Abstract:Human identification based on biological features is one of the research hotspots nowadays. Considering electrocardio signals are relatively stable and they can be easily acquired, identification using electrocardio signals attracts the attention of many researchers. Traditional identification methods which are based on electrocardio signals usually extract the features artificially. The procedures are complicated and could be easily affected by noise. Because QRS complex is stable even though the duration of cardiac cycle changes, this research uses QRS complex to identify humans. The electrocardio signals are denoised by the wavelet threshold denoising method, and the QRS complexes are extracted to be transferred to binary images. These images are feed to convolutional neural networks to do the identification. This paper compares the performance of several neural networks of different hyperparameters, and finds the highest accuracy reaches 982%. Besides, this paper discusses some other human identification methods which are based on electrocardio signals. Results show that the method proposed in this paper is better than the others.

Abstract:The braincomputer interface system based on eventrelated potentials is difficult to detect the idle state of the brain, which limits the freedom of the subjects to output instructions at any time. Oddball paradigm can induce the N200 potential, P300 potential and transient visual evoked potential simultaneously. the working state and idle state of the brain are distinguished according to the frequency domain characteristics of transient visual evoked potential. In the working state, the timedomain features of N200 and P300 potential are used to identify the control intention of the subject. Through the experiment of sending instructions and watching video feedback on seven healthy subjects, the switch between the working state and the idle state of the brain is realized. The accuracy of the method to identify the state or intention of the brain is 98.21%, which is 50.89% higher than that of the method to identify the idle state based on the eventrelated potential.

Abstract:In order to easily and effectively identify HCM and normal heart sound, this paper proposed a new method based on timefrequency domain (TFD) feature extraction algorithm of heart murmur for HCM heart sound. Wavelet transform and principle component analysis were applied to preprocessing. The time domain envelope of the signal based on frequency conversion homomorphic filtering (FCHF) was extracted. The segmentation and localization were performed to extract the systolic energy Es and diastolic murmur energy Ed. The heart murmur scaling factor (SF) was extracted by power spectral estimation. The SF was used to weight Es and Ed, gaining the quantitative indicators for representing heart murmurs. 100 normal heart sound and 181 HCM heart sound were classified for verifying the validity of quantitative indicators. The accuracy on average was 92.97%, the best performance was 95.37%. Result represented that the extracted features can effectively classify normal heart sound and HCM heart sound. The algorithm extracted quantitative indicators of representing heart murmurs can effectively represent heart murmur. The proposed algorithm is used to provide technical basis for classification and recognition of HCM heart sounds.

Abstract:In view of the possibility of false peaks on the surface electromyography (sEMG) of patients with spasticity, leading to decreased signal differences before and after stretch reflex onset (SRO), a method for detecting SRO based on empirical mode decomposition (EMD) denoising and modified sample entropy recognition is proposed. First, the EMG signal is decomposed via EMD. Then, the soft threshold is set to denoise the decomposed signal on the basis of the sEMG signal of the subjects in resting state. Lastly, modified sample entropy is used to identify SRO. The experimental results show that the EMD algorithm can effectively remove noise from the EMG signal, and the average recognition rate of SRO under the optimal parameter of the modified sample entropy is 94%.

Abstract:In recent years, many scholars have applied machine learning algorithm to electromyogram (EMG) data analysis and achieved good results, but the main direction is gesture recognition, few scholars have applied machine learning to EMG clinical diagnosis. There are two problems:The amount of data needed is large; Machine learning is rarely used in auditory brainstem response (ABR) data analysis. Aiming at these two problems, this paper studies the application of machine learning method in computer-aided diagnosis of ABR data based on small data set. In this paper, 2 352 EMG examination reports of Sichuan Traditional Chinese Medicine Hospital were collected. A data set containing 233 ABR reports data was created by inclusion criteria and data cleaning. Then, four machine learning algorithms, linear regression, logistic regression, random forest and Artificial neural network, are used to analyze and process this data set. According to the performance comparison, the random forest is considered to be the best one, the accuracy, recall and precision of this algorithm is 0. 995 7, 0. 989 7 and 0. 950 0 respectively. In addition, this paper also compares the effect of each algorithm with and without data standardization, this experiment shows that data standardization can improve the accuracy to some extent. The random forest model outputs the importance of each indicator, the most important indicator in ABR are L_latency_5, L_latency_A and L_Interval_35, followed by L_latency_b and L_latency_4. The integration of the importance of these indicators into the upper computer software helps to improve the efficiency of clinical diagnosis and has certain diagnostic evaluation potential in clinical application.

Abstract:Finger vein recognition technology has become a research and application hotspot due to its advantages of non-contact, high security and living body detection. In view of the complexity and large amount of computation of the traditional methods for extracting the region of interest (ROI) of finger vein images, a fast ROI extraction method was proposed. The method used a three-point comparison method to locate the upper and lower boundaries of the finger and an image correction method based on the similar triangle theorem. Compared with the traditional methods, the complex process of edge optimization was eliminated and the multiplication of rotation correction was reduced, and the speed of extracting the ROI of the finger vein images can be increased by 2 ~ 3 times. The finger vein image database was used for the simulation experiments, the results showed that the accuracy of this algorithm for extracting ROI was 100%, and the recognition equal error rate was only 2. 45%, which indicated that this method has high universality and stability and can be widely used in finger vein recognition.

Abstract:Aiming at the problem of pattern confusion in traditional empirical mode decomposition (EMD) method and the fact that the overall mean empirical mode decomposition (EEMD) does not have completeness and computational complexity, an improved adaptive complementary set empirical mode decomposition - (CEEMD) method is proposed. Based on the analysis of the noise adding criterion, this method introduces peak error (PE) as the noise adding evaluation index to adaptively determine the optimal noise adding amplitude. Then, the original signal amplitude standard deviation and the noise added amplitude standard deviation are used. The ratio coefficient is used to adaptively obtain the overall average number of times for different signals. Finally, the method is applied to the MIT-BIH ECG database established by the Massachusetts Institute of Technology, and the denoising of the target signal is well completed. Experiments show that the average SNR of the proposed method reaches 19. 249 7, the RMSE is only 0. 047 3, and the average smoothness index R is only 0. 030 5. The algorithm effectively removes the original ECG signal noise, improves the signal smoothness and improves the calculation efficiency.

Abstract:Skeleton pruning is an important issue in skeleton extraction and application. A common pruning approach utilizes the thresholding of skeletal components by saliency indices based on region reconstruction. However, this approach suffers difficulties in algorithmic parameter setting, pruning outcome control, and the execution time. To deal with these difficulties, a pruning method is proposed that iteratively removes the skeletal components. The punctuating skeleton length saliency index is used, and in each iteration, the least salient skeleton branch is pruned out, until the number of the remaining branches reaches a user defined level. In order to accelerate the algorithm, the RunForest data structure is adopted for region reconstruction operations, and the reconstruction triggering strategy (RTS) is proposed to reduce the number of reconstructions needed. Experimental results on a real-world image base show that the recall of the skeletal branches of the proposed method is higher than the existing algorithm by 13 percentage points, and the precision, by about 3 points. The execution time of the algorithm with RTS is about 56% that of without. The results show that the proposed method is effective.

Abstract:Inductive coupled power transfer ( ICPT) technology provides a safe and efficient solution for power supplying of rotating structures. For the disadvantages of design installation and debugging in traditional coaxial and side-mounted structure, this paper presents a planar array coil inductive coupled power transfer structure. It analyzes the stability and offset resistance of the power supply in the mobile system by building the 3D FEM model. The experimental results show that the coupling structure eliminates the disadvantages of the position of the transmitting and receiving coils being relatively static when the traditional structure is powered. It has good robustness and portability to the moving object. It is easy to install and use, and better adapts to the power supply of mobile devices. When this structure is powered, the speed of the rotating shaft can be acquired at the same time by the small fluctuation of the induced voltage amplitude in receiving coil.

Abstract:With the continuous development of intelligent transportation systems ( ITS ), vehicular communication, as its vital technology, has received increasing attention. As the propagation channel of vehicular communication, the characteristic analysis of wireless channel is fundamental. This paper studies the signal propagation model of a mobile car passing through a base station. Based on the theory of scattering clusters, a narrowband single-loop model suitable for large-scale multiple-input multiple-output ( MIMO) channels is established. By setting the scattering cluster parameters that match the actual scene to simulate and analyze the channel, the model will be more in line with the characteristics of the actual channel and have a better evaluation effect than the classic single-loop model. In the simulation model, the time autocorrelation function ( ACF), spatial cross-correlation function and frequency crosscorrelation function (FCF) of the channel are mainly studied. At the same time, a comparative analysis of different scattering cluster scenes proves that the model is more practical.

Abstract:It is difficult to diagnose the early weak fault of rolling bearing because it is easily affected by noise. In view of the shortcomings of the original ITD and cubic spline interpolation improved ITD algorithm and the difficulty in selecting the filter length parameters of the maximum correlation kurtosis decomposition (MCKD) algorithm, an improved ITD (QH-ITD) algorithm based on the quartic Hermite interpolation and an improved MCKD (AMCKD) algorithm based on variable step length search parameters optimization are proposed. Firstly, QH-ITD algorithm is used to decompose the fault signal of the original rolling bearing, then the kurtosis index and the correlation number are used to screen the corresponding component signals for reconstruction, then the AMCKD algorithm is used to reduce the noise of the reconstructed signal, finally, the Teager-Kaiser energy operator is used for demodulation, the fault characteristic information is extracted and the fault type is determined. It is verified that the proposed method can effectively diagnose and identify the early weak fault of the rolling bearing by simulating the damage fault diagnosis experiment and the early weak fault diagnosis experiment of the bearing with the whole life cycle.

Abstract:In order to study whether the electromagnetic signal can be used as a reliable earthquake precursor signal, an electromagnetic sensing probe based on inductive magnetic sensor for earthquake monitoring is designed. The probe is composed of a signal processing and an acquisition circuit, which can support electromagnetic signal acquisition in very low frequency (VLF) and ultra low frequency (ULF) from 10 Hz~ 10 kHz in a wide dynamic range 0. 1~ 1 000 nT. The sensitivity of the probe is greater than 20 mV/ nT@ 0. 1 Hz~ 10 kHz, 18-bit resolution. Besides, the probe has passed many reliability tests and has been applied to 221 stations including the Sichuan and Yunnan National Test site for electromagnetic monitoring. Field observation experiments show that the probe can capture a highly correlated electromagnetic signal effectively.

Abstract:This model design of high frequency wireless induction signal transmission device is based on HFSS simulation software, in order to improve the complex manufacturing process of special wired drilling tool. By establishing induction and resonance simulation models, the simulation signal transmission characteristics are calculated and compared with the physical model to verify the influence of the parameters of the design model, such as turns, pitch, wrapping material, on the transmission characteristics. By analyzing and correcting the design scheme through the simulation results of the model, a simulation model with high consistency with multiple physical models is obtained, which can guide the design of physical models, and greatly improve the efficiency, reduce the cost and save the time of R&D. Through the model simulation method designed in this paper, the feasibility of wireless induction coupled transmission is verified. It is highly consistent with the physical model, deviation can be less than -5 dB,which provides an important basis for the physical production.

Abstract:In the process of unmanned aerial vehicle (UAV) return without a reference map, the scene matching between the real-time map and the waypoint is the key to the success of the UAV return. In order to improve the real-time and robustness of scene matching, a UAV scene matching algorithm based on FAST corner detection and FREAK descriptor is proposed. Firstly, in order to improve the shortcomings of FAST corner detection method such as no scale invariance and redundant feature points, a multi-scale gridded feature detection method based on FAST corner is proposed. Next, the FREAK binary descriptor is simplified to improve the matching speed. Then, the K-nearest neighbor ratio method and RANSAC method are used for the initial and fine matching of the features, and a positioning model is established to obtain the actual distance between the waypoint and the current position of the UAV and orientation information. Finally, experiments are performed to verify the performance of the algorithm. The deviation of the positioning direction of the proposed algorithm is within 1 °, and the deviation of the image plane distance is stable within 0. 6 pixels, the running time is 0. 43 s, which is much shorter than the processing time of SIFT and SURF algorithms. In the case of conditions such as scale transformation and noise, compared with algorithms such as SIFT and SURF, the proposed algorithm has achieved a better correct matching rate and better robustness. The experimental results show that the proposed algorithm is robust and fast, especially in perspective transformation, it is more suitable for UAV vision-assisted navigation.

Abstract:Overlapping regions often occur when non-overlapping community structure is obtained by clustering granulation method. The nodes in the non-overlapping parts of the community of the left side between two communities with overlapping parts were defined as positive regions. Then, the nodes on its right are denoted as the negative region, and nodes in the overlapping parts are denoted as the boundary region. In order to achieve better community structure, it is necessary to divide the nodes in the boundary region into nonoverlapping parts. Submodular optimization is widely used in machine learning, If the objective function has sub-modularity, it exists a simple greedy algorithm which can approximate the optimal solution of the problem with constant factor in polynomial time. In this paper, submodular optimization is introduced into the processing of nodes in overlapping communities. and a community detection algorithm (SO-CDA) for boundary region processing based on submodular optimization is proposed. The device location function is defined for submodular optimization, the partition of overlapping nodes is converted to the maximization of submodular function. The experimental results on seven real networks show that SO-CDA can effectively divide communities and achieve more stable performance.

Abstract:Helicopter rotor blades are prone to fatigue damage in flight. To solve the damage location problem, a damage monitoring and locating system was constructed. With the acoustic emission signals of the damage sources extracted by the kernel principal component analysis (KPCA), combining the support vector machine ( SVM) and its regression function, the damage sources of the rotor blades were located. The regional location accuracy after feature extraction is 100% and the average regression error is 7%, which are better than the original data location. Therefore, this method can effectively locate the damage source of the rotor blade, reduce the dimension of input data and the amount of calculation.

Abstract:Autonomous mobile robots have been widely used in national defense, disaster relief and other fields. As a typical environmental target, the stair area needs to be accurately recognized by the robot. Obstacles placed in the stair area will destroy the stairs’ plane and edge features that traditional staircase recognition algorithms need to extract, resulting in the staircase area cannot be accurately recognized. Aiming at this problem, a point cloud-based stair area detection and recognition algorithm in a complex environment is proposed. The algorithm first uses the region growing method to segment the target region and selects the regions suspected to be the vertical step of each level of the stair by fitting the plane normal direction of each region, then processes the each level stair area to segment obstacles and obtain the boundaries on the vertical plane of each level of the stair. Finally, the stair model and obstacle position are obtained according to the boundary position of each level. The experimental results show that the algorithm has better robustness, can recognize stairs in various complex environments and get barrier-free stair area. The constructed 3D model of stairs has a size error of less than 7%, which is higher accuracy. The algorithm can achieve better detection and recognition results compared with traditional stairs recognition algorithms.

Abstract:Minimum entropy deconvolution (MED) is an effective technique for detecting impulse-like signals such as bearing fault or gear fault signal, but there is still a deficiency in this method, that is, a parameter of the filter length in this method has to be set before using. Unfortunately, the selection of this parameter value can only be chosen through the human experience. In order to overcome this limitation, an optimal selection indicator based on Kurtosis, permutation entropy (PE) and signal energy is proposed in this study. By virtue of this indicator, the optimal filter length can be selected to filter the raw signal better. Then, an enhanced energy operator named envelope-derivation energy operator ( EDEO) is used to extract the fault characteristic frequency from the filtered signal. The experimental results show that, compared with the conventional methods, this proposed method can effectively extract the bearing fault characteristic frequency under harsh working conditions and obviously highlight the amplitude of the bearing fault frequency.

Abstract:The traditional Nyquist sampling theorem brings a pressure in the sampling and storage devices for array parameter estimation in radar signal. A joint carrier frequency and two-dimensional direction of arrival (2D-DOA) estimation method are proposed based on double L shaped array random demodulation ( RD) structure. This method based on the estimating signal parameter via rotational invariance technique (ESPRIT) decomposition method. The algorithm constructs a cross-correlation matrix between different antennas. The corresponding pairing method is also provided. Simulation results show that this method can estimate the carrier frequency and 2D-DOA of the target signal from the sub-Nyquist samples, and reconstruct the time domain waveform of the original signal.

Abstract:Aiming at the problem that the temperature field analysis modeling is difficult due to the irregular air gap between the two windings in the stator slot of the switched reluctance motor, an improved modeling method for the double winding air gap is proposed. The calculation method of the total cross-sectional area of the air gap between the two windings of the motor is analyzed. The model structure of the double-winding air gap and the method for determining the optimal air gap width are studied. Based on this, a threedimensional finite element model of the switched reluctance motor is established. The temperature field is analyzed by finite element method, and the temperature field distribution of the switched reluctance motor is obtained. The temperature field under different air gap width is analyzed and the numerical fitting method is used to obtain the air gap width and the corresponding temperature. The function relationship, using the function relationship and the corresponding measured temperature value, can obtain the optimal air gap width of the double-winding air gap model. This method effectively improves the accuracy of the temperature field analysis results of the switched reluctance motor, it has a good application value.

Abstract:Solar energy is an attractive source of electricity. Solar photovoltaic cells are the basis of solar power generation systems. However, various types of defects in solar photovoltaic cells seriously affect the photoelectric conversion efficiency and service life of photovoltaic cells. To effectively detect these defects, a defect detection method based on a block case deletion model is proposed. First, the solar photovoltaic cell image using Fourier transform is preprocessed, it removes the bus bar and adjusts the brightness and contrast, and divides the image into blocks. Then, in the processed image, all abnormal blocks are found and all of them are removed by using the case deletion model. The background of the image is reconstructed from the remaining image patches by a non-linear regression model. Finally, the defect area is highlighted by the difference between the image waiting for checking and the resulting background image. The experimental results show that the proposed method can effectively detect many kinds of defects in Solar cells, such as micro-cracks, breaks and fragment, etc. the method is used to experiment with 313 solar photovoltaic cell images. For 158 non-defective images, the test results are normal. The remaining 155 images containing defects such as cracks and broken gates have only 5 images mis-detected, and the detection rate of defective images is 96. 77%.

Abstract:With the increasing complexity of spacecraft structure and thermal design, it is more difficult to implement surface temperature measurement technology, and the temperature measurement area tends to diversify. Therefore, there is an increasing demand for the application of non-contact temperature measurement technology in spacecraft thermal test. This paper takes the application of infrared temperature measuring equipment in vacuum and high-low temperature environment as the research object, designs the thermal protection scheme and device of the equipment, and simulates and analyses the thermal protection of infrared temperature measuring equipment based on node network method. Through physical test, the device can effectively realize the thermal protection of equipment in vacuum and high-low temperature environment, and ensure that the equipment is in the normal working temperature range and its temperature measurement algorithm model is not affected, the accuracy of temperature measurement is better than ±2℃ . It meets the requirements of equipment in space environment test.

Abstract:In order to identify three types of carbon structural steels whose metallographic structures are ferrite and pearlite. This paper proposes a metal identification method based on convolutional neural network. Convolutional neural networks can efficiently implement classification with complex environmental information, ambiguous inference rules, and flawed samples. The metal identification platform was built based on eddy current non-destructive testing technology and convolutional neural network. First, 8 high-frequency points are randomly selected from the bandwidth of the eddy current sensor, and the metal information that under each frequency point is separately collected by this eddy current sensor. Then, this information is imaged through data processing such as Fourier transform and coordinate transformation. Finally, the identification model is obtained by convolutional neural network. The results show that the proposed scheme can identify metals without damaging the metal compared to the traditional method. The accuracy of the CNN model for all three metals increased to 92. 33%, which is superior to the BP neural network (86. 20%).

Abstract:As the outputs of resolvers are amplitude-modulated signals containing angular position information, demodulation is necessary to convert analog modulated signals to digital ones. In resolver-to-digital converters, phase locked loop (PLL) is usually adopted as the algorithm for demodulation. However, conventional PLL is only a classical type-II tracking loop which cannot overcome the contradiction between dynamical and steady-state performance. Especially for the measured angular position with high dynamic variations, the demodulation error is often large. In order to improve the demodulation accuracy, the problem of resolver-digital conversion is transformed into the problem of angle tracking control, and a sliding-mode variable-structure demodulation algorithm is proposed. This algorithm can retrain the effects of the model uncertainties on demodulation accuracy caused by speed variation by using switching control term. Experimental results show that the proposed algorithm is feasible.

Abstract:Rolling bearing prognostic and health management (PHM) method can extract a large number of fault characterization data. Those data are of great potential value, because of their characteristics of high-dimensionality and high-redundancy. However, direct analysis and utilization of them are impossible. Therefore, aiming at reducing the redundancy data and screening sensitive features, a two-stage feature selection algorithm is proposed. In the first stage of the method, the Laplacian score (LS) is used to sort the original features based on their locality preserving power, and the mutual information-based clustering algorithm is utilized to remove the redundant features of the original feature set. In the second stage, the Mahalanobis-Taguchi system (MTS), as a useful multivariate pattern recognition method, is employed to comprehensively evaluate the remaining features, unearthing features which are prone to fault classification. The verification results of the bearing degradation simulation test data show that the proposed two-stage feature selection algorithm can effectively remove redundancy and improve the accuracy of fault monitoring. This method can be effectively applied to the initial fault detection of rolling bearings.

Abstract:Accurate prediction of the initial position of the permanent magnet synchronous motor has an important impact on the control of motor starting process, but identification of the rotor′s initial position without sensor has many advantages on the motor control. In this paper, an initial position prediction method for high precision PMSM based on pulse voltage vector method is proposed. Based on the analysis of the relationship between the magnetic circuit saturation with rotor position of the PMSM, by adding the forward and reverse pulse voltages to any two phases of the motor, the induced voltage without the current phase is measured, and the relationship between the magnitude of the induced voltage and the rotor position is obtained. The neural network is trained the test data to fit this relationship to form a rotor initial position estimating device. The simulation experimental results verify that the method can overcome various errors caused by the motor centralized parament model and has extremely high prediction accuracy.