis, a new type of non-invasive monitoring system for cerebral blood oxygen is constructed, which can inhibit the interference of skin melanin components to non-invasively measure the cerebral blood oxygen information. Finally, through the rank sum analysis of the Valsalva exercise and noload control experiments, the effectiveness of the system is preliminarily verified.

Abstract:The application of electromagnetic tomography (EMT) in the detection of the distribution of magnetic catalysts in biodiesel preparation is studied. Based on a sensor array with 8-coil TMR coil structure, the simulation experiment is designed, and the imaging results of catalysts with different particle sizes are compared. The research shows that the coil structure is suitable for the distribution detection of magnetic catalysts. In order to improve the shortcomings of the traditional imaging algorithm, the application of the matching pursuit algorithm (CoSaMP) in image reconstruction is studied, and compared with the traditional Tikhonov algorithm and Landweber algorithm, the results show that the CoSaMP algorithm has better edge processing ability and faster imaging speed. Compared with the other two algorithms, the image error is reduced by 30. 4% on average, and the imaging speed is increased by 46%, which effectively improves the quality and speed of reconstructed image.

Abstract:In robotic spray painting, 3D model of workpieces plays an important role, requiring information in geometry and shape for path planning or pose estimation. To solve the problem that it is difficult to effectively acquire a 3D workpiece model in practical spraypainting applications, this paper proposes a fast-modeling method based on complementary perspectives with multiple RGB-D cameras. In this method, two RGB-D cameras installed oppositely are used to acquire the local point cloud of the workpiece in a complementary perspective. Then, a special simple two-sided calibration plate is designed to offline calibrate the external parameters of the camera system based on the idea of coplanar joint optimization of intersection points, which is to accurately estimate the transformation matrix between cameras without coincident field of view. Finally, with the calibrated transformation matrix, the local workpiece point clouds with non-overlapping areas online captured from different camera perspectives are fused into a complete 3D workpiece model. Experimental results show that the proposed method can effectively construct the model of polyhedral workpiece with complicated structure, less than 160 ms of reconstruction time, and below 5. 3 mm of error.

Abstract:Aiming at the problems of low accuracy and robustness of rail detection and recognition methods and poor fitting of curved rail, a rail detection and recognition algorithm based on straight-curve hybrid model is proposed. In the beginning, the image is preprocessed and Canny edge detection is completed by adjusting the lag threshold. The Progressive Probabilistic Hough Transform is used to detect the direct straight track, divide the near and far field of view and determine the vanishing point. The straight track in the near field is fitted by a linear model, and the feature points of the rail are obtained by the circular linear approximation of the far field according to the detection results, and verified according to the gray characteristics of the rail. The least square method is used to complete the curve fitting. The switch of straight-curve model is completed according to the established rules. Experimental results show that the detection accuracy of the proposed algorithm is 90. 1%, which is suitable for different environments and has good robustness.

Abstract:The surface detection of optical fiber connector belongs to precision instrument detection, accordingly, making it possible for the large amounts of dust in the factory environment that exerts detrimental influence on the recovery of optical fiber connector. Nonetheless, the current detection technology possesses long running time, poor retention ability for image details, and is problematic to overcome interference in the actual working environment. To this end, we propose a self-identification noise reduction technology based on optimised extreme learning machine. Firstly, the interference data is processed by dimensionality reduction. Secondly, select the dimensionality reduction data as the training data, and use the extreme learning machine optimised by AdaBoost algorithm to locate the noise. Ultimately, the positions of noise points are repaired by filtering algorithms. The experimental results demonstrate that the selfrecognition noise reduction algorithm based on AdaBoost-Elm is equipped with high noise recognition ability and its ANRR reaches 97. 33%. Additionally, the average value of BBS and NRIQAVR based on AdaBoost-Elm noise reduction algorithm are 131. 14 and 2. 61 respectively, which are better than global filtering algorithm. In the end, we simulate the factory environment and use mean filtering based on AdaBoost-Elm to perform 3D restoration test on the sharply polluted fiber optic probe under different light intensity conditions. It is found that its BBS reaches around 130 and its NRIQAVR is lower than 2. 57, which has apparent merits compared with the median filtering based on Elm.

Abstract:Because the traditional detection algorithm is not accurate enough to detect cotton packaging defects and the recognition rate of small target defects is not high enough, an improved Faster R-CNN deep learning network is proposed to detect five defects such as damage, stain, hole, impurity and thread end in cotton packaging. Image enhancement is realized by preprocessing the image, then the RPN and ROI structure in Faster R-CNN are improved. In order to strengthen the detection ability of small target defects, the feature pyramid network structure is fused in the backbone network, and finally the ROI is bilinear interpolated to solve the problem of pixel deviation caused by multiple quantization. Experiments show that the average accuracy of the improved network for cotton packaging surface defect detection is 91. 34%, which is 9. 08% higher than the traditional algorithm.

Abstract:In order to effectively suppress the interference of the C-band of the Indian National Satellite Communication and X-satellite frequency bands to the ultra-wideband communication system, a new type of dual notch bands UWB filter is proposed. The filter adopts the intersection with a stepped T-type multimode resonator (MMR) and a defective ground structure (DGS) to achieve ultra-wideband characteristics. Using asymmetric coupling lines and coupling split ring resonators on both sides of the MMR, notchs are generated in the two frequency bands of 6. 67 ~ 7. 06 GHz and 7. 47 ~ 7. 57 GHz respectively. The measured results are in good agreement with the simulation results. The passband range of the filter is 3. 03~ 11. 50 GHz, the 3 dB bandwidth reaches 123%, the insertion loss is only 0. 87 dB, and the center frequencies of the two notches are at 6. 87 GHz and 7. 52 GHz, respectively. The depth of notch bands is greater than 20 dB, and the overall size is compact, only 16 mm×8 mm.

Abstract:Aiming at the problem that the early fault features of rolling bearings are weak and difficult to be effectively identified, this paper proposes a fault diagnosis method for rolling bearings which is based on tSNE-ASC feature selection and DSmT fusion decision. Multiple sensors were used to collect bearing acoustic signals under different fault modes, and each signal was decomposed by VMD to obtain multiple IMF components. Feature extraction was carried out for each IMF component, and data set matrix of each feature was constructed. TSNE was used to reduce the dimension of the matrix of each feature data set to two dimensions and calculate the average contour coefficient (ASC). According to the fact that ASC greater than critical value, the sensitive features of acoustic fault signal are extracted. The primary diagnosis of bearing fault is realized based on diagnosis model. DSmT is used to fuse the primary diagnosis result of acoustic signal and get the final diagnosis conclusion. Experimental results show that the tSNE-ASC feature selection method can effectively extract sensitive features in the mixed domain, and has high diagnostic accuracy in different working conditions and different diagnostic models. DSmT decision fusion effectively reduces the uncertainty of single signal diagnosis, and has high diagnostic accuracy under the condition of variable load and non-stationary speed up and down.

Abstract:Carbon fiber reinforced polymer is widely used in high-tech fields such as aerospace, fatigue damage will occur during service, it will bury potential safety hazards. Therefore, its health needs to be monitored, and damage probability imaging algorithms can be used to intuitively reflect structural health. However, the traditional damage probability imaging algorithms have a high damage probability in the non-damaged area and it is difficult to accurately locate the damage. In view of the above problems, a damage probability imaging algorithm based on Lamb wave energy and time of flight is proposed. The area measured is evenly divided into N pixels, calculate the Lamb wave energy and time of flight damage factor of each channel, determine and superimpose the probability value in the affected area of each channel damage factor, and obtain the damage probability of each pixel and image it. The experimental results show that compared with the frequently-used damage probability imaging algorithms based on energy damage factor and cross-correlation damage factor, the proposed method can intuitively reflect the defects of carbon fiber reinforced polymer, and the recognition effect is better, the imaging error is significantly reduced, the error is reduced by 4. 420, 2. 117, 2. 055 and 4. 732, 2. 380, 2. 647 respectively, which can identify defects more accurately and guarantee the safe application of carbon fiber reinforced polymer structure effectively.

Abstract:The return oil pipeline of aeroengine works in the mixed flow state of lubricating oil and air. The voidage measurement of oilgas two-phase in the pipeline is of great significance for understanding its characteristics, analysis and design of the pipeline. In this work, according to the practical state of oil return tube, an 8-electrode electrical capacitance tomography (ECT) sensor used for hightemperature operating condition was designed. With the array capacitance data from the designed ECT sensor, a voidage measurement model based on principal component regression (PCR) was established, and a new temperature compensation method was presented by calibrated capacitance value at different temperature in oil-filled full tube. The experiments were carried out, and the results showed that the presented voidage measurement model and its temperature compensation method were effective, and that the maximum absolute error of the voidage measurement was less than 10% with the temperature range of 20 ℃ ~ 180 ℃ .

Abstract:Kalman filtering is a common noise reduction method in the process of positioning using ultra-wideband (UWB). However, this algorithm has poor filtering performance of nonlinear system. The moving track of positioning target is easy to exceed the layout area of base station and be disturbed by abnormal noise, which will affect the accuracy and stability of positioning system. In order to solve this problem, a symmetric strong tracking ( SST) square root volume Kalman ( SCKF) algorithm was proposed. By introducing a symmetric time-varying fading factor to adjust each covariance matrix, the working mode of the multiple fading factor matrix in the error covariance matrix is changed, and then the filter gain is adjusted. Although the computational complexity increases slightly, the adaptability and robustness of the positioning model are enhanced. Simulation results show that under the interference of abnormal noise, the improved algorithm (SST-SCKF) can effectively improve the positioning accuracy compared with SCKF/ SCKF for multiple fading factors ( ST-ASCKF) algorithm, and the positioning trajectory is smoother than SCKF with single fading factor ( STSCKF). The positioning scheme based on UWB technology was designed by SST-SCKF algorithm. The dynamic simulation experiments show that the SST-SCKF algorithm proposed in this paper has better filtering performance than SCKF/ STSCKF/ ST-ASCKF. This SST-SCKT algorithm provides better noise reduction for personnel UWB positioning under complex environmental noise and makes the positioning more accurate.

Abstract:In view of the current vehicle tracking research algorithms, the Kernel correlation filtering algorithm ( KCF) has the shortcomings of single feature extraction and the inability to adapt the scale in a complex background, this paper proposes a multi-feature fusion scale adaptive algorithm. The algorithm uses color histogram information as color features, takes high-level convolution features with more semantic information and low-level convolution features with higher resolution as depth features, and performs adaptive feature fusion with color features. Then, the context image is used to constrain and optimize the target background information, and the response confidence is measured by the average peak correlation energy detection, and finally the high-confidence tracking result is used to avoid the problem of the model being vulnerable to interference. In addition, in order to achieve high target tracking accuracy, the algorithm in this paper uses a hierarchical model update strategy to update the extracted features. Experiments on the OTB100 data set show that the accuracy of the algorithm in this paper is better than other mainstream tracking algorithms Staple, SAMF, KCF, TLD, DSST and CSK are 4. 9%, 5. 7%, 10. 2%, 10. 3%, 23. 4%, 29. 7% higher.

Abstract:In view of the problems of large model volume, high computation and unable to deploy to mobile terminal in the blind recognition of modulation mode in traditional convolutional neural network, a modulation recognition method of attention and Ghost convolution neural network (AG- CNN) based on dual attention mechanism and ghost module is proposed. The modulation signal is mapped to complex space, the map points are processed by the normalized point density, and the higher order feature density constellation is obtained. The feature is used as input of AG-CNN model for learning training, the trained model is finally used to identify the unknown signal received by the receiver. The experimental results show that the recognition rate of density constellation map with sampling point of 10 000 is over 99. 95% by AG-CNN model. Compared with CNN model with the same number of layers, the convolution layer parameter is compressed by 6. 01 times and the calculation amount is 6. 76 times. Compared with VGG-16, Inception V3, ResNet-50, Shufflenet, Eficientnet and other convolutional network models, the number of parameters and floating-point operations decreases significantly, and in the case of saving learning parameters and reducing the complexity of the model, it shows excellent classification performance.