Search

Magnetic sensor array is usually designed for object detection and localization, where sensor coordinate is required to be consistent. Misalignment angles between coordinates should be calculated and calibrated. A new calibration method is proposed in order to calibrate the misalignment error between host frame and sensors. Host platform frame and perpendicular platform are used to obtain misalignment angles of sensor array. Experiment results show that it can estimate all angles of magnetic sensors and suppress misalignment error. Misalignment rms error is a reduction from 1109.7 to 27.6 nT, which means 2.48% indeed, thus improving the remote sensing accuracy of array which is unified to host frame coordinate.

Magnetic field measurement is an important analysis method in mining industry. Magnetic distortion of inertial navigation system (INS) is one severe error of portable geomagnetic field vector measurement instrument. A correlation algorithm is proposed for component compensation, in which magnetometer attitude information is provided by INS, and the geomagnetic field vector without magnetic anomaly is measured. Experiment result shows that distortion field vector of INS can reach hundreds of nT, which are suppressed to about 4.8 % after compensation, thus it is effective for component compensation. The magnetic anomaly measurement using magnetic intensity test (TMI) method and the proposed instrument are analyzed. In application, the proposed compensation way shows effectiveness comparing with several traditional compensation methods. In addition, the potential applications in mining industry are discussed.

Multi-Input Multi-Output Synthetic Aperture Radar (MIMO-SAR) emits orthogonal waveform, which reduces the probability of signal interception. Using multiple antennas to receive echoes, spatial filtering can be performed to avoid interference. MIMO-SAR waveform and structure characteristics determine that it has a certain anti-jamming ability with good prospect of application. For effective jamming MIMO-SAR, a theoretical model of rebound jamming against MIMO-SAR is established by taking the space-borne MIMO-SAR as object, the similarities and differences of rebound jamming against MIMO-SAR and conventional SAR are analyzed. Then, the effects of different signal systems on rebound jamming are discussed. Finally, the interference parameter selection for rebound jamming against the MIMO-SAR is analyzed. Simulation result verifies the validity and correctness of the proposed jamming method.

Narrowband interference (NBI) is a common interference pattern in direct-sequence spread-spectrum (DSSS) communications. NBI mitigation is a very important issue to ensure the effectiveness of DSSS communications. However, the existing NBI-mitigation algorithms for DSSS communications are confined to a high sampling rate. In order to solve this problem, compressive sensing (CS) is applied to NBI mitigation in DSSS communications, the impact of NBI on the reconstruction of the DSSS signal after compressed sampling is analyzed, and a newly emerged sparse approximation technique, block sparse Bayesian learning (BSBL), is utilized to estimate NBI in DSSS communications. An NBI-mitigation method in DSSS communications based on BSBL is proposed that uses the BSBL framework algorithm to reconstruct the NBI from the compressed signal and then cancels the interference in the time domain. After the NBI mitigation, the conventional CS algorithm is used to demodulate the DSSS signal in the compressed domain. In order to further improve the performance of the method, a filtering BSBL (FBSBL) framework is proposed using the block-sparsity feature of NBI and noise-like feature of the DSSS signal in the frequency domain. A DSSS communications NBI mitigation algorithm based on FBSBL_EM is designed. An equivalent measurement matrix is constructed use the constructed filtering matrix, which could reduce the noise and DSSS signal components in the procedure of compressed sampling. This leads to a highly accurate NBI reconstruction and further improves the NBI mitigation performance. The reported simulation results demonstrate that the proposed method is effective in cancelling the NBI in DSSS communications and significantly outperforms other conventional methods. A comprehensive analysis and comparison of the algorithm is presented and discussed. The results obtained in this study may be extended to other spread-spectrum communication systems, like the code-division multiple-access (CDMA) system.

Vehicle detection is an important method for understanding high-resolution remote sensing images. Deep convolutional neural network (DCNN)-based methods have improved many computer vision tasks and have achieved state-of-the-art results in many object detection datasets. Object detection of remote sensing images has been radically changed by the introduction of DCNN. Considering correlation between the scale distribution of objects and spatial resolution of remote sensing images, we propose an improved vehicle detection method based on a YOLOv3 model. A multi-scale clustering anchor box generation algorithm is proposed to obtain the anchor box parameters that match the resolution of each layer of the feature pyramid of model. This allows us to get more accurate anchor parameters. Focal loss is introduced into the default loss function to reduce the weight of negative samples, which were easily classified, that focus the model training process on samples that are difficult to classify. For the imbalance problem of positive and negative samples in the detection method based on the prior anchor box, focal loss is used to focus the model training process on samples that are difficult to classify. The experiment is performed on a dataset consisting of remote sensing images obtained from Worldview-3, and the results show that compared with the basic YOLOv3 algorithm, the average accuracy of vehicle detection is improved by 8.44%. The accuracy of vehicle detection of high-resolution remote sensing images is significantly improved while maintaining the speed of single-stage target detection. This approach is tested on an xView dataset consisting of remote sensing images obtained from Worldview-3. In addition, through using the proposed method, the average precision of vehicle detection increased by 8.44%. The experimental results show that the proposed method can be used for object detection in high-resolution remote sensing images effectively, and this method can significantly improve the performance of the model without sacrificing inference speed.

Owing to the ferromagnetism and stray fields of inertial navigation system, component compensation of magnetic field distortion is significant for geomagnetic field vector measurement system. However, traditional scalar compensation methods cannot be used for component compensation of magnetic field distortion, so some improvement works have been researched for component compensation in the study, the major of which is estimating compensation parameters using the Lagrange multiplier method. Experiment results show that the performance of the component compensation method is much better than traditional scalar compensation method. After compensation, the maximal measurement errors of north, vertical, east components and total intensity caused by magnetic field distortion are reduced to 10.4 nT (0.91% of the raw error), 32.5 nT (5.20%), 12.9 nT (1.27%), and 18.4 nT (2.02%), respectively. In addition, compared with previous component compensation method, the proposed method has three advantages: (i) simplified equipment, (ii) easier operation process, and (iii) better generality.

In this work, the authors improve the magnetic anomaly detector by using Karhunen-Loeve expansion, for expanding the detection distance of visually obscured ferromagnetic object when the target signal is contaminated by coloured noise. In the detector, the orthonormal basis functions are modified by the autocorrelation function of geomagnetic noise, and the anomaly is detected by matched filtering. Compared with the traditional detector, simulation results show that the modified detector has better performance with coloured Gaussian noise; especially, the improvement is more evident when the noise exponent is bigger. The real geomagnetic noise is proved to be coloured and obey approximate Gaussian distribution by the measurements in experiment. Tests with this noise indicate that the proposed detector achieves an incremental detection probability of 0.1 - 0.2 in the case of low signal-to-noise ratio, which makes it more attractive in practice.

Stationary human target localization is a key issue in through-the-wall sensing applications. When there are multiple stationary human targets, wrong data association may result in localization of false alarms. This paper proposes an approach of multiple stationary targets localization based on data association using UWB MIMO radar with one transmitting and two receiving antennas. First, the number and distances of stationary human targets are detected on each channel separately, then correlation coefficient matrix is introduced to evaluate the association probability of targets from different channels, finally the targets are located with no artifacts. Simulation and experimental results verified the superiority of the proposed approach.

For target location behind wall, the non-line-of-sight (NLOS) propagation caused by wall effect is the main factor that influences the localization performance. Based on the signal transmitting model, the NLOS is only determined by wall thickness and dielectric constant. In this paper, we use generalized Cramer-Rao lower bound (G-CRLB) as a basic tool to analyze target location performance. We deduced the G-CRLB of target localization behind wall. The Fisher information matrix (FIM) is divided into two parts: the observation matrix and prior information matrix, and the former corresponding to the measuring error, while the latter corresponding to wall parameter error. Simulation results evaluated the effect of wall parameters and measuring error on G-CRLB qualitatively.

Technological limitations in sensor manufacturing and unwanted magnetic fields will corrupt the measurements of three-axis magnetometers. An experiment with four different magnetic disturbance situations is designed, and the influence of hard-iron and soft-iron material is analyzed. The calibration method with magnetic disturbance parameters is proposed for calibration and magnetic disturbance compensation of three axis magnetometers. It is not necessary to compute pseudo-linear parameters, thus the integrated parameters are computed directly by solving nonlinear equations. To employ this method, a nonmagnetic rotation equipment, a CZM-3 proton magnetometer, a DM-050 three-axis magnetometer, two magnets and two steel tubes are used. Calibration performance is discussed in the four situations. Compared with several traditional calibration methods, experiment results show that the proposed method has better integrated compensation performance in all situations, and error is reduced by several orders of magnitude. After compensation, RMS error is reduced from 10797.962 nT to 15.309 nT when the big magnet and steel tube are deployed. It suggests an useful method for calibration and magnetic disturbance compensation of three-axis magnetometers.

Ferromagnetic interferential field from platforms is one of the most dominating error sources for magnetometer. For magnetic vector and gradient tensor measurement, what is cared about most is the effect of compensating magnetic field vector. In this paper, a magnetic compensation method is proposed, which uses host platform’s attitude from inertial sensor as auxiliary information and sets up a vectorial compensation model. By introducing three intermediate parameters, the issue of parameter estimation is linearized and solved with least squares method. Simulations show that errors of magnetic field vector and magnitude can both be reduced to several nT after compensation. Experiment has been conducted with a geomagnetic vector measurement system and results suggest that the method is an effective way for compensating both magnetic field magnitude and vectors.

The magnetometer error and distortion magnetic field of an inertial navigation system are the major two factors influencing the measurement accuracy of geomagnetic vector information measurement systems. The calibration and compensation methods are proposed in this letter. As a first step, a calibration model for a three-axis magnetometer is established, and a nonlinear least square algorithm is used to estimate thoroughly the error parameters. Then, a distortion compensation model based on relative attitude information is proposed, in which only four groups of attitude information are needed to estimate precisely the distortion parameters. Finally, the whole system is used for magnetic anomaly detection. Experimental results suggest that the magnetometer error can be accurately calibrated, and the magnetic field distortion can be suppressed significantly. After calibration and compensation, the intensity and magnetic anomaly vectors such as west vector, south vector, and vertical vector can be accurately measured by the geomagnetic vector measurement system. It demonstrates that the proposed method can effectively improve the accuracy of the geomagnetic vector measurement system.

Geomagnetic field is widely used for remote sensing, environment observation, geoscience and geology research. A portable geomagnetic field vector measurement system is designed, but misalignment greatly influences measurement accuracy. It is calibrated by establishing interim coordinate without measuring the projection of geomagnetic field and gravity. In addition, earth's magnetic dip, the accurate attitude of equipment and careful operation in collection process are not need. It is easy to implement the experiment, and misalignment angles between sensors are calculated. Magnetometer, inertial navigation system, host platform frame and perpendicular platform are used in experiment. After calibration, North, Vertical and East vector measurement error are reduced to 3.9%, 4.7% and 6.6% respectively. Inclination and declination error are reduced from 4.374° and 4.974° to 0.173° and 0.309° respectively. Thus, it is convenient and effective for misalignment calibration.

Magnetic interference is one key factor reducing the vector measurement accuracy of geomagnetic field measurement instrument, where scalar suppression algorithm is ineffective. A convenient component suppression method is proposed to avoid complex equipments and strict operations in previous component methods. The key factor is to overcome the problem that additive unknown parameters are added when attitude changes. System attitude change information is proposed for component suppression, where the attitude change is provided by inertial navigation system (INS). After suppression, magnetic interference of north, vertical and east direction are suppressed to 2.1%, 4.28% and 2.53% respectively. It provides a new and convenient way to improve the accuracy of geomagnetic field vector measurement.

Magnetic field distortion of INS is the major factor influencing the accuracy of geomagnetic field information measurement system. Simulation and experiment results show that traditional scalar compensation methods are disabled for component compensation. A component compensation method is proposed, in which parallelepiped frame and perpendicular platform are used with designed operation process. Comparing with traditional methods, the component compensation method is effective for distortion parameter estimation, and it shows better component compensation performance. Experimental test result demonstrates that distortion field components of INS are suppressed approximately two orders after compensation, and the North, Vertical and East component measurement error of the geomagnetic field are reduced to 2.3%, 3.3% and 4.5% of the former values respectively. Declination error is reduced from 7.074° to 0.331° (4.6% of the former value). This compensation method contributes to the accuracy improvement of geomagnetic field vector measurement system.

The algorithm, the model, and the measurement strategy are important for the calibration of three-axis magnetometers. A new calibration model with clear physical meaning is proposed, and scale factors, offsets, and nonorthogonal angles are directly illustrated. One limitation of traditional iteration calibration algorithms is the influence of the initial parameters. In this paper, the Levenberg–Marquardt algorithm is proposed to solve the troublesome procedure of initial parameter selection, so that it can improve the calibration performance of three-axis magnetometers. The validity of this method is proved by simulation, and the estimated parameters are found to be close to the prearranged parameters. The experimental system mainly consists of a three-axis fluxgate magnetometer (DM-050), two-dimensional nonmagnetic rotation equipment, and a proton magnetometer. The magnetic field intensity is obtained by the proton magnetometer. Experimental results show that the root-mean-square error is reduced from 84.177 to 4.076 nT. In addition, the influence of the initial parameters and measurement strategy is analyzed. Experimental results show that the Levenberg–Marquardt algorithm is not sensitive to the initial parameters and the 36 static measurements strategy is more reliable than the rotation measurement strategy. © 2014 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.

Magnetic distortions such as eddy-current field and low frequency magnetic field have not been considered in vector compensation methods. A new compensation method is proposed to suppress these magnetic distortions and improve compensation performance, in which the magnetic distortions related to measurement vectors and time are considered. The experimental system mainly consists of a three-axis fluxgate magnetometer (DM-050), an underwater vehicle and a proton magnetometer, in which the scalar value of magnetic field is obtained with the proton magnetometer and considered to be the true value. Comparing with traditional compensation methods, experimental results show that the magnetic distortions can be further reduced by two times. After compensation, error intensity and RMS error are reduced from 11684.013 nT and 7794.604 nT to 16.219 nT and 5.907 nT respectively. It suggests an effective way to improve the compensation performance of magnetic distortions.

Calibration strategy will influence the calibration performance of three-axis magnetometers. In this paper, three different calibration strategies (symmetrical calibration, orthogonal calibration and random calibration) are introduced and verified by simulation and experiments. The experimental system mainly consists of three-axis fluxgate magnetometer (MAG3300), 2D nonmagnetic rotation equipment and proton magnetometer. The scalar value of magnetic field was obtained with proton magnetometer and considered to be the true value. Calibration performance of three different calibration strategies were analyzed and compared. Experimental results show that after calibration the scalar RMS error has dropped 88.5%, 85.5% and 87.6% respectively. To check the generality of the estimated error parameters, another site was chosen to be a validation position, and the calibration performance of three calibration strategies were tested and compared. After calibration, the scalar RMS error has dropped 83.2%, 78.9% and 78.8% respectively. Thus the symmetrical calibration is considered to be the best calibration strategy among three calibration strategies.

The accuracy of three-axis magnetometers is influenced by different scale and bias of each axis and nonorthogonality between axes. One limitation of traditional iteration methods is that initial parameters influence the calibration, thus leading to the local optimal or wrong results. In this paper, a new method is proposed to calibrate three-axis magnetometers. To employ this method, a nonmagnetic rotation platform, a proton magnetometer, a DM-050 three-axis magnetometer and the differential evolution (DE) algorithm are used. The performance of this calibration method is analyzed with simulation and experiment. In simulation, the calibration results of DE, unscented Kalman filter (UKF), recursive least squares (RLS) and genetic algorithm (GA) are compared. RMS error using DE is least, which is reduced from 81.233 nT to 1.567 nT. Experimental results show that comparing with UKF, RLS and GA, the DE algorithm has not only the least calibration error but also the best robustness. After calibration, RMS error is reduced from 68.914 nT to 2.919 nT. In addition, the DE algorithm is not sensitive to initial parameters, which is an important advantage compared with traditional iteration algorithms. The proposed algorithm can avoid the troublesome procedure to select suitable initial parameters, thus it can improve the calibration performance of three-axis magnetometers.

Nonlinearity is a prominent limitation to the calibration performance of vector magnetometers. A new calibration model is proposed to suppress the nonlinearity of magnetometers and improve calibration performance, in which the nonlinearity coefficients of scale factors are considered. The experimental system mainly consists of a three-axis fluxgate magnetometer (MAG3300), a 2D nonmagnetic rotation equipment, and a proton magnetometer (CZM-3), in which the nonmagnetic rotation equipment is used to change the position of three-axis fluxgate magnetometer, and the scalar value of magnetic field is obtained with the proton magnetometer and considered to be the true value. The principle of this new calibration method is analyzed, and the calibration procedures are introduced. Experimental results show that after nonlinearity suppression, the root-mean-square error of calibration can be reduced by three times. It suggests an effective way to improve the calibration performance of three-axis magnetometers.

The accuracy of three-axis magnetometers is limited by different scales, bias of each axis and nonorthogonality between axes, which is usually lower than that of scalar magnetometers. In this paper, the nonlinear least square method is proposed to calibrate three-axis magnetometers. The validity of this method was proved by simulation, in which the estimated parameters of the error model are close to prearranged parameters. In experiment, a three-axis fluxgate magnetometer (DM-050), a two dimensional nonmagnetic rotation equipment and a proton magnetometer (GSM-19T) were used. The scalar value of magnetic field was obtained by proton magnetometer and considered to be the true value. The calibration performance of unscented Kalman filter (UKF), two-step algorithm and nonlinear least square were compared. Experimental results show that the error average and standard deviation of nonlinear least square are the least among the three methods. After calibration, the average of scalar error is reduced from −76.2 nT to −0.00093 nT and the standard deviation is reduced from 10.832 nT to 4.298 nT. The results suggest an effective way for the calibration of three-axis fluxgate magnetometers.

Three-axis magnetometer error and measurement noise influence the accuracy of magnetic measurements. Genetic algorithm (GA) is proposed to calibrate the magnetometer error, and wavelet is proposed for noise cancellation. The noise of a Mag3300 magnetometer and a DM magnetometer were tested within a horizontal barrel shield equipment. Five kinds of wavelet analysis and two kinds of wavelet package were used for noise cancellation, and the performance of different wavelets was compared. Noise of the Mag3300 magnetometer and DM magnetometer were suppressed from 29.6 and 2.3466 to 3.7 and 1.0789 nT, respectively. The scalar error of the Mag3300 magnetometer was tested using a two-dimensional nonmagnetic rotation equipment and a GSM-19T proton magnetometer. Scalar calibration performances of the unscented Kalman filter (UKF), the two-step algorithm, and GA were compared. Experimental results show that GA provides less error intensity (about 370 and 70 nT) than UKF and the two-step algorithm. In addition, the influence of wavelet on scalar calibration using UKF, the two-step algorithm, and GA was analyzed. Results show that wavelet improved the scalar calibration performance. Mean error of the Mag3300 magnetometer scalar was 795.5 nT. When combined with the wavelet package, the error was suppressed to −22.3, 3.3, and −0.4 nT using UKF, the two-step algorithm, and GA, respectively. The results suggest an effective way for magnetometer calibration using GA and wavelet. © 2016 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.

Magnetometer, misalignment error and distortion field can reduce the accuracy of gradiometers. So, it is important to calibrate and compensate gradiometers error. Scale factor, bias, nonorthogonality, misalignment and distortion field should be considered. A gradiometer is connected by an aluminium frame, which contains two fluxgate magnetometers. A nonmagnetic rotation equipment is used to change gradiometer attitude, and the compensation parameters are estimated. Experiment results show that, after calibration and compensation, error of each axis is reduced from 888.4 nT, 1292.6 nT and 168.9 nT to 15.3 nT, 22.1 nT and 9.9 nT, respectively. It shows that the proposed method can calibrate gradiometer and compensate distortion field. After calibration and compensation, the object remote sensing performance is improved.

Magnetoresistive sensors have plenty of applications in magnetic sensing areas, but magnetic and nonmagnetic 1/f noise severely degrades their low-frequency performances, especially detection sensitivity with hundreds of loss. In this paper, vertical motion flux modulation scheme with a piezoelectric silicon bridge is recommended for 1/f noise reduction, due to its effectiveness, simplicity and stability as well as moderate modulation efficiency. In our prototype sensor, the detection sensitivity was greatly improved to ~80 pT /√Hz at 1 Hz, which was near 300 times upgraded, when the detected magnetic field was transferred up to the resonance frequency of the piezoelectric silicon bridge.

Magnetic anomaly detection based on magnetic gradient tensor has become more and more important in civil and military applications. Compared with methods based on magnetic total field or components measurement, magnetic gradient tensor has some unique advantages. Usually, a magnetic gradient tensor measurement array is constituted by four three-axis magnetometers. The prominent problem of magnetic gradient tensor measurement array is the misalignment of sensors. In order to measure the magnetic gradient tensor accurately, it is quite essential to calibrate the measurement array. The calibration method, which is proposed in this paper, is divided into two steps. In the first step, each sensor of the measurement array should be calibrated, whose error is mainly caused by constant biases, scale factor deviations and nonorthogonality of sensor axes. The error of measurement array is mainly caused by the misalignment of sensors, so that triplets’ deviation in sensors array coordinates is calibrated in the second step. In order to verify the effectiveness of the proposed method, simulation was taken and the result shows that the proposed method improves the measurement accuracy of magnetic gradient tensor greatly.

Ideally, the output total value of three-axis magnetometer is stable at a certain place no matter what direction the three-axis magnetometer is placed. Actually, diversionary error of the three-axis magnetometer can not be avoided. In stable magnetic circumstance, tradional LMS adaptive algorithm is with good performance to calibrate the three-axis magnetometer. Experiment result indicates that diversionary error is decreased from 28 nT to 3.5 nT; but the performance is bad in disturbing magnetic field circumstance. Hence, the new method of combining adaptive filter with FIR digital filter is introduced. Then, advantages of the new method is proved by comparing with traditional LMS adaptive filter. Original diversionary error of three-axis magnetometer is 34.4722 nT in disturbing magnetic field circumstance, which is calibrated to 25.7352 nT using traditional LMS adaptive filter and 2.3502 nT by the new method of combining adaptive filter with FIR digital filter, respectively. Obviously, the method proposed in this paper has better calibration performance in disturbing magnetic circumstance.

Bias of magnetometers and target total value is obtained precisely via calibration equipment. Then, real time calibration weight matrix is obtained using LMS adaptive algorithm. It is proved through experiment that bias is obtained with good stability and accuracy compared with parameter estimation; after calibration, diversionary error is reduced from 33nT to 4nT. Furthermore, diversionary error is calibrated well using a proton magnetometer without rotating the magnetometer over one circle. Experiment results show that it not only reduces diversionary error fluctuation but also eliminates system error compared with other methods.

Step-size is the most important parameter in LMS adaptive algorithm, which directly influences the performance of adaptive filter. In this paper, filter parameter is theoretically calculated and subsection various step-size (SVSS) is introduced to improve stability of adaptive filter. Simulation results indicate that the new method of SVSS can solve well the contradiction between convergence rate and stability in noise cancelling. Furthermore, SVSS is introduced in calibration of three-axis magnetometer based on simulation and experiment. Obviously, experiment result shows the advantage of SVSS compared with fixed step-size and various step-size (VSS) method.

The magnetometer array is effective for magnetic object detection and localization. Calibration is important to improve the accuracy of the magnetometer array. A magnetic sensor array built with four three-axis DM-050 fluxgate magnetometers is designed, which is connected by a cross aluminum frame. In order to improve the accuracy of the magnetometer array, a calibration process is presented. The calibration process includes magnetometer calibration, coordinate transformation and misalignment calibration. The calibration system consists of a magnetic sensor array, a GSM-19T proton magnetometer, a two-dimensional nonmagnetic rotation platform, a 12 V-dc portable power device and two portable computers. After magnetometer calibration, the RMS error has been decreased from an original value of 125.559 nT to a final value of 1.711 nT (a factor of 74). After alignment, the RMS error of misalignment has been decreased from 1322.3 to 6.0 nT (a factor of 220). Then, the calibrated array deployed on the nonmagnetic rotation platform is used for ferromagnetic object localization. Experimental results show that the estimated errors of X, Y and Z axes are −0.049 m, 0.008 m and 0.025 m, respectively. Thus, the magnetometer array is effective for magnetic object detection and localization in three dimensions.

Fluxgate magnetometers are widely used for magnetic field measurement. However, their accuracy is influenced by temperature. In this paper, a new method was proposed to compensate the temperature drift of fluxgate magnetometers, in which a least-squares support vector machine (LSSVM) is utilized. The compensation performance was analyzed by simulation, which shows that the LSSVM has better performance and less training time than backpropagation and radical basis function neural networks. The temperature characteristics of a DM fluxgate magnetometer were measured with a temperature experiment box. Forty-five measured data under different magnetic fields and temperatures were obtained and divided into 36 training data and nine test data. The training data were used to obtain the parameters of the LSSVM model, and the compensation performance of the LSSVM model was verified by the test data. Experimental results show that the temperature drift of magnetometer is reduced from 109.3 to 3.3 nT after compensation, which suggests that this compensation method is effective for the accuracy improvement of fluxgate magnetometers.