Abstract:Due to its advantages of small trauma and fast recovery, minimally invasive puncture surgery has been widely used in clinical practice in recent years, mainly for tissue biopsy, local ablation and local anesthesia. At present, the surgical instruments used in puncture surgery are oblique pointed flexible needles, which can avoid important organs and tissues and cause less damage to patients than rigid needles. In the process of surgery, effective path planning algorithms can assist doctors to improve the safety and accuracy of surgery, and the error of existing intraoperative path planning algorithms is in the order of millimeters, which limits its application in clinical surgery, and solutions need to be explored. Therefore, the research progress of flexible needle path planning in puncture surgery is sorted out, the factors affecting path planning are analyzed, the path planning algorithm of puncture needle is introduced in detail, the key problems that need to be solved are pointed out, and finally the future development direction of path planning of flexible needle surgical puncture needle is discussed from different aspects such as optimal path selection, parameter feedback and intraoperative realtime path planning method.

Abstract:The passive ankle exoskeleton can provide an efficient assistance at ankle push-off. However, current research mainly focused on its assistance effects during treadmill walking, the actual motion performance during overground walking has not been explored. In this study, we developed a low-cost passive ankle exoskeleton, and firstly investigated its effects on muscle activations and joint kinematics during overground walking. Five healthy young subjects were enrolled in our experiment where they were required to perform overground walking under three wearing conditions: normal walking, wearing an exoskeleton, and wearing an exoskeleton with zero torque assistance. The joint kinematics and electromyography ( EMG) signals of the major muscle groups of both lower limbs were collected simultaneously. Features of muscle activation and joint angles were extracted. A two-way analysis of variance ( Two-way ANOVA) was employed to investigate the effect of wearing condition on these features. The results showed that the mechanical frame can support the human body resulting in reduced muscle activation of medial gastrocnemius (MG) during the stance phase. Compared to normal walking, the exoskeleton significantly decreased the peak torque and power of the ankle joint while the average activation of the MG was decreased by 20% during the stance phase. Additionally, the exoskeleton may change the hip and knee joints' kinematics. The study demonstrated that the passive ankle exoskeleton effectively assisted the ankle joint during the stance but the wearer may alter gait performance for adapting exoskeleton.

Abstract:In order to make up the deficiency of discrete state recognition in motion process prediction, a motion intent recognition method for human joint angle prediction is proposed. The driver circuit and echo acquisition program were designed for the A-mode ultrasound probe to measure the muscle thickness. The motion data of 6 volunteers were collected. A customized mapping relation model was generated after fitting the data of muscle thickness and joint angle during the same movement. The average value of the coefficient of determination R 2 is 0. 916 9, which shows a good matching degree. The system’s output response frequency of predictive value can reach 30 Hz when the model was solidified into the program, which indicates that the method can track the continuous state changes in the process and recognize the human motion intent. Compared with the discrete state recognition method, it can effectively improve the recognition precision and real-time performance.

Abstract:Aiming at the problem of limited detection accuracy in the application of brain-computer interface (BCI) technology in target detection, a new encoding and decoding method based on the decision layer fusion of P300 and error-related potential (ErrP) in eventrelated potential (ERP) was proposed. In the encoding aspect of the BCI system, the P300 and ErrP features are respectively evoked by the target image and visual feedback. In the decoding aspect, four schemes are used for target detection: individual P300 feature, individual ErrP feature, feature layer fusion of P300 and ErrP, and decision layer fusion of P300 and ErrP. The average results of 10 healthy subjects with four schemes show that the balance accuracy of decision layer fusion of P300 and ErrP is the highest, reaching 80. 03%±5. 20%, which is improved by 4. 38% compared with the method of using individual P300 feature and is improved by 11. 29% compared with the method of using individual ErrP feature. The feasibility of hybrid BCI technology in target detection tasks is verified.

Abstract:To address the problems of background interference and limited dataset size, we propose a method for locating the region of interest in thyroid ultrasound images. The method utilizes an attention mechanism based on cross-scale attention interaction strategy to improve the fusion efficiency of hierarchical features in the localization model. The feature network of the localization model is enhanced through knowledge distillation to solve the problem of overfitting. A t-mask is designed based on the statistical distribution of anatomical thyroid morphology, and a joint attention mask is calculated to guide the network in learning key channels and pixel information of thyroid ultrasound images, thereby achieving the localization of the region of interest. Experimental results demonstrate that the average precision (AP) for thyroid ultrasound image region of interest localization reaches 92. 7% when the IoU threshold is set to 0. 5, which is clinically significant and valuable for assisting doctors in diagnosing thyroid diseases.

Abstract:Due to the nonlinearity and non-smoothness of surface electromyography ( sEMG) signals, which leads to the limitations of traditional muscle fatigue classification methods, a muscle fatigue classification method based on the combination of Fourier decomposition method (FDM) and machine learning is proposed based on this. The FDM is used to decompose the sEMG signal into a series of Fourier intrinsic band functions (FIBF), determine the optimal decomposition level, extract the ratio of the total power of each FIBF component to the total power of the sEMG signal (FTPR) as classification features using the FDM, and compare the effectiveness of each machine learning classification algorithm and the effect of data length on the classification accuracy. It was shown that the FDMbased feature extraction method can effectively identify muscle fatigue states, and an average classification accuracy of 98. 17% was obtained using a support vector machine classifier with a data length of 3 000 and a 10-level decomposition level of FDM. Each FIBF component was analyzed individually, and it was found that the FTPR under the 5th FIBF component had the best class separability, and the FTPR of the 1st to 2nd components would become larger when muscle fatigue was present, and the FTPR of the 4th to 10th components would become smaller, i. e. , the frequency amplitude of the sEMG signal in the 0 ~ 117 Hz interval would increase when muscle fatigue was present, and the frequency amplitude in the 175. 5 ~ 585 Hz interval would decreases. By comparing the muscle fatigue classification effects of different feature extraction methods, the experimental results show that the FDM and FTPR features can significantly improve the classification accuracy. Therefore, the proposed method can be used for muscle fatigue state recognition.

Abstract:Current research on gesture recognition based on convolutional neural network (CNN) focuses on increasing the depth of the network, and pays less attention to improve the distribution of sample data which can brought the performance improvement. Aimed at these problems, a kind of electromyography feature matrix ( EFM) sample that quantifies the correlation of surface electromyography (sEMG) features is fed into the efficient channel attention (ECA) mechanism CNN, which is used to identify the 52 types gesture in NinaproDB1. Firstly, the time window is used to truncate the low-pass filtered sEMG and calculate various signal time domain features. Then, the cartesian product is used to combine and multiply different features. The EFM is obtained after normalizing the feature multiplication values. At the same time, ECA mechanism is introduced to make the network focus on the important deep features, thereby improving the effect of gesture classification. sEMG, EMG time-domain features and EFM are fed to the attention mechanism CNN respectively for gesture recognition. The recognition accuracy of EFM is the highest and reached 86. 39%, which is higher than the accuracy of gesture recognition research methods in recent years. The effectiveness of the proposed method is verified, and a feasible new scheme for accurate multi-category gesture classification is provided.

Abstract:To solve the problem of obtaining the optimal path for mobile robots running during goal-oriented running in an unknown environment, a path planning model in biological inspired goal-oriented navigation based on Q-learning is proposed in this paper. The model includes three parts: Spatial exploration based on Q-learning, running control based on cognitive map and optimal path selection. Firstly, in space exploration, the location state is represented by place cells’ firing statues, and the state-action is learned by using dynamic ε value, which can generate cognitive map and provide the optimal path in space exploration stage. Secondly, in the running control based on cognitive map, the running direction is selected respectively according to the principle of maximum action cells’ firing and the principle of group action cells, and the multi-scale position update intervals are used to update the position. As a result, the optimal path based on different cognitive maps can be obtained. Finally, path planning’s result from space exploration stage and running control stage is compared, and the optimal path is selected. Simulation results show that the proposed model is feasible. A better path planning result can be obtained by using the dynamic ε value in space exploration. Besides, a feasible and effective path can be provided for goal-oriented running after sufficient space exploration.

Abstract:This paper proposed an internal defect volume automatic measurement algorithm based on the improved spatial intuitionistic fuzzy C-means clustering (NL-SIFCM) and 3D region growing for measuring the volume of holes and cavities in 3D CT images of workpieces. Firstly, the acquired 3D CT images are pre-processed. Subsequently, NL-SIFCM was used to segment on 3D CT image to obtain a binarized defect image set, while a fast algorithm was obtained for 3D CT image having spatial similarity between slices. Finally, the binary defect image set for 3D region growing to obtain the defect voxel number and spatial structure, and the defect spatial structure is displayed in the 3D visualization software to assist inspectors in analyzing defects. The experimental results show that the measured volume of the standard spherical volume used to simulate defects has a relative error of less than 1. 0%, indicating a high level of measurement accuracy. The applicability of the algorithm has been validated by actual workpiece inspection, demonstrating its effectiveness in meeting the demands of CT inspection.

Abstract:The accuracy of time measurement units ( TMU) in integrated circuit automatic test equipment ( ATE) is facing higher demands due to the rapid development of integrated circuits. To cope with the increased accuracy of digital IC time parameter measurements, a high precision time measuring unit was designed in this paper, which used the electronic pin test chip MAX9979 to apply excitation and capture response, and combined with Xilinx Artix-7 FPGA internal curing time digital converter (TDC). Time-todigital converter implemented by interpolation using a combination of coarse and fine counting, the coarse counter is implemented by a 32-bit direct counter with a reference clock of 200 MHz, while the fine counter consists of a delay chain cascaded by a CARRY4 cascade, with a dedicated configuration for CARRY4 to reduce the measurement error caused by its overrun function, and the delay chain was calibrated using the code density calibration method. The experimental results show that the TMU range is 21. 475 s, the average resolution is 34. 7 ps, the DNL is less than 2. 5 LSB, the INL is better than 4. 5 LSB, and the accuracy is 39. 7 ps.

Abstract:In order to improve the hidden fault detection rate of electromagnetic relay, a fault detection method based on super-path time measurement for electromagnetic relay is proposed in this paper. Firstly, the fuzzy threshold is optimized offline. Secondly, the superpath time data is predicted online using sliding window linear regression, and extended to predicted interval based on fitting error. Then, the relay super-path time measured data and predicted interval are converted into current fault evidence and predictive fault evidence using fuzzy threshold, and cumulative fault evidence is updated by iterative fusing cumulative, current and predictive fault evidence using interval evidence reasoning. Finally, the result of fault detection can be obtained from cumulative fault evidence according to fault criteria. The proposed method considers the slow-changing features of hidden relay fault and uncertainty in super-path time measurement process, so that the fault evidence is closer to the actual working condition of relay. The experimental results show that the proposed method has significant advantages in evidence accuracy and convergence speed, and can effectively improve the detection accuracy of electromagnetic relay hidden fault.

Abstract:In the process of aluminum alloy casting and high temperature rolling, non-contact nondestructive testing technology is used to realize online monitoring and detection, which is of great significance to reduce production costs, ensure the continuity of production lines, and improve product yield. Firstly, the finite element model of the laser EMAT testing process of aluminum alloy, which is excited by pulsed laser beam and received only by the coil electromagnetic ultrasonic transducer (EMAT), is established. The influence of water film surface restraint mechanism and silicon steel magnetic structure on the amplitude of excited multi-mode ultrasonic wave is analyzed. The influence of the outer diameter, inner diameter, wire diameter, number of layers of the excitation coil and the receiving coil of the coil electromagnetic ultrasonic transducer (EMAT) only on the ultrasonic receiving efficiency is studied. Secondly, the Laser EMAT test of aluminum alloy was carried out to verify the influence of water film surface constraint mechanism, coil only EMAT design parameters and silicon steel magnetic focusing structure on the detection echo amplitude. The results show that the ultrasonic echo signal amplitude increases by 37. 76% and the signal-to-noise ratio increases by 17. 3 dB when silicon steel is used as the magnetic back plate of the excitation coil under the constraint of the water film surface. When the laser energy is constant, the spot size is constant, the outer diameter of the excitation coil is 12. 3 mm, the inner diameter is 1. 6 mm, the wire diameter is 0. 4 mm, and the number of layers is 2, the coil impedance is consistent with the internal resistance of the circuit. The coil obtains the most energy and provides the strongest radial bias magnetic field. When the outer diameter of the receiving coil is 14. 1 mm, the inner diameter is 1. 7 mm, and the wire diameter is 0. 26 mm, the ultrasonic receiving efficiency is the best.

Abstract:In view of the poor performance of traditional packet-switched network-on-chip ( NoC) in a large amount of data communication, this paper proposed a design scheme of double ring network-on-chip ( DRNoC) based on packet circuit connected (PCC) switching. First, this double ring topology is composed of inner and outer rings, which can realize two-way communication within or between rings, and the number of nodes on the ring can be expanded. Secondly, DRNoC router channels can be configured as bridge nodes or link point routers. Compared with 2D-Mesh router, the number of channels is reduced, and the structure is simpler and the resource consumption is less. Finally, a double ring dynamic routing algorithm (DDRA) for DRNoC is proposed. This algorithm does not need to decode and judge the output direction at each routing node. When the establishment of the first packet is blocked, other routing paths are selected according to the network situation, which ensures the cross-ring transmission of data on the basis of the same ring transmission to the greatest extent, reduces the waiting time for the establishment of the first packet and improves the throughput. Experiments show that in the case of a large amount of data communication, DRNoC equipped with DDRA algorithm can reduce the hardware resource overhead and the average packet delay of the network and improve the average throughput, effectively improve the network performance.

Abstract:To address the problem that unmanned aerial vehicle (UAV) is vulnerable to GPS spoofing, an UAV GPS spoofing detection model based on long short-term memory ( LSTM) is proposed. In order to improve the training accuracy of the model, the training dataset was firstly enhanced using time series generative adversarial networks (TimeGAN) to compensate for the lack of training data and to compare the performance difference between the enhanced dataset and the original dataset. The LSTM model was then built, and experimental results show that the accuracy, precision, recall and F1 value trained by the TimeGAN+LSTM model under simulation experiments are 98. 08%, 98. 55%, 98. 07% and 98. 31%. Finally, the comparison with the traditional machine learning model proves that the proposed spoofing detection model has better performance metrics. The model can achieve effective detection of UAV GPS spoofing signals.

Abstract:With the development of integrated circuit technology and increased integration, circuit delay has significantly decreased. The research on traditional time-to-digital converters (TDC) tends to focus on circuit designs that combine high resolution and high accuracy. In recent years, as Moore’ s law has gradually become less effective and with the rise of the Internet of Things ( IoT), lightweight, miniaturized, and low-power edge devices have rapidly developed. The research focus on miniaturized TDCs for on-chip delay measurement has gradually shifted towards high-precision and low-power designs. Based on the Xilinx Virtex-6 XC6VLX240T FPGA development platform, a coarse measurement structure using a self-timed ring ( STR) instead of direct counting method and a fine measurement structure consisting of two symmetrical delay chains are proposed. The coarse measurement structure’ s STR is combined with the fine measurement’s symmetrical delay chains using edge coincidence detection units and latch units. The design results show that the range of the structure can reach 491 ns, with a resolution of 14. 8 ps and a maximum accuracy of 12. 9 ps. The power consumption is 0. 068 W, indicating that the proposed two-stage differential structure has the characteristics of high precision and low power consumption.

Abstract:Due to the strong nonlinearity of material used in soft gripper, it is difficult to establish a precise mechanism model for measuring the bending angle of the soft gripper. To address this challenge, a mechanism and data-driven soft sensor model for the bending angle of the soft gripper is proposed. The model consists of a mechanism model and an adaptive block increment stochastic configuration networks (ABSCN) compensation model, which includes information scent evaporation and inertia weights. The mechanism model parameter is identified using least squares, and ABSCN is used to predict compensation for high order unmodeled dynamics. By adaptively optimizing the number of incremental block configurations in block incremental stochastic configuration networks (BSC), the compactness of the model is improved, and the training time is reduced. Finally, through simulation experiments and comparison with real data using a hybrid model, it is shown that the proposed method significantly improves the accuracy.

Abstract:In order to solve the problem of insufficient defect feature extraction in the complex mine environment, a fast and intelligent guide defect recognition algorithm RDM-YOLOv5 is proposed by integrating feature enhancement and cascading attention mechanism, aiming to solve the current situation of low efficiency of manual inspection. Firstly, in order to improve the information representation ability of the backbone network feature map, the feature enhancement module RLKM is designed, which enhances the backbone network’s ability to extract target features through reparameterized large kernel convolution, and effectively reduces the amount of model parameters. Then, after extracting the high-level and low-level features through the backbone network, under the action of the cascaded attention mechanism DCAM composed of the designed channel attention mechanism and coordinate attention mechanism, the deep semantic information of the defective target is further mined, and the feature information of the small target are significantly enhanced. Finally, in order to improve the detection accuracy while ensuring the lightweight of the detection network, a lightweight convolution GSConv is introduced into the feature enhancement network to reduce the computational cost while maintaining the accuracy of model detection. The experimental results show that compared with YOLOv5s, the detection accuracy and speed of RDM-YOLOv5 are increased by 3. 7% and 11. 4%, respectively, and the number of model parameters is reduced by 15. 4%. It can basically meet the needs of accurate identification and rapid location of guide surface defects in practical applications.

Abstract:A 3D object detection and size measurement algorithm is designed for object detection and size measurement tasks in 3D scenes, which fuses LIDAR and camera sensors. A 2D object detector based on convolutional neural networks is used to extract the 2D detection box of the object. The 3D point cloud containing the object is obtained by combining the 2D detection box in the image and the geometric projection relationship. The object clustering point cloud is obtained by the Euclidean clustering method, realizing 3D object detection. An improved size measurement scheme based on the 2D detection box of the object is proposed to replace the original 3D box information obtained after point cloud clustering, improving the accuracy of object size measurement. The accuracy of object detection and size measurement is evaluated and tested on existing datasets. Experimental results show that the average detection accuracy of the 2D object detector YOLO v7 reaches 81% on the detection dataset, and the measurement error of the improved size measurement scheme is within 5% for object size measurement. It also has good performance in object detection and size measurement for objects that are farther away or smaller. Keywords:object detecti

Abstract:Complex permittivity is a crucial parameter for high-frequency substrate. Precise measurement of dielectric constant and loss is essential for the practical application of high-frequency substrate. A transmission line circuit based on microstrip lines was designed to obtain the loss characteristics of the substrate. Microstrip transmission lines of 25. 4 and 127 mm in length were simulated, fabricated, and measured to obtain the return loss S11 and insertion loss S21 within DC-20GHz. The measured data indicates that the results of S11 are below -15 dB, and the insertion loss of the transmission line is 24. 02 dB/ m at 20 GHz. By processing error analysis, the change of simulated S11max within DC-20GHz can reach about 6 dB when circuit parameters changed by 50 μm. Finally, the relative dielectric constant and loss tangent of the substrate are obtained by combining the resonant ring method with the same substrate. The results demonstrate that this method yields a high degree of accuracy for the loss tangent, with an error of less than 10% at 2, 10 and 20 GHz.

Abstract:Aiming at the problems that the ballast resistance in the jointless track circuit is easily affected by the working environment and the implementation of field measurement is difficult, the ballast resistance estimation method is proposed. Firstly, according to the two-port network theory, the transmission model of the jointless track circuit under the adjustment state is constructed, and the influence law of ballast resistance and compensation capacitance on the voltage and current value at each end of the track circuit equipment is analyzed by simulation. Secondly, based on field microcomputer monitoring data, the error between the simulated value and the measured value of the voltage and current at each end of the track circuit equipment is calculated. Finally, the particle swarm optimization algorithm is used to estimate ballast resistance value by regarding the track circuit transmission model and the error value as the adaptation function and the algorithm adaptation value respectively. And the results are used for performance verification. The experimental results show that the estimated result of this method is verified and the mean absolute percentage error(MAPE) reaches 14. 31%. Thus, the calculation accuracy of this method is better than that of the rail entry voltage method, which verifies the validity and feasibility of this method.

Abstract:The groove of ultra-narrow gap welding is narrow and deep, so it is difficult to evaluate the welding quality directly through vision. To solve the above problems, this paper proposed an ultra-narrow gap welding quality evaluation model based on chaotic multistrategy disturbed sparrow search algorithm ( CMDSSA) to optimize support vector machine ( SVM). Firstly, the sparrow search algorithm ( SSA) is improved, and the Logistic-Tent chaotic mapping and multi-disturbance strategy are introduced to improve the optimization performance of the sparrow search algorithm. Then, the superiority of CMDSSA algorithm is verified by comparing with SSA, CSSOA, PSO, GA and WOA algorithms. Finally, CMDSSA was used to optimize the penalty factor C and the kernel parameter g of SVM, and a CMDSSA-SVM quality evaluation model was constructed to evaluate the welding quality. The results show that the evaluation accuracy of CMDSSA-SVM is 97. 541%, which verifies the high accuracy and feasibility of the proposed method for ultranarrow gap welding quality evaluation.

Abstract:Aiming at the high-precision trajectory tracking control in joint space for cable-driven aerial manipulator subjected to lumped disturbances, this article proposes an adaptive robust control strategy based on time delay estimation technique. In the control frame, a time delay estimation technique is introduced to compensate the unmodelled characteristic, external disturbances and dynamic coupling effects. A fractional-order nonsingular terminal sliding mode controller has been designed to improve the convergence speed of the system states and ensure the control precision of trajectory tracking. Meanwhile, an adaptive law is designed to enhance the robustness of the controller. The stability analysis of the closed-loop system has been conducted based on Lyapunov theory. Lastly, the effectiveness of the proposed controller has been verified through a visual simulation and ground tests. The results show that our proposed controller has higher tracking precision, better robustness and stronger ability of disturbance rejection.

Abstract:Microlens array plays an important role in the field of modern optics. It is of practical significance to evaluate the quality of array devices by measuring residual stress and guiding the optimization of manufacturing process. Most of the existing measurement studies focus on the stress measurement of the whole array device, but the stress distribution of the microlens unit has not been systematically measured. Based on the principle of birefringence, this paper uses photoelastic system/ polarizing microscope respectively to measure and study microlens array devices with different base thickness ( 0. 5 mm/ 1 mm), different unit arrangement ( tangent arrangement / tight arrangement) and different microlens unit calibers (126 μm/ 1 mm) at macro / micro scale. The stress distribution law of the whole device ( macroscopic) and the stress distribution of the lens unit ( microscopic) are compared and summarized. The experimental results show that the residual stress distribution of the microlens array is uniform, and there is a significant convergence between the macro residual stress distribution and the micro residual stress. Under the premise of the same thickness, the distribution of residual stress is less affected by the arrangement and aperture size. When the thickness of the array decreases from 1 mm to 0. 5 mm, the residual stress in the same area of the square aperture microlens array increases by about 250%. For the thickness of the circular aperture microlens array, the residual stress increases by about 150% in the same measuring area. The residual stress increases significantly within the same aperture / configuration / region, decreases from 1 mm to 0. 5 mm, the residual stress in the same observation area increases by about 150%.

Abstract:Arc fault is one of the main causes of electrical fire. In the electric system of electric vehicle, the DC series arc fault usually occurs at the loose contact point or the line connection damage, which will cause serious accidents such as fire and explosion. In order to quickly and accurately detect the series electric arc fault of electric vehicles, an experimental platform for electric car fault arc has been constructed to collect time series data on trunk current under various operating situations and create a sample library. Through the lightweight convolutional neural network, a series fault arc detection model based on the improved Mobilenet network is established. By comparing and analyzing the learning rate, network layer number and sample length, the model is optimized. The optimization model can realize the detection of series fault arc, the selection of fault lines of electric vehicles through the trunk current, and the detection accuracy reaches 96. 39%. This paper provides a feasible scheme for the arc fault detection of electric vehicle’s electrical system.

Abstract:Aiming at the problems of extensive classification of health status of S700K switch machine, slow diagnosis speed and low efficiency; a diagnosis method based on CEEMDAN and kernel based extreme learning machine (KELM) is proposed. Firstly, the power data of S700K switch machine is decomposed by adaptive noise complete set empirical mode decomposition, and six intrinsic mode functions (IMF) are obtained. Then, the fuzzy entropy (FE) value of the intrinsic mode function is calculated as the characteristic parameter to characterize the health state of the switch machine. Finally, the kernel limit learning machine improved by sparrow search algorithm (SSA) is used to diagnose nine health states, and compared with SVR and ELM models. The simulation results show that the accuracy rate and the recall rate of the improved kernel based extreme learning machine model are 97. 8%, 98. 0% and 97. 8% respectively. Compared with SVR and ELM models, SSA-KELM model improves the diagnostic accuracy rate by at least 2. 2% on the basis of ensuring the running speed.

Abstract:Attitude and heading reference system (AHRS) can be mounted at Human body to sense and determine the orientation of Human in motion. The algorithms of orientation determination in literature still have problems of insufficient precision. Thus, we design a square root unscented Kalman filter (SRUKF) based on minimum sigma points set with high precision and good rapidity. To cope with the problem of sigma points set determination, we survey the matching method of higher moments than 2 in covariance matrix. As a result, we propose a SRUKF algorithm based on n+2 sigma points set, which retains rapidity of n+1 sigma points set, and also achieves the precision of 2n+1 set. The covariance matrix of sigma points reaches the L / 2 order approximation of original function if the sigma points have L order precision. Particularly, the covariance reaches 2th order if the sigma points have 2th order precision. Furthermore, we analyses the convergence of the filter process step by step with nonlinear drift function. The simulation using a standard experimental data is conducted in comparison with other filters. The orientation precision is similar to that of 2n+1 while the execution time is similar to that of n+1. The results demonstrate that the designed algorithm could achieve stable and reliable orientation determination of AHRS.

Abstract:Helmets and work clothes are important guarantees for the safety of substation staff. In order to solve the problem of low detection accuracy of existing detection models, this paper proposes a substation staff wear detection algorithm based on multi-branch deep convolution and dual attention. The algorithm proposes a multi-branch deep convolution (MBDC) network to add a deep separable convolution layer to enhance the completeness of feature extraction. Then, it is proposed that multimodal interaction attention (MIA) can increase the detection ability of the model to small targets, and combine the MIA mechanism with efficient channel attention (ECA) mechanism to form a dual attention mechanism to enhance the recognition accuracy of the model for small targets and occluded targets. Finally, the focus loss function and SIOU ( scylla intersection over union) are introduced as the loss function to solve the problem of positive and negative sample imbalance and accelerate the convergence speed. The experiment shows that the average accuracy of the algorithm in this paper is 84. 88%, 9. 92% higher than the original algorithm, and the overall performance is better than the comparison algorithm.