Abstract:The microstructure parameters of steel can characterize a variety of physical properties. Steel strips and key pressure vessels in service need to be tested online for their microstructure parameters. This paper analyzes the on-line detection method of metal material microstructure. The research status and existing problems of electromagnetic methods ( hysteresis loop, magnetic Barkhausen noise, multi-frequency eddy current, harmonic analysis of tangential magnetic field strength) and ultrasonic methods (sound velocity method, attenuation coefficient method, nonlinear ultrasonic method and combination of multiple ultrasonic parameters ) in microstructure detection of metal materials are summarized. The relationship between the physical properties of metal materials and the microstructure parameters and the unclear relationship are discussed, and the development potential and direction of metal microstructure detection are prospected.

Abstract:Aiming at the problem that the traditional floating small target feature detection method is difficult to extract the target feature effectively, this paper analyzes the feature of small target on the sea surface, and studies the principle of fine composite multi-scale dispersion entropy (RCMDE). A small target detection method based on RCMDE-XGBoost is proposed. The signal was de-noised by using variational mode decomposition, the multi-scale features of the target were extracted by fine composite multi-scale dispersion entropy, the multi-dimensional feature matrix was constructed and input into XGBoost network for feature classification, and the small target detection on the sea surface was realized through model training. Using the IPIX radar measurement database, the detection rate of #54, #311, #320 HV polarization mode reaches 93. 33%, 92. 38%, 95% respectively, which is 12% higher than the graph connected density detection method on average, proving the effectiveness of RCMDE-XGBoost detection method.

Abstract:A lightweight traffic sign detection network incorporating foreground attention, YOLOT, is proposed to address the problem that object detection algorithm models are prone to error and miss detection on traffic sign detection. Firstly, the introduction of the SiLU activation function to improve the accuracy of model detection; secondly, a lightweight backbone network based on the ghost module is designed to effectively extract object features; thirdly, introduction of foreground attention perception module to suppress background noise; fourthly, we improve the path aggregation network by adding a residual structure to the feature fusion process; finally, we use VariFocalLoss and GIoU to calculate the classification loss of objects and the similarity between objects. Extensive experiments are conducted on several datasets, and the results show that the accuracy of the method in this paper is better than the current state-of-the-art methods. Ablation experiments are conducted on the CCTSDB dataset, and the final accuracy reaches 98. 50%, with an accuracy improvement of 1. 32% compared to the baseline model, while the model is only 4. 7 MB, and the real-time detection frame rate reaches 44 frames per second.

Abstract:A contour-based automatic perspective correction algorithm for circular meters is proposed for the problem of geometric distortion of the collected images in the process of using machine vision to recognize the readings of circular meters. The algorithm is divided into two parts: correction matrix estimation and image correction. First, the correction matrix is calculated by the outline of the circular meter, and then the distorted image is corrected by bilinear interpolation, the algorithm only needs to obtain the coordinate set of the outline of the circular meter to correct the distorted image. The simulation experiment results show that the correction algorithm based on contour can improve the correction accuracy by 2% ~ 6% compared with the traditional four-point projection transformation algorithm, and the actual instrument calibration experiment also shows that the contour calibration algorithm has higher accuracy and stability. It has practical application value for automatic reading of circular meters.

Abstract:Aiming at the problem of insufficient precision and poor generalization in the traditional way of wind turbine surface defect detection, an improved YOLOv5s wind turbine surface defect detection model is proposed. In terms of network structure, an improved MobileNetv3 network is introduced into the backbone feature extraction network to coordinate and balance the lightweight and accuracy relationship of the model. Secondly, the BiFPN fusion method is adopted to enhance the multi-scale adaptability of the neural network and improve the fusion speed and efficiency. Finally, for the lightweight adaptive adjustment of feature weights, the ECAnet channel attention mechanism is used to further improve the feature extraction ability of the neural network. In terms of loss function, the loss function of bounding box regression is modified to αIoU Loss, which further improves the accuracy of bbox regression. The experimental results show that the improved algorithm based on YOLOv5s can quickly and accurately identify the defect targets on the surface of the wind turbine in complex environments, and can meet the practical application requirements of real-time target detection.

Abstract:In order to guide the automatic decimal alignment of the parallel robot in the loading process, a high precision binocular measurement method of tanker mouth pose is proposed. The sub-pixels of the upper and lower edges of the tanker mouth image are extracted by image preprocessing. The binocular projection elliptical cone model of the spatial circle is established, and the bottom surface of the elliptical cone is constructed in the world coordinate system. Through the mapping relationship between the image pixels and the elliptic cone bottom, the optimal coordinate system of the cone bottom is found to obtain the target pose. The position of the pixel corresponding to the real center is calculated by pose correction, and the three-dimensional coordinates of the center in the world coordinate system are obtained by binocular triangulation. Aiming at the ambiguity problem of spatial circle projection, an ambiguity elimination method based on the tanker mouth height is proposed. The results of the simulation and experiment show that the algorithm has high accuracy and can meet the requirements of large angle pose measurement. The maximum error of the measured pose is 1. 8°, and the maximum error of the real center pixel extraction is 0. 98 pixels. The method does not need any auxiliary measurement and constraint conditions, and effectively improves the automatic efficiency in the loading process.

Abstract:Accurate perception of water level changes is one of the key segments to achieve precision water affairs control and flood disaster, but harsh scenes such as low illumination, haze, rain and snow, freezing, lighting, and waves bring a great challenge to water level accurate detection. Aiming at the problem of accurate detection of water level in existing methods, this paper constructs a Unet model fused with transformer residual channel attention mechanism (called “TRCAM-Unet”), then, a water lever intelligent detection method in harsh environments based on TRCAM-Unet is proposed. The key technologies include that: Multi-level feature fusion is achieved by full scale connection structure. The relevance of regional feature is strengthened by transformer module. Strengthening the extraction ability of useful information and weakening the interference of useless information by residual channel attention module. The experiments and practices of water level semantic segmentation in harsh scenes shows that TRCAM-Unet achieved 98. 84% MIOU scores and 99. 42% MPA scores, the maximum error of water level detection outside 150 meters was not above 0. 08 m, mean water level deviation (MLD) had only 1. 609×10 -2 meters, it is much better than the mainstream semantic segmentation models such as Deeplab, PSPNet, Unet. This study has important application value for water level accurate detection in harsh scenes and flood disaster early warning.

Abstract:Deep convolutional neural networks are one of the main methods for underwater image enhancement, but their expensive memory consumption and computational requirements hinder their deployment in practical applications. To this end, a lightweight dense residual convolutional neural networks (DRCNN) is proposed for underwater image enhancement. DRCNN uses depthwise separable convolution to extract high-level features to reduce computational cost; promotes information interaction between different channels through dense connection and residual learning, but also improves model representation; and fuses the input degraded image with the intermediate feature map to preserve image global similarity while preventing model gradients from vanishing. The experimental results demonstrate that DRCNN can significantly improve the quality of underwater images. When compared to the existing algorithm, DRCNN parameters are reduced by 85%, PSNR and SSIM values are increased by 3% and 2% respectively, and test speed is improved by 3%. DRCNN achieves better performance with fewer parameters, which is advantageous for real-time applications on low-resource devices.

Abstract:Vehicle re-identification is one of the important applications in the field of intelligent transportation. Most of the existing vehicle Re-ID methods focus on pre-defined local area features or global appearance features. However, in the complex traffic environment, it is difficult for traditional methods to acquire pre-defined local regions, and it is difficult to capture valuable vehicle global feature information. Therefore, an end-to-end dual-branch network with multi-view fusion hybrid attention mechanism and global feature enhancement is proposed. The network aims to obtain more complete and diverse vehicle features by enhancing the feature representation ability and feature quality of vehicles. This paper uses the view parsing network to segment the four views of the vehicle image, and uses the view stitching method to alleviate the problem of information loss caused by inaccurate segmentation. To better highlight salient local regions in stitched views, this paper proposes a hybrid attention module consisting of a channel attention mechanism and a self-attention mechanism. Through this module, the correlation between the key local information and the local information is obtained from the stitched view respectively, so as to better highlight the detailed information of the vehicle part in the stitched view. Besides, this paper also proposes a global feature enhancement module to obtain the spatial and channel relationship of global features through pooling and convolution. This module not only extracts the semantically enhanced vehicle features, but also makes the vehicle features contain complete detailed information, and solve the influence of the acquired vehicle images by factors such as changes in viewing angle and lighting conditions. Extensive experiments in this paper on the Veri-776 and VehicleID datasets show that mAP, CMC@ 1, and CMC@ 5 reach 82. 41%, 98. 63%, and 99. 23%, respectively, which is better than the existing methods.

Abstract:A method using a V shaped chessboard was proposed to address the dynamic pose determination problem of 2D LiDARs mounted on uniaxial turntables in scene scanning systems. From the image of the laser stripes on the calibration planes captured by a camera, the line features of the stripes were extracted. Then the laser plane equation and two-line equations of laser stripes in the camera coordinate system were computed. Using scanning data from the LiDAR, the transformation between the LiDAR and the camera coordinate systems was acquired. By controlling the turntable rotate three times or more, the functions of the turntable axis, the LiDAR center, and the LiDAR coordinate system w. r. t the rotation angle of the turntable were estimated. In the pose determination experiment, the average error between the calculated LiDAR centers and the samples is less than 1. 2 mm, the average angle errors between the calculated LiDAR coordinate axes and the samples are less than 0. 7°. In the experiment of object measurement using a 2D LiDAR mounted on a uniaxial turntable, the object size error is less than 3 mm. The experimental results showed a good accuracy of the presented method.

Abstract:A visual navigation model incorporating split attention mechanism and next expected observation ( NEO) is proposed to address the problem that deep reinforcement learning visual navigation algorithm degrades navigation accuracy, real-time and reliability of image matching due to navigation scene changes. The features of current and target states are first extracted using the ResNest50 backbone network to reduce network redundancy. The shallow target feature information is captured intensively using a cross-stagepartial-connections CSP to enhance the learning ability of the model. Then an improved loss function is proposed to make the inference network closer to the true posterior so that the agent can make the best decision in the current environment and further improve the navigation accuracy of the model in different scenarios. The training and testing are conducted on AVD dataset and AI2-THOR scenes, and the experimental results show that the navigation accuracy of the algorithm in this paper is as high as 96. 8%, with an average SR improvement of about 3% and an average SPL improvement of about 6%, which meets the requirements of navigation accuracy and realtime matching.

Abstract:Unsupervised person re-identification was attracted much attention due to its good scalability in real surveillance scenarios. The existing unsupervised person re-identification methods mainly trained the network by obtaining rough global features through the basic backbone network, and rarely used local refinement branches and global feature sharing to form more discriminative feature descriptors. This paper proposed a feature extraction network based on local refinement multi-branch and global feature sharing. The network combined rough global features and fine features in local refinement branches to obtain diverse feature expressions of person. In addition, in order to improve the ability of the branch network to capture information of potential key areas, an attention block of channel refinement information fusion was placed before the branch operation to enhance the network’ s attention to person features and perform dedicated learning of refined features. The experimental results on Market-1501, DukeMTMC-reID and MSMT17 datasets verified the effectiveness of the proposed method. The average accuracy (mAP) was improved by 4. 4%, 3. 2% and 6. 4% respectively, and the average accuracy on Market-1501 dataset reached 83. 3%.

Abstract:With the continuing popularity of deep learning techniques, convolutional neural network (CNN) has become the main tool to solve the remote sensing image scene classification tasks. However, current research interests are highly focused on the topic of how to fuse multi-branch-based CNN and how to apply attention models. Despite that these approaches enhance the classification accuracy markedly; it leads to high computational complexity. In this paper, the above problems are addressed by means of introducing a modified loss function and designing a novel data augmentation strategy, which can significantly improve the classification performance of CNN without increasing the computational complexity. First, a stage-based focal loss function is presented to adaptively mining the hard sample during the training process. Second, a parallel training strategy is conducted to feed the original image samples and samples after Gridmask operation into the sharing CNN separately. Experimental results show that the proposed algorithm achieves 96. 72% and 93. 95% detection accuracy on two large-scale databases of AID and NWPU-RESISC45, respectively, and can significantly improve the performance of remote sensing image scene classification.

Abstract:Aiming at the complex phenomena such as variable background, bending and overlapping arrangement of hydraulic pipeline and the low accuracy of pipeline segmentation by existing image segmentation methods, a hydraulic pipeline segmentation method based on U-net network, combined with Mobilenetv3 network, squeeze-and-excitation networks module and self-calibration convolutional module is proposed. The method selected Mobilenetv3-large model as the backbone network, and the feature maps are processed with Lraspp network. In the decoding process, the squeeze-and-excitation networks module and self-correction module are integrated to improve the feature extraction ability. Finally, the combination of Dice function and BCE function is used as the loss function of the network, which effectively improves the convergence ability of the network. Experimental results show that the mean values of the proposed method in the intersection over union and pixel accuracy are 90. 8% and 95. 2%, respectively. The model size is 16. 9 M, and the reasoning time for each image is 20 ms, which can be applied to the scene requiring real-time deployment. It provides a basis for the accurate identification of hydraulic pipeline leakage.

Abstract:Aiming at the problem that the quality of the captured image may be poor in the inspection of transmission lines, and the problem that the detection accuracy of traditional methods is not high due to the line defect that the targets are small and densely distributed, a transmission line defect detection method based on super-resolution reconstruction and multi-scale feature fusion is proposed. First, the super-resolution network is used to reconstruct the inspection image, improve the clarity and enrich the feature information contained in the image. Then the improved YOLOX network is used to detect defects in the inspection image, and the convolution block attention mechanism is embedded in the backbone network to strengthen the positioning ability of the model for overlapping small targets. In order to further improve the detection ability of small targets, a shallow detection scale is added to YOLOX’s feature fusion network for feature fusion. Finally, by using CIOU to optimize the loss function of the bounding box, improve the convergence ability of the model and reduce the missed detection rate of the defect targets. The experimental results show that the proposed method can accurately detect the transmission line defects on the basis of improving the inspection image quality, with an accuracy of 93. 27%. Compared with classical models such as SSD, it has stronger extraction ability and robustness for small and dense defect targets.

Abstract:In order to overcome the problem that difficult to accurately estimate the contrast and width of the edge, and easy to be affected by noise, which resulting in the reduction of the edge extraction accuracy in the current edge detector. A multi-scale differential edge detection algorithm with invariant scale and contrast was designed. Firstly, a mathematical function was used to represent the edge and to calculate the position, width, contrast, offset and direction of the closed form. The noise was filtered out as a low contrast feature. Secondly, a precise scale normalization method is defined to make the features of different dimensions comparable and improve the accuracy of the classifier. Then, through the derivative of gradient amplitude squared and the Laplaian calculation of gradient amplitude squared, the influence of contrast parameters was eliminated, and the edge detector with constant scale and contrast was constructed to output the edge. Experimental results show that the proposed method presents higher edge extraction effect, and the edge is more clear and complete compared with the current edge detection technology.

Abstract:Aiming at the problem that the non-rotational symmetry of the three output signals of the 3×3 optical fiber coupler is easy to cause the error of phase demodulation, this paper puts forward the theory of the asymmetric optical fiber coupler phase demodulation system based on the two output signals of the 3×3 optical fiber coupler, and verifies the phase demodulation ability of the system for the single frequency periodic signal and the third harmonic signal through software simulation analysis. The phase demodulation performance of the optical fiber vibration sensing system is tested on the phase-sensitive optical time domain reflectometer ( Φ-OTDR). The experimental results show that the system can realize phase demodulation in the range of 200~ 4 000 Hz, the absolute error of vibration positioning is within ±10 m, and the spatial resolution of the system can reach 21 m. On the basis of reducing the hardware cost of the Φ-OTDR system, this method can provide a new solution and reference scheme for the problem of unsteady splitting ratio and phase difference of optical fiber coupler in Φ-OTDR system, and can be applied and popularized in the field of traffic safety monitoring

Abstract:High humidity in the welding rod can lead to obvious quality problems such as bubbles and porosity in the weld. In order to meet the non-destructive detection of welding rod humidity in the field, a flexible interdigital capacitive rod humidity detection method is proposed. Firstly, a finite element model was established based on the new flexible interdigital capacitive sensor to study the influence of the structural parameters of the interdigital electrode on the sensor performance. Then, the parameters of the structure of the interdigital capacitive welding rod humidity sensor were optimized using BP neural network and genetic algorithm, and the optimal design of the humidity sensor was obtained. Finally, the U-shaped elastic mechanism was designed to completely wrap the welding rod and complete the humidity testing and calibration experiments for J507 and J607 welding rods. The simulation and experimental results show that the sensor detection depth is about 1. 4 mm (less than the thickness of the electrode flux) when the structural parameters of the interdigital electrode are optimal, and the goodness of fit of the electrode humidity value to the capacitance value reaches 0. 98. This method has good detection capability for welding rods with humidity values in the range of 0. 117% to 2. 5%, which provides a new idea for realizing on-site, nondestructive detection technology of welding rod humidity and helps to avoid welding defects during the welding process.

Abstract:The accurate state of health (SOH) estimation of Lithium-ion battery is one of the core issues faced by battery management systems. Considering that it is difficult to directly measure the battery capacity in practice, and the capacity regeneration problem always cause SOH estimation errors, a SOH estimation method of Lithium-ion battery is proposed based on incremental energy analysis and bidirectional gate recurrent unit ( BiGRU)-Dropout. The incremental energy curve is used to analyze the battery’ s degeneration characteristic, and the maximum peak height is extracted as a new health factor of battery SOH. Through the BiGRU network built by flip layer and gate recurrent unit layer, the mapping relationship between health factor and SOH is obtained. At the same time, Dropout mechanism network layer is added to prevent overfitting, and a SOH estimation model is established to accurate estimate the battery SOH. The results indicate that the proposed method can estimate battery SOH quickly and accurately under different charging rates.

Abstract:For the difficulties of classical bi-stable stochastic resonance (CBSR) system in amplification and detection of weak signals, an underdamped exponential tri-stable stochastic resonance (UETSR) system in a Levy noise background is proposed. The UETSR system is constructed by combining the bi-stable potential and exponential potential function, and using the property that non-Gaussian noise can effectively improve the signal-to-noise ratio. Firstly, the steady-state probability density function of the system is derived. The mean signal-to-noise ratio improvement (MSNRI) is adopted as an index to measure the stochastic resonance performance. The quantum particle swarm algorithm is used on parameters optimization. The effect of each parameter of the system on the output variation pattern of the UETSR system with different parameters α and β of Levy noise is investigated. Finally, the UETSR, CBSR and classical tri-stable stochastic resonance system (CTSR) are applied to the bearing fault diagnosis, and the amplitudes at the inner and outer ring fault frequencies after the system output increased by 197. 58, 1. 153, 18. 81 and 238. 87, 26. 63, 39. 72, respectively, compared to the input signal. The spectral level ratios of the highest peak to the second highest peak were 5. 44, 4. 03, 3. 85 and 5. 10, 3. 79, 5. 05. The experimental results show that SR phenomena can be induced by different system parameters, and the UETSR system outperformed the CBSR system and the CTSR system. The above conclusions prove that the system has excellent performance and strong practical significance

Abstract:Aiming at the problem that the composite interpolation envelope empirical mode decomposition (CIEEMD) method is lack of self-adaptability in the selection of non-stationary coefficient threshold, an improved composite interpolation envelope empirical mode decomposition (ICIEEMD) method is proposed. Firstly, the fractal box dimension is calculated from the vibration signal covered by grids with side length of ε, and the non-stationary threshold is adaptively selected. After decomposition, some intrinsic mode functions (IMF) are obtained. Secondly, combining with the correlation coefficient, the kurtosises of time domain signal and of envelope spectrum to establish the composite index of correlation coefficient and TE kurtosises (C-indexTE), then the effective IMF components were selected and reconstructed into a new signal. The energy spectrum of the reconstructed signal is obtained by using Teager energy operator, and the fault feature extraction of rolling bearing is realized. Finally, based on the simulation signal and the experimental data set of rolling bearing, the experimental analysis is carried out. The proposed method can extract more clear fault feature frequencies than the CIEEMD and spectral kurtosis methods, which proves the effectiveness and feasibility of the proposed method.

Abstract:With the rapid development of current air combat weaponry, trajectory prediction for high-altitude, high-speed, large maneuver targets is occupying an increasingly important strategic position. In order to solve the current problem of insufficient target trajectory prediction, this paper proposes a model integrating wavelet decomposition (WD) and long short term memory (LSTM) network to predict the trajectory of maneuvering targets. First, the input trajectory time series is decomposed into one low frequency component (CD1) and three high frequency components ( CA1, CA2, CA3) by wavelet decomposition. Then, the component prediction is performed by taking advantage of the long short term memory network for time series processing. Finally, the component prediction results are reconstructed and compared with the original trajectories for verification, and the results show that the proposed model has high accuracy for trajectory prediction. In order to exclude the chance of experimental results, two sets of data are used for validation in this paper. The comparison experiments show that the proposed model has less prediction error compared with the other two models.

Abstract:As the key equipment of high-speed railway train ground communication, the accuracy of balise information transmission directly affects the safe and efficient running of trains. In order to study the influence of train speed and balise tilting, pitching and yawing angular deviation on its transmission performance, the working principle of balise transmission system is analyzed according to its working process, and a dynamic model of BTM antenna receiving up-link signal and a theoretical model of balise angular deviation are established. Calculate the magnetic flux passing through the BTM antenna and the induced electromotive force of the BTM receiving antenna, simulate the amplitude curve of the induced electromotive force of the BTM receiving coil under different train speeds and balise installation angular deviation, and calculate and analyze the changes of the balise contact length, BTM operating time, receiving bits and other performance indicators under the above conditions. The research shows that: The contact length of balise decreases with the increase of the train speed. For every 50 km/ h increase in speed, the magnetic field intensity in the contact zone of balise decreases by 0. 2 dB. The BTM operating time, the number of dynamic receiving bits, and the number of safe frames of dynamic receiving telegram are inversely proportional to the train speed. When the train speed is constant, the number of dynamic receiving bits of BTM decreases linearly with the increase of balise tilt angle and yaw angle. In the process of on-site construction and maintenance, we should focus on checking the transponder tilt deflection. Under the condition of meeting the bit error rate of BTM received message, the applicable speed limit of the current transponder transmission system is 420 km/ h.

Abstract:As an important component of the hydraulic system, check valve affects the work efficiency of the hydraulic system and the safe and stable operation of the equipment when internal leakage occurs. Aiming at the leakage prediction in check valve, a nondestructive leakage prediction method in check valve based on multi-source signal and cuckoo search support vector regression ( CSSVR) is proposed. Firstly, a leakage detection experimental platform in the check valve is established, and the vibration signal and acoustic emission signal leaking in the check valve under the different working conditions and the different fault characteristics are obtained on the platform. Secondly, wavelet packet energy analysis method is used to extract the root mean square (RMS) of optimal frequency band reconstruction signals which are as the predictors with the inlet pressure to establish a multi-source signal prediction model based in CS-SVR. Finally, with compared and analyzed the models, the results show that the cuckoo search (CS) has a greater advantage over the grid search (GS) and particle swarm optimization (PSO) for SVR model parameters. Prediction methods based on multi-source signal inputs are more accurate than single-source input prediction methods, and the proposed method can realize the prediction of the internal leakage rate of the check valve with different pressure, different fault categories and different fault degrees. The proposed method average relative error is 8. 97% and has high robustness, which lays a foundation for the development of non-destructive testing application technology for valve leakage.

Abstract:Using undisturbed soil samples with different soil textures, the mass volume moisture content is measured and the volume moisture content of capacitive frequency domain reflectometry (FDR) soil moisture sensor is compared, and the calibration equation of the sensor is obtained. The whole process of comparative measurement from water content saturation to almost no free water was repeated for several times for each sample, and the dynamic measurement repeatability on different soil moisture was obtained. At the same time, in each comparative measurement of soil samples, rotate the soil moisture sensor to obtain the repeatability of static measurement in different directions on the same soil moisture value. Based on the measured data, the repeatability difference between the dynamic error and the static error is analyzed. The experimental research shows that the undisturbed soil calibration of FDR soil moisture sensor has good full-scale repeatability under the condition of little difference in ambient temperature. The maximum undisturbed soil sample dynamic error of volume moisture content is 3. 57% ( cm 3·cm -3 ), the minimum is 2. 29% ( cm 3·cm -3 ), and the average is 2. 81% (cm 3·cm -3 ); The maximum error of directional static volume moisture content is 0. 22% (cm 3·cm -3 ), the minimum is 0. 01% (cm 3· cm -3 ), and the average is 0. 02% (cm 3·cm -3 ). The undisturbed soil has good dynamic repeatability and can be used as the standard material for post station calibration. The error of directional static volume moisture content is small and can be ignored in the measurement.

Abstract:In view of the shortcomings of that it is difficult to calculate the initial position of switched reluctance motor in the inertial state, an initial position estimation strategy based on pulse injection method is proposed in this paper. The initial position of the motor can be obtained in both static state and inertia state when the strategy is adopted. Firstly, the selection principle of pulse current excitation time and injection frequency is analyzed, so that the switches and other devices can operate normally in a safe working environment. Then, the feasibility of experimental operation can be guaranteed. Secondly, the pulse injection method is used to implement the average inductance partition of the switched reluctance motor, and the position positioning of the motor at rest and with inertial speed is studied respectively. Finally, the experiment is carried out on a three-phase 6 / 4 switched reluctance motor. The experimental results show that the method can effectively measure the initial position of the motor in the inertial state or static state without adding additional hardware circuits. At the same time, the control algorithm is simple, easy to operate, and has strong universality and portability.