Abstract:Harmonic injections on the demand side are a growing threaten to the power quality in the power utilization grid. Harmonic identification is important to improve power quality in large scale power utilization grid. The smart electrical information network can measure the electrical information in the network with distributed smart meters. This distributed measuring network can improve the harmonic identification accuracy. This paper proposed a distributed harmonic identification method using the ChowLiu information tree. The harmonic currents at the Point of Common coupling(PCC) are extracted by independent component analysis. The ChowLiu information tree offers the topological structure of the signals and recovers the magnitude and location of the harmonic sources. A practical testbed for networked measurement is built to verify the proposed harmonic source identification method. Experimental results show that the proposed method can identify the harmonic sources in a high accuracy under networked sensing environment with acceptable real-time performance.

Abstract:With the rapid development of logging while drilling (LWD) technology, the realtime transmission of multidimensional formation data is particularly important. Aiming at improving the low data rate, high cost, and nonreal time for traditional transmission approaches, this paper proposes a novel acoustic transmission scheme based on noncontinuous orthogonal frequency division multiplexing (NCOFDM). Firstly, the frequency response function of drill strings channel is simulated through transfer matrix method (TMM), and on this basis, a channel estimation method is proposed to obtain frequency response of each subchannel. Then, a received array with two accelerometers is utilized to receive the acoustic signals with phase shifting, and the surface strong noise interference of uplink transmission can be reduced through the correlation of the received signals. Finally, a set of test device of communication is designed to validate the transmission performance under the channel condition of experimental drill strings. Simulation and test results show that the designed transmission scheme has high data rate and high reliability.

Abstract:There are two advantages including high accuracy and strong freeinterference for the magnetoacoustic synchronous method, which is a fine location method and is widely used. However, it is difficult to effectively detect the audio signal. Because the wavelet packet transform in detecting whether the normal signal contains a anomaly transient or not , an audio signal enhancement algorithm based on wavelet packet transform decomposing signal, adaptive filter estimating noise and genetic algorithm optimizing signal combined reconstructing signal is presented. The adaptive filter needs less for prior knowledge of the signal and noise, and genetic algorithm does not depend on the specific problems. Experimental results validate its high accuracy and strong robustness. In particular, it can be applied to the extracting audio signal under the condition that the cable fault discharge is weak. Therefore, it is a potential eligible solution to strict constraint of the background noise and narrow recognition range in pinpointing of cable failure location.

Abstract:NonGaussian impulsive noises widely exist in real world, which seriously impacts on the performance of the adaptive filtering algorithms based on l2 norm optimization criterion. In various types of adaptive filtering algorithms, the affine projection sign algorithm (APSA) combines the good convergence feature of the affine projection algorithm (APA) and the suppression capability of the sign algorithm (SA) for nonGaussian impulsive noises. Thus, it has good performance in nonGaussian impulsive noise conditions. However, the step size of APSA is fixed and does not consider the system sparsity, so that there are certain limitation in parameter selection and convergence speed. Therefore, in this paper, the variable stepsize method and proportion matrix thought are combined together, the step size function is introduced, and a robust affine projection sign adaptive filtering algorithm  the variable stepsize improved proportion affine projection sign algorithm (VSSIPAPSA) is proposed, which can not only alleviate the contradiction between convergence speed and steadystate misadjustment, but also increase the adaptability of the system to different sparsity and noise characteristics. Theoretical analysis and simulation results demonstrate the effectiveness and robustness of the algorithm.

Abstract:机械系统存在的外部环境干扰、变工况条件以及无法直接测量等因素，导致获取的数据常常不满足传统机器学习的两个前提：训练与测试数据分布相同以及目标诊断数据量充足，从而影响诊断模型的泛化能力。针对上述问题，提出一种基于辅助数据的增强型最小二乘支持向量机（LSSVM）迁移学习策略，用于数据量不足时的轴承故障诊断。其中利用递归定量分析（RQA）提取非线性特征并与传统时域特征相结合以提高诊断精度。诊断分类器通过改进传统LSSVM模型，在原目标函数和约束条件中分别增加辅助集的惩罚函数和约束条件，最终得到加入辅助集的函数估计，从而将该算法推广至迁移学习。此外，类内类间距离指标用于描述特征区分性，并提出4种辅助数据集的使用方法，从而构建迁移学习为框架的诊断模型。球形轴承的振动信号试验结果表明，相比传统机器学习，在目标振动数据较少条件下所提模型在轴承故障诊断时性能提升显著。

Abstract:Accurate identification of distribution network fault type can provide directive guidance for the operation and maintenance personals of transmission lines. In this paper, a new fault rootcause identification method of distribution network based on the time and frequency characteristic analysis of fault waveforms is proposed. Through model building and theoretical analysis of different types of fault waveforms, the characteristic parameters are proposed, which can characterize different kinds of fault waveforms from time domain, frequency domain and arc model. The formulas for calculating characteristic parameters from fault waveform data are proposed. Multiple characteristic parameters are fused, and based on which the classifier is built; the fault types of the distribution network caused by different rootcauses are identified automatically through detecting and analyzing the characteristic parameters of the input waveform data. Finally, the proposed classification method was tested and verified using 136 groups of different field fault waveform data provided by EPRI; the test results show that the successful identification ratio reaches to 90%, which verifies the feasibility of using fault waveform time and frequency characteristics to realize the fault type identification of distribution networks.

Abstract:In order to improve the detection efficiency and accuracy of ellipse/similar ellipse in complex image, an improved ellipse detection method is proposed based on the randomized Hough transform (RHT) with character string constraint. The noneffective samplings and number of accumulations of RHT are greatly reduced with the help of the character string geometric constraint and the normal constraint at the endpoints of the character string. Through analyzing the effective distribution of the pixels in edge image, the twodimensional array accumulator for storing the character string endpoint information is established, then the ellipse power is used to eliminate the interference of the false ellipse centers before extracting the boundary points, which not only improves the reliability of space sampling points, but also decreases the accumulation probability of noneffective sampling points. Experiments indicate that the proposed algorithm possesses high operation speed and detection accuracy in ellipse detection, as well as has strong resistance to large deformation, severely contour missing and noise.

Abstract:Aiming at the problem that quadrature error exists in the output signals of grating reading heads, a quadrature error realtime compensation method based on Coordinate Rotation Digital Computer, (CORDIC) algorithm is proposed. Aiming at the problems that there exist large error interval in the sinecosine signal angle solution and low angle solution sensitivity in the sinecosine signal peak interval in the CORDIC algorithm, a vector pattern biiteration method is introduced to suppress the angle solution error caused by the CORDIC algorithm due to too quick iteration convergence speed, and the local table lookup method is used in combination to eliminate the angle solution error in the signal peak interval. The quadrature error compensation process consists of three steps: phase solution, phase compensation and signal reconstruction. Taking the solved angle value as the object, the angle errors are compensated in real time in complete period. The CORDIC rotation mode is used to reconstruct the cosine signals from the compensated angle values so as to realize the realtime compensation of the quadrature errors of Moiré signals. This compensation method was implemented on an FPGA platform, and the phase difference measurement and compensation performance of this method were verified. Test results indicate that with the signal quadrature error of ［1°, 10°］, the phase measurement error is within ±0.04°; for the signals with different frequencies and phase differences, after compensation the maximum phase error is within ±1°, the mean error is within ±0.1°, and the mean square deviation is within 0.5°, which proves that the proposed method can effectively achieve the realtime compensation of the quadrature errors of Moiré signals.

Abstract:When dynamical balance machine is used to measure the imbalance of the workpiece with large eccentric mass, due to the large imbalance the vibration signal usually exceeds the linear range of both the dynamical balance machine and measurement system, which results in grave error. In order to improve the accuracy of imbalance measurement, a method is proposed, which utilizes a controllable electromagnet to generate controllable electromagnetic force for attenuating the unbalance vibration synchronously, so as to avoid the vibration signal exceeding the liner range. The basic operation principle of the method is discussed, the equivalent magnetic circuit model of the electromagnet is established, and the relationship between the coil current and dynamic electromagnetic force is analyzed with finite element modeling. The influence of some factors, such as air gap and temperature on electromagnetic force is analyzed. The control system and control strategy are also introduced. Dynamical balance test experiment was conducted, the test results indicate that the proposed measurement method can effectively improve the dynamical balance test accuracy of the workpiece with large eccentric mass.

Abstract:Highprecision radar cross section (RCS) measurement has high requirement for background environment. It is difficult to eliminate the clutter effect through background vector cancelling when strong interference exists in background environment. This paper proposes a high accuracy RCS measurement method based on imaging extraction to separate and extract the target scattering signal from background clutter, thereby the measurement accuracy is improved. Firstly, the mathematical relationship between the image and RCS is derived, then the echo measured in the revolving stage mode is utilized to conduct the imaging processing and obtain the 2D image of the target area. Then, the 2D image of the target is extracted from the imaging area, and the RCS of the target is obtained through spectrum transform and calibration. The simulation results show that in RCS measurement under interference the proposed method can improve the measurement accuracy by 3~5 dB, and measure the target with weak scattering. Experiment results show the effectiveness and accuracy of the imaging extraction method.

Abstract:Aiming at the problem that in the generation process of travelling magnetic field in existing magnetic field linear time grating displacement sensor, the tooth and slot influence the velocity uniformity of travelling magnetic field, this paper proposes a linear time grating displacement sensor based on linear array of planar coils. The sensor structure without tooth and slot improves the velocity uniformity, can realize the high precision measurement under large polar distance. In the sensor, two phase exciting coils exerted with orthogonal signals are arranged alternately to form a linear array of planar coils, the generated travelling magnetic field is acquired by the pickup coils to obtain electric travelling signal, and the displacement quantity is obtained after signal processing. The magnetic field finite element analysis software is used to conduct modeling simulation on the sensor, the measurement error is acquired from the simulation results; the measurement error is analyzed and traced through theoretical analysis, and the sensor structure is optimized based on the analysis result. A sensor prototype was developed according to the analysis and optimization results, and the accuracy experiment was conducted. The experiment results show that the measurement accuracy of the sensor is ±1 μm in the range of 240 mm, and high precision measurement is achieved.

Abstract:Aiming at the problems that current absolute linear magneticfield type time grating displacement sensors do not provide incremental linear displacement feedback for full closedloop numerical control system, in this paper the measurement standard transformation method is adopted to process the spatial displacement information of the absolute linear displacement magneticfield type time grating sensors in time domain. The inherent relevant of the sampling data series of the absolute linear displacement time grating sensor is analyzed using time series algorithm, and the selfadaptive recursive algorithm is established. The measured linear absolute displacement data sampled in timetriggered pattern are employed as sample set, and the absolute linear displacement for the next sampling period is calculated through recursion. Then the incremental time clock pulses represented by the absolute linear displacement of the time grating sensor are sent continuously using pulse width modulation pattern during the next sampling period. In this way, the absolute linear time grating sensor is transformed into an incremental linear one. Experiment results prove that the dynamic displacement measurement error of the proposed incremental linear time grating sensor is within ±2 μm in the range of 76.604 mm. With the proposed method, the absolute linear time grating sensor can be applied for incremental linear displacement feedback required in full closedloop numerical control system.

Abstract:A force estimation method based on the surface electromyograph(sEMG) and generalized regression neural network (GRNN) is proposed for the demand of the force control of the intelligent EMG prosthetic hand. First, the experimental platform is introduced. The acquisition of the sEMG, the feature extraction of sEMG and the construction of GRNN are described. Then, the sEMG in the hand motions are detected by the EMG sensors with which eight different positions of arm skin surface are attached on. A three dimension force sensor is adopt to measure the force output by the human's hand. The multi channels of the sEMG and the force are measured synchronously. Characteristic matrix of the sEMG and the force signal are used to construct the GRNN. The mean square error is employed to assess the accuracy of the estimated force. Experiments are implemented to verify the effectiveness of the proposed estimation method. The experimental results show that the force output by the human's hand can be estimated by the used of sEMG and GRNN.

Abstract:In this paper we introduce a new passive multioutput lowfrequency vibration sensor that can measure acceleration and speed of the waves. Using the passive servo feedback control technique, the sensor shows the characteristics of a speed pendulum, which, however, is not the usual speed pendulum in largedamping state. A detailed mathematical analysis is carried out on the characteristics of the pendulum body. Simultaneous measurements of both acceleration and speed can be achieved using only one transducing mode. It is proved that the speed output is the output of the speed pendulum speedometer and the acceleration output is the output of the speed pendulum accelerometer. We also analyze the characteristics of the sensor, especially analyze the acceleration output characteristics in detail. Finally, the sensor was tested, and good frequencyresponse characteristics of acceleration output and speed output of the sensor in the frequency range from 0.1 Hz to 100 Hz were obtained, which satisfies the requirement of lowfrequency engineering vibration measurements, and the goal of using one sensor to measure both acceleration and speed at the same time is achieved.

Abstract:Aiming at the problems of nonlinear model, strong coupling, as well as model parameter uncertainty and external disturbance of workclass remotely operated vehicle (ROV) in path tracking process, a command filtered adaptive backstepping control strategy is proposed based on nonlinear disturbance observer (NDO). The NDO is used to observe the model uncertainty and external disturbance. The command filter is adopted to avoid directly computing the analytic derivative of the virtual control quantity. The unobservable disturbance is compensated with adaptive law. Lyapunov stability theory is used to prove the asympotic stability of the path tracking error system. The simulation experiment results show that the designed controller can achieve precise path tracking and has good robust characteristic.

Abstract:In order to solve the image spin problem in scanning imaging process of aerial camera, a fourchannel bilateral control system is adopted to compensate the image spin. In the system, the internalloop compensator is designed based on robust internalloop compensator structure. Firstly, Hinfinity mixed sensitivity optimization method is adopted to solve the robust controller in robust internalloop compensator. Secondly, the filter of the system is determined according to the reference model, then the systematically designed internalloop compensator is obtained. It is very important to improve the interference suppression performance as much as possible while guarantee the robust stability of the system, then the bilateral control image spin compensation accuracy of the system is improved through compromising between interference suppression and robust stability. Experiment results show that the proposed method can effectively suppress the influence of equivalent disturbance and improve the image spin compensation accuracy. The maximum and root mean square (RMS) values of the position compensation errors under dynamic scanning are (1.81×10-3)°, (5.224 74×10-4)°, respectively. Compared with traditional control system using disturbance observer, the compensation errors under dynamic scanning for the proposed system are reduced by 41.99%, 41.73%, respectively. So the image spin compensation accuracy of the fourchannel bilateral control system is improved.

Abstract:The hysteretic nonlinearity of the piezoelectric actuator (PEA) will greatly reduce the pointing accuracy of a precision pointing system for intersatellite laser communication, and affect the acquisition of beacon and link stability. To address this issue, an improved PrandtlIshlinskii (PI) model based on the PLAY hysteresis operator and parameter identification method is presented to characterize the hysteresis of the PEA. Based on this basis, a feedback forward linearization method for the piezoelectric is proposed and its effectiveness of mathematical model and linearization is verified. The experimental results show that the inverse of this model with maximum error less than 1% is employed for feedforward compensation of the PEA hysteresis to obtain an approximately linear system. The improved PI model entails less computation than the original PI model. The performance the feedforward compensation is verified by experiments using control signals with different frequencies, and with constant and decreasing amplitudes respectively. The results show that the feedforward inverse model compensation can reduce the linearity error of the PEA from 5% to less than 1%, which PrandtlIshlinskii model in computational complexity simplified from O(n) to O (1).

Abstract:Nonlinear piezoelectric hysteresis reduces the openloop control precision and closedloop operation bandwidth of the piezoelectric deformable mirror (DM), which limits its applications in adaptive optics (AO) system. In order to overcome the hysteresis influence of the piezoelectric deformable mirror, the PI hysteresis model was adopted to describe the nonlinear hysteresis behavior of a unimorph piezoelectric deformable mirror and achieve the high precision openloop control of the unimorph piezoelectric deformable mirror. Firstly, the mathematical model of PI hysteresis was established. The least square method was used to identify the weight value of the PI hysteresis model. The weight value and threshold value of the PI inverse model were calculated. The control voltage of the unimorph piezoelectric deformable mirror after hysteresis elimination was obtained. Then, an adaptive optical test platform based on Hartmann wavefront sensor was built. The openloop control experiment of the defocus wavefront using a ring actuator of the unimorph piezoelectric deformable mirror was carried out. The experiment results show that the voltagedeformation hysteresis of the unimorph piezoelectric deformable mirror is reduced from 9.3% to 1.2% after hysteresis elimination. The openloop reconstruction accuracy of the defocus wavefront is improved by more than 70%, which proves that the proposed PI hysteresis model can be effectively applied in the openloop control of the unimorph piezoelectric deformable mirror.

Abstract:K2 algorithm is the classical learning algorithm of Bayesian network structure. Aiming at the problems that K2 algorithm depends on the maximum number of parent nodes & node order and ant colony optimization algorithm has large search space, this paper proposes a new Bayesian structure learning algorithm  MWSTACOK2 algorithm. Firstly, through calculating the mutual information, the algorithm establishes the Most Weight Supported Tree (MWST) and obtain the maximum number of parent nodes. Secondly, ant colony optimization algorithm is adopted to search the Most Weight Supported Tree and obtain the node order. Finally, combining with K2 algorithm, the proposed algorithm can obtain the optimal Bayesian network structure. The simulation experiment results show that the proposed algorithm not only solves the problem that K2 algorithm relies on prior knowledge, but also reduces the search space of ant colony algorithm, simplifies the search mechanism and obtains good Bayesian structure. The proposed algorithm was applied to the operation data of the cement rotary kiln in Jidong Cement Company, established the Bayesian network structure model of the cement rotary kiln and achieved precise and rapid fault diagnosis.

Abstract:The metal magnetic memory (MMM) method can be used to predict and evaluate the microdamage region of ferromagnetic metal components in early stage; however, the magnetic memory signal is easily disturbed by external strong magnetic field, which can lead to the deviation of the detection result. In order to study the influence rule of magnetic field intensity on magnetic memory signal, the Kp perturbation algorithm is used in this paper, the multiprimary cell magnetic mechanical model is established based on the number of effective Bohr magnetons p in the K space, and the quantitative changing relationship of magnetic mechanics under the action of external magnetic field is calculated. The research results show that under the action of external magnetic field, the electron orbit motion is strengthened, the crystal lattice structure is distorted, and the magnetic moment of the atom is increased. When the magnetic field is small, the magnetic memory signal increases with the increasing of the external magnetic field intensity linearly; When the magnetic field intensity reaches a critical value, the magnetic moment of the atom is approximately equal to the magnetic moment of the independent atom, and the magnetic memory signal tends to be a constant value; When the magnetic field intensity is greater than the critical value, the magnetic memory signal of the stress concentration area will be covered by the magnetic field. The experiment results of the magnetic memory testing obtained in this paper and theoretical analysis results have very good consistence.

Abstract:Due to the internal pressure and other environmental effects, the pressurized water pipe has potential problems such as leakage and burst. Since there is a certain pressure in the pipe, the leakage point will cause the inner pipe walls to vibrate when the pipeline leaks. Thus in this paper, a new method is proposed to monitor the pressurized water pipe with a distributed optical fiber vibration sensor(DOFVS) based on the phasesensitive optical timedomain reflectometer. The indoor test results indicate that the proposed method can detect the aperture which is equal or larger than 4 mm for a steel pipe with DN91 cm×EN2 cm and 0.4 Mpa inner pressure. The field measurement results show that, the leaking signal could be effectively acquired when the leakage rate is larger than 11 L/s, where the pipe is DN200 cm×EN2 cm steel made pressure pipe and the inside pipe pressure is around 0.27 MPa. Besides, we adopt the wavelet threshold denoising method to effectively suppress the high frequency noises in the monitoring signal, and this method achieve good results.

Abstract:Oil pipeline leak is a continuous and dynamic process affected by many factors (e.g., corrosion, wear, weld defects, vibration, erosion and manmade destruction). The method based on pressure signal detection and Gaussian assumption signal analysis cannot meet the characteristics of multivariable, strong coupling and dynamics. In this article, the operating and environmental parameters associated with the pipeline leaking are comprehensively considered. A novel oil pipeline leak detection method based on Dynamic Kernel Independent Component Analysis (DKICA) is proposed to solve the timingsequenceautocorrelation problem of the pipeline monitoring parameters and enhance the detection accuracy. Firstly, the optimal order of the model parameters is confirmed by the determination characteristics of dynamic (DOD) algorithm to reduce the autocorrelation among the monitoring parameters. Secondly, the Kernel Independent Component Analysis (KICA) is utilized to extract the independent component in kernel principal space. Finally, the pipeline leak is monitored by T2, SPE and the combined index of the independent components. Experimental results indicate that both the missing and false detection accuracies of the combined index D2 are much lower than those of the SPE and T2 separately. Additionally, both the missing and false detection accuracies of the 2order DKICA are much lower than those of KICA, due to the consideration of the dynamic characteristics. It verifies the feasibility and effectiveness of the proposed method based on DKICA for the oil pipeline leak detection.

Abstract:Aiming at monitoring the batch process with complex nonlinear characteristic, a multiway kernel entropy component analysis (MKECA) method based on the angle structure statistic is proposed. In this method, the process data is firstly preprocessed, and then the principal component matrices of the batch process data are extracted by KECA. Research shows that KECA reveals angular structure relating to the Renyi entropy of the input space data set, and angular structure statistic is constructed using the principal component matrix structure. And then the control limits are calculated by the kernel density estimation algorithm. Finally， through the simulation of the penicillin fermentation and the actual production process of recombinant, the experiment results show that the proposed method effectively uses the structural information of the principal components compared to the traditional method of process monitoring. So error rate and false alarming rate are significantly lowered.

Abstract:In order to detect weak signal with unknown frequency, this paper presents a novel method for detecting the unknown frequency signal by combining the variance peak value of the Duffing oscillator and genetic algorithm (GA). Firstly, the impact of periodic driving signal with different frequencies, initial phase angle and noise on the system operating state are analyzed. Second, the corresponding relationship between the system output variance and the system running state is studied. The effect of the test signal frequency, as well as the phase difference between the periodic driving signal and the test signal on the state variables variance and the state transition time are discussed. Finally, the new method is proposed, which using a number of Duffing oscillators with different initial phases to cover all phase signal detection. Moreover, the proposed method combines with genetic algorithm to obtain the detected signal frequency by optimizing the calculation of peak value of system output varaince under different input frequency signals. This method resolves the limitation of existing weak signal detection method based on the chaotic oscillator, in which the signal frequency needs to be known. Experimental results show that the method can accurately and flexibly detect the frequency of the test signal and has strong adaptability. This workt provides a novel approach for the weak signal detection.

Abstract:Recovery voltage method (RVM) is to study insulation aging status of power transformer. Based on a lot of test data of recovery voltage, the fuzzy rough set theory is proposed and used to assess the oil paper insulation status of transformer. And the assessment system of oil paper insulation of transformer is constructed. Firstly, the evaluation index of oilpaper insulation status of transformer is identified based on recovery voltage feature parameters. Secondly, the membership function of fuzzy partitioning feature parameters for transformer test data is obtained by using FCM clustering algorithm. Then, the fuzzy attributes are reduced for the assessment table of oil paper insulation statue in accordance with the discernibility matrix, and the assessment rules of oilpaper insulation status are extracted. Finally, the assessment system of oilpaper insulation status is built based on the historical database. The experiments demonstrate that the assessment system is effective and feasible which provides a new solution for the assessment of transformer oilpaper insulation status. This work is of practical value in actual engineering applications.

Abstract:Echo receiving system is an important part in the optical system of frequency modulated continuous wave noncooperative target ranging; and its receiving characteristics have certain influence on the ranging precision of the system. In order to conduct the design and analysis of the system, optical design software was used to conduct the modeling of the receiving system with Cassegrain telescope and multimode fiber as the main body, and the parameters of the Cassegrain telescope were determined through analyzing and optimizing the aberration and optical receiving efficiency. The Gaussian scattering model for the backscattering of the target was established; a complete receiving optical path was built to conduct the experiment; the received power of the system was analyzed quantitatively under different conditions, and the rules between receiving efficiency, focusing position and target distance, surface scattering were investigated. The research results show that for the sample blocks processed with the same method, the system receiving power is negatively correlated with the roughness and target distance, while the system focusing distance is positively correlated with the roughness and target distance.

Abstract:The traditional clustering process of multiparametric flow cytometry data analysis is complicated, nonautomated and timeconsuming. To overcome this limitation, an automatic clustering method based on kernel entropy analysis (KECA) is proposed. The feature vector with the greatest contribution to the Renyi entropy is selected as the projection direction to carry out the feature extraction. A classifier based on cosine similarity and the Kmeans algorithm is designed to get the label of each cell, and a method for determining the optimal number of clusters based on the angle of vectors is adopted. Experimental data of peripheral blood lymphocyte is processed, and the results indicate that the proposed method can realize automatically clustering with simple operation and the overall accuracy rate of clustering can reach 97%, which can improve the efficiency of cell analysis.

Abstract:When the geometric shadow method is applied to retrieve the significant wave height from marine radar images, the accuracy of shadow segmentation directly affects the performance of wave height estimation. Aiming at this problem, the Rune G′s shadowing algorithm (for the sake of convenience of description, the algorithm proposed by Rune′s method) is improved; an adaptive method that can accurately segment the shadow of the radar images is proposed based on the idea of differential edge detection. This novel method also inherits the advantage of the Rune′s algorithm, and could estimate the significant wave height without any external reference. Compared with the Rune′s method, the partial shadow segmented using the proposed algorithm is more reasonable. The image data acquired from Pingtan Test Base, Fujian were used to verify the effectiveness of the proposed algorithm. The results show that the improved estimation algorithm increases the accuracy of the wave height.

Abstract:Aiming at the problems of low accuracy and complex calculation in existing subpixel edge location algorithm, a subpixel edge location algorithm is proposed based on Gauss integral curved surface fitting. According to the characteristic of unilateral step edge, the edge normal section line Gauss integral model is constructed, on the basis of determining edge transition zone, the pixel information in fitting curved surface area is transformed into the active coordinates of edge curve, and the transformed pixel coordinates and gray information are fitted according to Gauss integral model, the accurate subpixel edge of the image is located. With the vision measurement system described in this paper, the experiment on the gauge block line edge was conducted, and the result was compared with that for traditional Gauss curved surface fitting subpixel edge location algorithm. The result proves that the subpixel edge location algorithm based on Gauss integral curved surface fitting has high location accuracy, the linear error of the first grade gauge block is within 1 μm, and the computing speed is doubled. When the algorithm is adopted to locate subpixel edge, the error caused by light source intensity can be effectively compensated through modifying the average of the Gauss integral model. Therefore, this algorithm can be applied to the high accurate measurement of mechanical parts, such as gears and etc.

Abstract:Aiming at the problems of strong speckle noise in sidescan sonar images and object segmentation difficulty, a segmentation algorithm based on spaceconstrained fast fuzzy Cmeans clustering (SCFFCM) and Markov random field (MRF) is proposed in this paper. Firstly, the strong speckle noise in sonar images is removed in nonsubsampled contourlet transform (NSCT) domain based on Bayesian maximum posteriori probability theory. Secondly, SCFFCM algorithm is proposed to accelerate the segmentation speed and give a good initial segmentation. Thirdly, the constrained field of MRF model is calculated from the initial segmentation, the combined weights of fuzzy clustering and Markov random field are adaptively updated according to the image gray fluctuations within the neighborhood; then the joint field of FCM fuzzy field and MRF constrained field is solved, and the segmentation result is obtained based on the maximum probability criterion. Finally, considering the noise points and ‘hole’ phenomenon in the segmentation result, a postprocessing method based on morphology is adopted to remove the isolated noise points and complete the ‘hole’ filling. Segmentation experiments on simulated and actual sidescan sonar images were conducted. Experiment results show that the proposed algorithm has stronger antinoise capability, higher segmentation precision and faster calculation speed compared with FCM and some other improved FCM algorithms.

Abstract:Corneal arcus is a white ring shape variation formed on the edge of the cornea, which is mainly caused by the abnormal lipid metabolism in human body. This corneal abnormality is significantly associated with the lipid disorder and atherosclerosis. Using image analysis method to detect corneal arcus can help people find the abnormal lipid metabolism in human body conveniently and in time. In the image acquired in natural eye open state, the corneal arcus is often occluded by the eyelid randomly, and disturbed by light spot and blood vessel. In order to improve the robustness of the algorithm and reduce the locating error caused by the random occlusion of the eyelid a corneal arcus segmentation method based on multiscale color replacement is proposed. Firstly, the image is quantized. Secondly, the image is processed with the defined color replacement strategy under different scale templates. Finally, the object segmentation is achieved. 1 968 images in our database were used to conduct experiment. Experiment results indicate that with the proposed method the segmentation accuracy for the 1 968 images reaches to 97.0%. The proposed method has high robustness and is not sensitive to noise.

Abstract:In complex environment, there are many kinds of defects on the surface of metal. Existing metal surface defect segmentation algorithm has the deficiencies of low efficiency and narrow application scope, to solve these problems, an adaptive segmentation algorithm for metal surface defects is proposed in this paper. In this algorithm, firstly, the graylevel images of the metal surfaces are transformed from eight directions. And then, according to the graylevel fluctuation of the multiple images, the threshold and step length in neighborhood gray level difference segmentation algorithm are adaptively changed, and corresponding image in each direction is segmented. Lastly, all the processed images are compressed to a single image with PCA algorithm. Experiment results indicate that compared with existing segmentation algorithms, the proposed algorithm not only can be applied to segment various kinds of metal surface defects, but also has high segmentation accuracy.

Abstract:Structure from motion (SFM) refers to a process in which the 3D structure is created by analyzing 2D image sequences, which is very important in many applications of computer vision. Feature tracking is one of the core components of largescale SFM. However, the robustness and time efficiency of the existing algorithms are needed to be improved. To address these issues, a fast and robust feature tracking (FRFT) algorithm is presented. Firstly, images moments are used to define a main direction for AGAST feature point, which can help to construct a rotation invariance descriptor. Secondly, in the space of the difference of Gaussian, the difference between the center point and its neighbor points is used to construct a descriptor for the OAGAST keypoint, which can avoid the influence of illumination and scale change on the feature matching. Thirdly, to improve the time efficiency of feature matching, the large feature set is clustered to some small ones, and KDTree is used to accelerate feature matching for improving the time efficiency of FRFT. Finally, the proposed method is evaluated with four ways, and compared with the stateoftheart methods. Experimental results show that the proposed FRFT method outperforms the stateofthe art ones on robustness and time efficiency.