Multi-branch decision network identification method based on space object radar narrowband features

Abstract

The invention discloses a multi-branch decision network identification method based on space object radar narrowband features, and relates to the technical field of space target monitoring, and the method comprises the steps: obtaining narrowband radar data of a to-be-identified space object, and determining a structured data set of the to-be-identified space object based on the narrowband radar data; performing sliding window division on the structured data set to obtain a plurality of sliding data blocks; calculating operation data features in each sliding data block to obtain a motion feature set; calculating RCS sequence data features in each sliding data block to obtain an RCS statistical feature set; and splicing the motion feature set and the RCS statistical feature set to obtain a fusion feature, and determining an identification result of the to-be-identified space object based on the motion feature set and the RCS statistical feature set. According to the invention, through extracting the motion features of the narrowband radar data and the RCS statistical features and carrying out multi-branch decision fusion, the accuracy of space object identification is improved.

Description

Technical Field

[0001] This invention relates to the field of space target monitoring technology, and in particular to a multi-branch decision network identification method based on the narrowband radar characteristics of space objects. Background Technology

[0002] With the increasing demand for space target monitoring, radar technology has become a core means of space object detection and identification. Current space object identification tasks mainly rely on broadband radar signals. However, broadband radar systems generate massive amounts of echo data, leading to high complexity in signal processing links, high hardware resource consumption, and limited real-time performance, making it difficult to meet the application requirements of rapid screening and continuous monitoring of large-scale space targets. Meanwhile, although narrowband radar signals have the advantages of small data volume and ease of processing, traditional narrowband identification methods can only extract single-dimensional features from limited echo information, failing to fully explore target characteristics and resulting in low identification accuracy, severely restricting the effective application of narrowband radar in the field of space object monitoring.

[0003] To address the aforementioned technical problems, existing solutions have significant drawbacks. Firstly, feature extraction relies on a single dimension, depending solely on isolated features such as motion trajectories or radar cross-section fluctuations. This makes it difficult to effectively distinguish spatial objects with similar radar cross-section characteristics but different orbital characteristics, or those with similar orbits but significantly different structures, resulting in insufficient inter-class separability. Secondly, temporal correlation is poor. Most methods only utilize single-frame observation data for feature extraction, failing to fully explore the inherent correlations and temporal evolution patterns between consecutive frames, leading to dynamic information loss and affecting recognition stability. Thirdly, robustness is insufficient. There is a lack of effective mechanisms to remove outliers in radar cross-section sequences; abnormal measurements directly participate in statistical calculations, severely interfering with feature accuracy and reducing recognition precision.

[0004] In summary, existing space object recognition technologies face challenges in narrowband radar applications, including limited feature dimensions, poor temporal correlation, and insufficient robustness, resulting in performance limitations that fail to meet practical application requirements. Therefore, there is an urgent need for a space object recognition method capable of extracting multi-dimensional features, enhancing temporal correlation, improving robustness, and achieving efficient feature fusion under narrowband signal conditions, in order to overcome the performance bottleneck of narrowband radar in space target monitoring. Summary of the Invention

[0005] The technical problem this invention aims to solve is to address the shortcomings of existing technologies, specifically the limited recognition accuracy of space objects under narrowband radar conditions due to single feature dimensions, poor temporal correlation, and insufficient feature fusion. Specifically, this invention provides a multi-branch decision network recognition method based on narrowband radar features of space objects, as detailed below: 1) In a first aspect, the present invention provides a multi-branch decision network identification method based on the narrowband radar features of space objects, the specific technical solution of which is as follows: S1, acquire narrowband radar data of the spatial object to be identified, and determine the structured data set of the spatial object to be identified based on the narrowband radar data; S2, the structured data set is divided into multiple sliding data blocks by a sliding window; the running data features in each sliding data block are calculated to obtain a set of motion features; S3, calculate the RCS sequence data features within each sliding data block to obtain the RCS statistical feature set; S4, the motion feature set and the RCS statistical feature set are concatenated to obtain the fusion feature, and the recognition result of the spatial object to be identified is determined based on the motion feature set, the RCS statistical feature set and the fusion feature.

[0006] The beneficial effects of the multi-branch decision network identification method based on the narrowband radar features of space objects provided by this invention are as follows: By segmenting continuous sequence data using a sliding window mechanism, temporal correlation features are effectively extracted, overcoming the problem of temporal information loss caused by traditional methods relying solely on single-frame information, thus improving feature stability. Motion features and RCS statistical features are extracted and fused features are constructed, combining motion and structural characteristics to overcome the limitation of single feature dimensions in distinguishing morphologically similar targets, enhancing inter-class separability. Multi-branch decision fusion based on motion features, radar cross-section statistical features, and fused features fully utilizes complementary information from different feature types, solving the problem of poor adaptability in simple feature splicing, and improving recognition accuracy and system robustness.

[0007] Based on the above solution, the present invention can be further improved as follows.

[0008] Furthermore, the structured data set includes at least: time information, distance information, azimuth information, elevation information, and RCS sequence information.

[0009] Furthermore, the calculation of the running data characteristics within each sliding data block specifically includes: Based on the distance, azimuth, and pitch information in the structured dataset, the GEO coordinate system of the spatial object to be identified is transformed to the ECEF coordinate system to obtain the ECEF coordinate sequence. The ECEF coordinate sequence within the sliding data block is subjected to multiple polynomial fittings to obtain a polynomial model. The midpoint of all observation times within the sliding data block is used as the fitting time, and the position and velocity information of the spatial object to be identified at the fitting time are calculated using the polynomial model. The position information and velocity information are transformed from the ECEF coordinate system to the ECI coordinate system to obtain the position vector and velocity vector in the ECI coordinate system. Based on the position vector and the velocity vector, the running data features are calculated, and a set of motion features is constructed.

[0010] Furthermore, the step of performing multiple polynomial fittings on the ECEF coordinate sequence within the sliding data block to obtain a polynomial model specifically includes: Multiple polynomial fittings were performed on the ECEF coordinate sequence within the sliding data block to identify and remove abnormal observation points whose residuals exceeded a preset threshold, thus obtaining valid observation points. Using the effective observation points, polynomial fitting is performed multiple times, and polynomial fitting and identification and elimination operations are performed at least once in a loop. The polynomial model is obtained by refitting based on the obtained effective observation points.

[0011] Furthermore, the operational data characteristics include at least: semi-major axis, eccentricity, orbital inclination, right ascension of the ascending node, argument of perigee, true anomaly, altitude of apogee, and altitude of perigee.

[0012] Furthermore, the RCS sequence data features include at least the maximum, minimum, mean, median, coefficient of variation, range, kurtosis coefficient, skewness coefficient, variance, and standard deviation of the sequence within each sliding data block.

[0013] Furthermore, the step of determining the recognition result of the spatial object to be identified based on the motion feature set, the RCS statistical feature set, and the fusion feature specifically includes: The motion feature set is input into the first branch of the decision tree model to obtain the first classification score; The RCS statistical feature set is input into the second branch of the decision tree model to obtain the second classification score; The fused features are input into the third branch of the decision tree model to obtain the third classification score; A decision fusion classification score is determined based on the first classification score, the second classification score, and the third classification score. The decision fusion classification score is converted into a probability distribution, and the category with the highest probability value is output as the recognition result of the spatial object to be identified.

[0014] 2) In a second aspect, the present invention also provides a multi-branch decision network identification system based on the narrowband radar features of space objects, the specific technical solution of which is as follows: a data acquisition module, a motion feature module, an RCS feature module, and a fusion identification module; The data acquisition module is used to acquire narrowband radar data of the spatial object to be identified, and to determine the structured data set of the spatial object to be identified based on the narrowband radar data. The motion feature module is used to divide the structured data set into sliding window segments to obtain multiple sliding data blocks; and to calculate the running data features within each sliding data block to obtain a motion feature set. The RCS feature module is used to calculate the RCS sequence data features within each sliding data block to obtain a set of RCS statistical features; The fusion recognition module is used to concatenate the motion feature set and the RCS statistical feature set to obtain fusion features, and to determine the recognition result of the spatial object to be recognized based on the motion feature set, the RCS statistical feature set and the fusion features.

[0015] Based on the above solution, the present invention can be further improved as follows.

[0016] Furthermore, the structured data set includes at least: time information, distance information, azimuth information, elevation information, and RCS sequence information.

[0017] Furthermore, the calculation of the running data characteristics within each sliding data block specifically includes: Based on the distance, azimuth, and pitch information in the structured dataset, the GEO coordinate system of the spatial object to be identified is transformed to the ECEF coordinate system to obtain the ECEF coordinate sequence. The ECEF coordinate sequence within the sliding data block is subjected to multiple polynomial fittings to obtain a polynomial model. The midpoint of all observation times within the sliding data block is used as the fitting time, and the position and velocity information of the spatial object to be identified at the fitting time are calculated using the polynomial model. The position information and velocity information are transformed from the ECEF coordinate system to the ECI coordinate system to obtain the position vector and velocity vector in the ECI coordinate system. Based on the position vector and the velocity vector, the running data features are calculated, and a set of motion features is constructed.

[0018] Furthermore, the step of performing multiple polynomial fittings on the ECEF coordinate sequence within the sliding data block to obtain a polynomial model specifically includes: Multiple polynomial fittings were performed on the ECEF coordinate sequence within the sliding data block to identify and remove abnormal observation points whose residuals exceeded a preset threshold, thus obtaining valid observation points. Using the effective observation points, polynomial fitting is performed multiple times, and polynomial fitting and identification and elimination operations are performed at least once in a loop. The polynomial model is obtained by refitting based on the obtained effective observation points.

[0019] Furthermore, the operational data characteristics include at least: semi-major axis, eccentricity, orbital inclination, right ascension of the ascending node, argument of perigee, true anomaly, altitude of apogee, and altitude of perigee.

[0020] Furthermore, the RCS sequence data features include at least the maximum, minimum, mean, median, coefficient of variation, range, kurtosis coefficient, skewness coefficient, variance, and standard deviation of the sequence within each sliding data block.

[0021] Furthermore, the step of determining the recognition result of the spatial object to be identified based on the motion feature set, the RCS statistical feature set, and the fusion feature specifically includes: The motion feature set is input into the first branch of the decision tree model to obtain the first classification score; The RCS statistical feature set is input into the second branch of the decision tree model to obtain the second classification score; The fused features are input into the third branch of the decision tree model to obtain the third classification score; A decision fusion classification score is determined based on the first classification score, the second classification score, and the third classification score. The decision fusion classification score is converted into a probability distribution, and the category with the highest probability value is output as the recognition result of the spatial object to be identified.

[0022] 3) In a third aspect, the present invention also provides a computer device, the computer device including a processor coupled to a memory, the memory storing at least one computer program, the at least one computer program being loaded and executed by the processor to enable the computer device to implement any of the above methods.

[0023] 4) In a fourth aspect, the present invention also provides a computer-readable storage medium storing at least one computer program, which is loaded and executed by a processor to enable a computer to perform any of the above methods.

[0024] It should be noted that the beneficial effects of the technical solutions of the second to fourth aspects of the present invention and their corresponding possible implementations can be found in the above description of the technical effects of the first aspect and its corresponding possible implementations, and will not be repeated here. Attached Figure Description

[0025] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a flowchart illustrating the steps of a multi-branch decision network identification method based on the narrowband radar features of space objects according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the identification process of a multi-branch decision network identification method based on the narrowband radar features of space objects according to an embodiment of the present invention; Figure 3 This is a structural block diagram of a computer device according to an embodiment of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0027] like Figure 1 As shown in the figure, a multi-branch decision network identification method based on the narrowband radar features of space objects according to an embodiment of the present invention includes the following steps: S1, acquire narrowband radar data of the spatial object to be identified, and determine the structured data set of the spatial object to be identified based on the narrowband radar data; S2, divide the structured data set into sliding window segments to obtain multiple sliding data blocks; calculate the running data features within each sliding data block to obtain a set of motion features; S3, calculate the RCS sequence data features within each sliding data block to obtain the RCS statistical feature set; S4. The motion feature set and the RCS statistical feature set are concatenated to obtain the fused feature. The recognition result of the spatial object to be identified is determined based on the motion feature set, the RCS statistical feature set and the fused feature.

[0028] The beneficial effects of the multi-branch decision network identification method based on the narrowband radar features of space objects provided by this invention are as follows: By segmenting continuous sequence data using a sliding window mechanism, temporal correlation features are effectively extracted, overcoming the problem of temporal information loss caused by traditional methods relying solely on single-frame information, thus improving feature stability. Motion features and RCS statistical features are extracted and fused features are constructed, combining motion and structural characteristics to overcome the limitation of single feature dimensions in distinguishing morphologically similar targets, enhancing inter-class separability. Multi-branch decision fusion based on motion features, radar cross-section statistical features, and fused features fully utilizes complementary information from different feature types, solving the problem of poor adaptability in simple feature splicing, and improving recognition accuracy and system robustness.

[0029] It should be noted that, for ease of understanding, the technical terms used in this solution will be explained one by one, and will not be repeated hereafter: Narrowband radar data refers to echo data acquired under conditions of narrow radar signal bandwidth. Compared with wideband radar data, it has the advantages of small data volume and easy real-time processing. However, traditional processing methods are difficult to fully extract the effective features of the target. This solution aims to overcome the performance bottleneck of narrowband data in identification applications.

[0030] Structured dataset: refers to a dataset extracted from narrowband radar data and formed through format standardization. It includes at least time information, range information, azimuth information, elevation information, and RCS sequence information, providing input data in a unified format for subsequent feature extraction.

[0031] RCS stands for Radar Cross Section. In space object monitoring technology, RCS is a measure of the echo power generated by a target object under radar wave illumination, reflecting the target object's ability to reflect radar waves. The value of RCS is closely related to factors such as the physical size, shape, surface material of the target object, and radar observation angle, and is a key parameter characterizing the radar features of space objects. In this scheme, RCS is used as a core physical quantity. By extracting its statistical characteristics (such as maximum, minimum, mean, median, coefficient of variation, range, etc.), it is used to analyze the structural characteristics and attitude change patterns of space objects, forming a multi-dimensional identification basis together with motion characteristics.

[0032] Sliding window: This refers to a data segmentation mechanism that divides continuous sequence data into multiple data blocks according to a preset window length and a preset overlap step, enabling batch analysis and continuous processing of time-series data. In this scheme, the sliding window size is set to 500 frames, and the sliding step size is set to 100 frames, dividing the sequence data in the structured dataset of the spatial objects to be identified into multiple sliding data blocks of 500 frames each. The sliding window size of 500 frames and the sliding step size of 100 frames can be adjusted according to actual conditions.

[0033] Operational data characteristics: These refer to the orbital parameter characteristics extracted by analyzing the motion state of a space object. They include at least the semi-major axis, eccentricity, orbital inclination, right ascension of the ascending node, argument of perigee, true anomaly, altitude of apogee, and altitude of perigee. These characteristics are used to characterize the orbital properties and motion laws of a space object.

[0034] Motion feature set: refers to the feature set composed of operational data features, used to characterize the motion features of spatial objects.

[0035] RCS sequence data characteristics: These are the characteristics obtained by statistical analysis of the RCS sequence information within each sliding data block. They include at least the maximum, minimum, mean, median, coefficient of variation, range, kurtosis coefficient, skewness coefficient, variance, and standard deviation of the sequence within each sliding data block. They are used to characterize the structural characteristics and attitude change patterns of space objects.

[0036] RCS statistical feature set: refers to the feature set composed of RCS sequence data features, used to characterize the RCS features of spatial objects.

[0037] Fusion features: refers to the joint feature vector formed by splicing the set of motion features and the set of RCS statistical features. It contains both motion features and RCS statistical features, and is used to comprehensively characterize the motion characteristics and structural characteristics of spatial objects, so as to achieve multi-dimensional information complementarity.

[0038] GEO coordinate system: refers to the geodetic coordinate system, which uses longitude, latitude and altitude to represent spatial position. It is the coordinate representation of the initial observation data of radar stations and space objects, and needs to be converted to the geocentric-fixed coordinate system (ECEF coordinate system) for subsequent calculation and processing.

[0039] ECEF coordinate system: refers to the geocentric and geofixed coordinate system, with the Earth's center of mass as the origin. The X-axis points in the direction of the intersection of the prime meridian and the equator, the Y-axis points in the direction of the intersection of the 90° east longitude meridian and the equator, and the Z-axis points in the direction of the North Pole along the Earth's rotation axis.

[0040] ECEF coordinate sequence: refers to the X, Y, and Z coordinate components of the ECEF coordinate system, used to describe the position information of a spatial object in the ECEF coordinate system.

[0041] Polynomial fitting: This refers to using multiple polynomial models to fit curves to a coordinate sequence. Through iterative fitting and outlier removal operations, outlier observation points with residuals exceeding a preset threshold are identified and removed, ultimately yielding valid observation points and a fitted model to eliminate outlier interference in the observation data. In this scheme, fifth-order polynomial fitting and three iterations are used to perform polynomial fitting and identification / removal operations, ensuring that outliers in the observation data are completely removed.

[0042] ECI coordinate system: refers to the geocentric inertial coordinate system, with the Earth's center of mass as the origin. The coordinate axes remain fixed relative to the background stars. It is used to describe the true motion of objects in inertial space and avoids the influence of the Earth's rotation.

[0043] Position vector and velocity vector: These are vector quantities that represent the position and motion state of an object. The position vector consists of three coordinate components, and the velocity vector consists of three velocity components. They are the basic input quantities for orbital mechanics calculations and the mathematical basis for calculating the characteristics of operational data.

[0044] Semi-major axis, eccentricity, orbital inclination, right ascension of the ascending node, argument of perigee, true anomaly, altitude of apogee, and altitude of perigee: These refer to the six classical orbital elements and related altitude parameters that describe the orbital characteristics of a space object. The semi-major axis represents the size of the orbit, the eccentricity represents the shape of the orbit, the orbital inclination represents the degree of inclination of the orbital plane relative to the equatorial plane, the right ascension of the ascending node represents the position of the ascending node on the equatorial plane, the argument of perigee represents the position of the perigee in the orbital plane, the true anomaly represents the instantaneous position of the object on the orbit, and the altitude of apogee and altitude of perigee represent the altitudes of the farthest and closest points from Earth on the orbit, respectively. They are calculated using standard orbital mechanics formulas based on the position and velocity vectors in the ECI coordinate system.

[0045] Coefficient of variation, range, kurtosis coefficient, and skewness coefficient: These are parameters used in statistics to describe the distribution characteristics of data. The coefficient of variation reflects the proportion of data dispersion relative to the mean; the range reflects the difference between the maximum and minimum values; the kurtosis coefficient reflects the sharpness of the data distribution; and the skewness coefficient reflects the symmetry of the data distribution. In this scheme, they are used to characterize the statistical distribution pattern of the RCS sequence.

[0046] Decision tree model: refers to a multi-branch decision network, a machine learning classification model that makes decisions based on a tree structure, and achieves category determination by recursively partitioning the feature space. In this scheme, an adaptive augmented decision tree model is used as the basic classification model.

[0047] First / Second / Third Branches: These refer to the three parallel processing branches in the decision tree model. The first branch is dedicated to processing the motion feature set, the second branch is dedicated to processing the RCS statistical feature set, and the third branch is dedicated to processing the fused features. Each branch independently outputs a classification score, realizing differentiated feature processing and adaptive learning.

[0048] First / Second / Third Classification Score: This refers to the confidence score vector of each category output by the three parallel processing branches in the decision tree model. It represents the probability that the input feature belongs to each category. The higher the value, the stronger the confidence.

[0049] Decision fusion classification score: refers to the comprehensive classification score obtained by summing the first classification score, the second classification score, and the third classification score. It is used to comprehensively reflect the recognition results of the three branches and realize multi-perspective information fusion.

[0050] Probability distribution: refers to the normalized probability value distribution of the decision fusion classification score through the softmax function transformation. The sum of the probability values ​​of all categories is 1, and the category with the highest probability value is the final recognition result, realizing the quantitative output of classification decision.

[0051] In another embodiment of this solution, S1 is specifically implemented as follows: The system acquires narrowband radar data of the target space object from the radar receiver. This narrowband radar data is raw narrowband echo data, specifically the electromagnetic wave signal reflected back to the radar receiver after the radar transmitter emits a narrowband signal and it is reflected by the space object. The signal bandwidth is narrow, and the data volume is relatively small. The narrowband radar data is then analyzed to extract time information, range information, azimuth information, elevation information, and RCS sequence information. The radar transmitter is one of the core components of the radar station, and both have the same location information.

[0052] The time information is a sequence of Coordinated Universal Time (UTC) timestamps of the space objects observed by the radar. The range information is a sequence of straight-line distance measurements between the radar and the space objects. The azimuth information is a sequence of horizontal angle measurements of the space objects relative to the radar station. The elevation information is a sequence of vertical angle measurements of the space objects relative to the radar station. The RCS sequence is a sequence of quantified measurements of the space objects' ability to reflect radar waves.

[0053] The extracted time, distance, azimuth, elevation, and RCS sequence information are aligned to ensure that the information corresponding to the same observation time matches. Then, the format is standardized to unify the data types and numerical ranges, resulting in a structured dataset. The structured dataset is represented as data = {T, R, A, E, RCS}, where T represents time information, R represents distance information, A represents azimuth information, E represents elevation information, and RCS represents the RCS sequence information.

[0054] In another embodiment of this solution, S2 is specifically implemented as follows: Set the sliding window parameters, and set the sliding window size to 500 frames and the sliding step size to 100 frames according to the actual task requirements. Then, divide the continuous sequence data in the structured dataset into multiple independent sliding data blocks. Each sliding data block contains 500 frames of observation data, and there are 100 frames of overlapping data between two adjacent sliding data blocks to ensure that the temporal correlation is effectively utilized.

[0055] Based on the distance, azimuth, and elevation angle information in the structured dataset, the spatial object to be identified is transformed from the GEO coordinate system to the ECEF coordinate system, resulting in an ECEF coordinate sequence. The process of transforming from the GEO coordinate system to the ECEF coordinate system is as follows: Transform the radar station's GEO coordinates to the ECEF coordinate system: The radar station's GEO coordinates include longitude. (radians), latitude (radians) and height (meters), Earth's radius is The WGS84 ellipsoid model is used, with an oblateness of [missing information]. First eccentricity square Calculate the radius of curvature of the zonal loop. : ; Calculate the ECEF coordinates of the radar station ( , , ): ; ; .

[0056] Transform the distance, azimuth, and elevation coordinates of the spatial object to be identified into the ECEF coordinate system: Let the distance be R, the azimuth be A, and the elevation be E. For a given distance R, azimuth A, and elevation E, calculate the ENU coordinate vector: ; Construct the rotation matrix H from the ENU coordinate vector to the ECEF coordinate system: ; The ECEF coordinates (X, Y, Z) of the spatial object to be identified are: ; The ECEF coordinate system coordinate sequence is obtained. In this process, the ENU coordinate vector refers to the vector representing the relative position of a spatial object in the radar station's local northeast-sky coordinate system. ENU is an abbreviation for the east, north, and sky components. The ENU coordinate vector serves as an intermediate representation during coordinate transformation. First, the polar coordinates (range, azimuth, elevation) obtained from radar measurements are transformed to local rectangular coordinates (east, north, sky), facilitating subsequent transformation to the ECEF coordinate system via a rotation matrix, thus unifying the local coordinate system of the radar station with the global Earth-fixed coordinate system. During the transformation, the ENU coordinate vector reflects the instantaneous relative position of the spatial object with respect to the radar station, serving as a bridge connecting the original radar measurements with the Earth-fixed coordinate system.

[0057] A fifth-order polynomial fit is performed on the ECEF coordinate system coordinate sequence within each sliding data block, and a model is used for each coordinate component. A preliminary fitting is performed, where P represents the X, Y, or Z coordinate components, and t is the time variable. , , , , as well as For the polynomial coefficients, calculate the residuals and standard deviations between each observation point and the fitted curve. Identify and remove outlier observations (i.e., outliers) whose residuals exceed a preset threshold. The preset threshold is set to three times the standard deviation (e.g., 3σ, where σ is the standard deviation). Refit the curve using the remaining valid observations. Repeat the polynomial fitting and identification / removal process three times to ensure that outliers are completely removed, resulting in the polynomial model. .

[0058] The midpoint t of all observation times within the sliding data block mid As the fitting time t fit The polynomial model obtained by fitting is used to calculate the result at the fitting time t. fit Location information: The result at fitting time t is obtained fit ECEF coordinate system position (X ecef ,Y ecef Z ecef ). Calculate at fitting time t fit The velocity information is given by the first derivative of the polynomial: The result at fitting time t is obtained fit ECEF coordinate system velocity (Vx) ecef Vy ecef ,Vz ecef ).

[0059] Based on Greenwich Mean Time (GAST), the calculated position information (X) ecef ,Y ecef Z ecef ) and speed information (Vx) ecef Vy ecef ,Vz ecef Transform from the ECEF coordinate system to the ECI coordinate system to obtain the position and velocity vectors in the ECI coordinate system: convert the Coordinated Universal Time (UTC) time t fit Transform to Universal Time UT1: UT1 = UTC + ΔUT1, so that t in UTC time... fitConvert to a more uniform world time UT1, where ΔUT1 is the difference between UT1 and UTC provided by the International Earth Rotation and Reference System Service (IERS). Calculate the Greenwich Mean Time (GMT) apparent sidereal time angle θ. GAST (In radians), calculations are performed using standard models or software libraries recognized by the International Astronomical Union (IAU) (such as SOFA or ERFA), taking into account effects such as precession and nutation, to finally obtain t. fit GAST angle value θ at time t GAST .

[0060] The formula for position vector transformation is: r eci =R z (θ GAST )×r ecef , where r eci r is the position vector in the ECI coordinate system. ecef R is the position vector in the ECEF coordinate system. z (θ GAST Let be the right-handed rotation matrix about the Z-axis, defined as: .

[0061] The velocity vector conversion formula is: v eci =R z (θ GAST )×(v ecef +w e ×r ecef ), where v eci The velocity vector in the ECI coordinate system, v ecef Let w be the velocity vector in the ECEF coordinate system. e This is the Earth's rotational angular velocity vector, with a value of approximately 7.292115 × 10⁻⁶. -5 rad / s, expressed as w in the ECEF coordinate system e =[0, 0,w e ] T ; × represents the vector cross product operation, w e ×r ecef The term refers to the entrapment speed caused by the Earth's rotation.

[0062] The position vector r in the ECI coordinate system is obtained through the above transformation. eci =[X eci ,Y eci Z eci and velocity vector v eci =[Vx eci Vy eci ,Vz eciBased on the obtained position and velocity vectors in the ECI coordinate system, the operational data characteristics are calculated using standard orbital mechanics formulas, as follows: Angular momentum vector: h=r eci ×v eci ; Orbital plane normal vector (pointing to the ascending node): n=[0,0,1]×h, assuming the Z-axis is the Earth's rotation axis / reference axis; Eccentricity vector: e vec =(v eci ×h) / )-(r eci / ||r eci ||), where, GM is the Earth's gravitational constant. Orbital energy (specific mechanical energy): ɛ = (v eci v eci ) / 2- / ||v eci ||; Semi-major axis: a=- / (2ɛ), for elliptical orbits, a>0; Eccentricity: e = ||e vec ||; Orbital inclination angle: i = arccos(h) z / ||h||), where h z The Z component of the angular momentum vector h; Right ascension of ascending node: Ω = arctan2(n y ,n x ), where n y Let n be the Y-component of the orbital plane normal vector n. x Let X be the X component of the orbital plane normal vector n, note the quadrant, range [0, 360)°; Argument of perigee: =arccos((n e vec ) / (||n|| ||e vec ||)), if e vec <0, then ; True near point angle: =arccos((e vec r eci ) / (e ||r eci ||)), if r eci If v < 0, then ; apogee altitude: a L =(aR s ) / 1e3; Perimeter altitude: a S =a L -a×e / 1e3, where R s Where is the Earth's radius, and 1e3 is the conversion factor. A set of motion features is constructed. .

[0063] In another embodiment of this solution, S3 is specifically implemented as follows: For each sliding data block, extract the RCS sequence information contained therein, denoted as {RCS1, RCS2, ..., RCS}. n} where n is the number of data points within the sliding data block, n equals 500, and the sequence is arranged in chronological order of observation time. The maximum, minimum, mean, median, range, variance, standard deviation, coefficient of variation, skewness coefficient, and kurtosis coefficient are calculated for the extracted RCS sequence.

[0064] The formula for calculating the maximum value is: =max(RCS1,RCS2,...,RCS n ), representing the maximum value in the RCS sequence within the sliding data block; The formula for calculating the minimum value is: Min = min(RCS1, RCS2, ..., RCS) n ), representing the minimum value in the RCS sequence within the sliding data block; The formula for calculating the mean is: μ = (RCS1 + RCS2 + ... + RCS) n ) / n represents the arithmetic mean of all values ​​in the RCS sequence within the sliding data block; The median is calculated as follows: after sorting the RCS sequence by numerical value, take the middle value. When the number of data points n is odd, the median is... =RCS （n+1） / 2 The median when n is even =(RCS n / 2 +RCS （n / 2+1） ) / 2 represents the middle level of the RCS sequence within the sliding data block; The formula for calculating the range is: = -Min indicates the range of values ​​for the RCS sequence within the sliding data block; The formula for calculating variance is: σ² = Σ( -μ) 2 / n, where i ranges from 1 to n, the summation symbol Σ represents the summation operation over all data points, μ is the mean, n is the number of data points, and variance represents the degree of dispersion of the RCS sequence relative to the mean within the sliding data block; The formula for calculating standard deviation is: , representing the average dispersion of the RCS sequence within the sliding data block; The formula for calculating the coefficient of variation is: This indicates the relative dispersion of the RCS sequence within the sliding data block, eliminating the influence of dimensions; The formula for calculating the skewness coefficient is: =Σ( -μ) 3 / (n·σ 3 ), which represents the symmetry of the RCS sequence distribution pattern within the sliding data block. Positive values ​​indicate a right-skewed distribution, and negative values ​​indicate a left-skewed distribution; The formula for calculating kurtosis coefficient is: =Σ( -μ) 4 / (n·σ 4 The value indicates the sharpness of the RCS sequence distribution pattern within the sliding data block. The larger the value, the more concentrated the distribution and the sharper the peak.

[0065] The 10 RCS sequence data features obtained from the above calculations are represented as { ,Min,μ, , ,σ²,σ, , , The process of calculating the RCS sequence data features for a single sliding data block is completed. This process is repeated for all sliding data blocks to obtain the RCS sequence data features for each block, thus constructing a set of RCS statistical features.

[0066] It should be noted that the processes of obtaining the motion feature set and RCS statistical feature set in S2 and S3 are not sequential and can be obtained synchronously in parallel.

[0067] In another embodiment of this solution, S4 is specifically implemented as follows: After obtaining the motion feature set through S2, the motion feature set is represented as follows: , It contains motion features in 8 dimensions. The RCS statistical feature set is obtained through S3, and the RCS statistical feature set is represented as F. k F k ={ ,Min,μ, , ,σ²,σ, , , This includes 10 dimensions of statistical features. The set of motion features... With RCS statistical feature set F k A concatenation operation is performed, using horizontal vector concatenation to form a fused feature F. fusion F fusion =[ ,F k ], F fusion It is a 1×18-dimensional feature vector, which includes 8 dimensions of motion feature set and 10 dimensions of RCS statistical feature set. The fused features comprehensively represent the motion characteristics and structural characteristics of the spatial object to be identified, and realize multi-dimensional information complementarity.

[0068] Set of motion features The first branch of the decision tree model is input. The first branch uses the AdaBoost decision tree model as the object recognition model. The model parameters are set as follows: the number of learners is set to K, which is the same as the number of target categories (assuming there are K target categories). The number of target categories represents the total number of categories of spatial objects to be recognized. The number of splits is set to K-1 (the number of target categories minus 1), representing the number of branches split by each internal node. The 1×8 dimensional motion features are then processed. Input the first branch of the decision tree model for model training and classification prediction, and output the target recognition result, which is the first classification score S1: S1 is a 1×K dimensional vector, where each element represents the confidence score of the spatial object to be identified belonging to the corresponding category. The higher the score, the greater the probability that it belongs to that category. The first classification score S1 reflects the spatial object recognition result based on motion features.

[0069] The RCS statistical feature set F k Input the second branch of the decision tree model. The second branch uses the AdaBoost decision tree model as the target recognition model, and the model parameters are consistent with those of the first branch. The number of learners is set to K, and the number of splits is set to K-1. The 1×10 dimensional RCS statistical features F... k Input the second branch of the decision tree model for model training and classification prediction, and output the target recognition result, which is the second classification score S2: S2 is a 1×K dimensional vector, where each element represents the confidence score of the spatial object to be identified belonging to the corresponding category. The second classification score S2 reflects the spatial object recognition result based on structural features.

[0070] F fusion feature F fusionThe third branch of the decision tree model is input, and the AdaBoost decision tree model is also used as the target recognition model for the third branch. The model parameter settings are consistent with those of the first and second branches, with the number of learners set to K and the number of splits set to K-1. The 1×18 dimensional fusion feature F is used. fusion Input the third branch of the decision tree model for model training and classification prediction, and output the target recognition result, which is the third classification score S3: S3 is a 1×K dimensional vector, where each element represents the confidence score of the spatial object to be identified belonging to the corresponding category. The third classification score S3 reflects the spatial object recognition result based on the joint information of motion features and structural features.

[0071] The obtained first classification score S1, second classification score S2, and third classification score S3 are summed using an element-wise vector addition operation to obtain the decision fusion classification score S, where S = S1 + S2 + S3. S is a 1×K dimensional vector, and each element of S is the sum of the confidence scores for the corresponding categories of the three branches. The decision fusion classification score S comprehensively reflects the recognition results of the three branches, achieving multi-view information fusion. The decision fusion classification score S is then subjected to a softmax transformation to convert it into a probability distribution form. The category with the highest probability value after the softmax transformation is taken as the recognition result L of the spatial object to be identified. The recognition result L is the category label of the spatial object to be identified, thus completing the spatial object recognition process.

[0072] Furthermore, the structured dataset includes at least: time information, distance information, azimuth information, elevation information, and RCS sequence information.

[0073] Furthermore, the characteristics of the running data within each sliding data block are calculated, specifically including: Based on the distance, azimuth, and elevation information in the structured dataset, the GEO coordinate system of the spatial object to be identified is transformed to the ECEF coordinate system to obtain the ECEF coordinate sequence. The ECEF coordinate sequence within the sliding data block is subjected to multiple polynomial fittings to obtain a polynomial model. The midpoint of all observation times within the sliding data block is used as the fitting time, and the position and velocity information of the spatial object to be identified at the fitting time are calculated using a polynomial model. The position and velocity information are transformed from the ECEF coordinate system to the ECI coordinate system to obtain the position vector and velocity vector in the ECI coordinate system. Based on position and velocity vectors, the characteristics of the running data are calculated, and a set of motion features is constructed.

[0074] Furthermore, the ECEF coordinate sequence within the sliding data block is subjected to multiple polynomial fittings to obtain a polynomial model, specifically including: Multiple polynomial fittings were performed on the ECEF coordinate sequence within the sliding data block to identify and remove abnormal observation points whose residuals exceeded a preset threshold, thus obtaining valid observation points. Using valid observation points, perform polynomial fitting multiple times, and iterate at least once to perform polynomial fitting and identification and elimination operations. Based on the obtained valid observation points, a new fitting is performed to obtain a polynomial model.

[0075] Furthermore, the operational data characteristics include at least: semi-major axis, eccentricity, orbital inclination, right ascension of the ascending node, argument of perigee, true anomaly, altitude of apogee, and altitude of perigee.

[0076] Furthermore, the characteristics of RCS sequence data include at least the maximum, minimum, mean, median, coefficient of variation, range, kurtosis coefficient, skewness coefficient, variance, and standard deviation of the sequence within each sliding data block.

[0077] Furthermore, the recognition result of the spatial object to be identified is determined based on the motion feature set, the RCS statistical feature set, and the fused features, specifically including: Input the set of motion features into the first branch of the decision tree model to obtain the first classification score; Input the RCS statistical feature set into the second branch of the decision tree model to obtain the second classification score; The fused features are input into the third branch of the decision tree model to obtain the third classification score; The decision fusion classification score is determined based on the first classification score, the second classification score, and the third classification score. The decision fusion classification score is converted into a probability distribution, and the category with the highest probability value is output as the recognition result of the spatial object to be identified.

[0078] The beneficial effects are as follows: By combining motion parameter features and RCS temporal statistical features, an 18-dimensional feature vector is constructed using a slider narrowband feature fusion mechanism for spatial object recognition. A dynamic sliding window-style parameter fitting and regional statistical feature calculation are designed, and motion features and RCS statistical features are calculated by sliding over 500 frames to solve the problem of data fragmentation. By establishing a three-branch decision fusion network, two different features and the fused features are trained and processed separately, and the comprehensive decision recognition results are combined to improve recognition accuracy.

[0079] Example 1, Figure 2 To identify the flowchart, such as Figure 2 As shown, this constitutes the technical implementation path from raw narrowband radar data to spatial object recognition results.

[0080] Input raw narrowband radar data of space objects. Preprocess the raw narrowband echo data to construct a structured dataset.

[0081] Set the sliding window parameters. Divide the sliding window based on the structured dataset, and adjust the sliding parameters according to the actual task requirements. Extract the running sequence data features and RCS statistical features within the sliding data block to obtain the motion feature set and RCS statistical feature set, respectively.

[0082] A three-branch decision network recognition model is constructed. The motion feature set and the RCS statistical feature set are concatenated to form a fused feature. The motion feature set is input into the first branch of the decision tree model to obtain the first classification score, the RCS statistical feature set is input into the second branch to obtain the second classification score, and the fused feature is input into the third branch to obtain the third classification score.

[0083] The obtained first classification score, second classification score, and third classification score are summed to obtain the decision fusion classification score. The decision fusion classification score is then subjected to softmax transformation to convert it into a probability distribution form. The category with the highest probability value is output as the recognition result of the spatial object to be identified, thus completing the spatial object recognition process.

[0084] In the above embodiments, although the steps are numbered S1, S2, etc., they are only specific embodiments given by the present invention. Those skilled in the art can adjust the execution order of S1, S2, etc. according to the actual situation, and these situations are also within the protection scope of the present invention. It can be understood that in some embodiments, some or all of the above embodiments may be included.

[0085] The present invention also provides a multi-branch decision network identification system based on the narrowband radar features of space objects. The specific technical solution is as follows: a data acquisition module, a motion feature module, an RCS feature module, and a fusion identification module. The data acquisition module is used to acquire narrowband radar data of the spatial object to be identified, and to determine the structured data set of the spatial object to be identified based on the narrowband radar data. The motion feature module is used to divide the structured data set into sliding window segments to obtain multiple sliding data blocks; and to calculate the running data features within each sliding data block to obtain a motion feature set. The RCS feature module is used to calculate the RCS sequence data features within each sliding data block, and obtain the RCS statistical feature set. The fusion recognition module is used to concatenate the motion feature set and the RCS statistical feature set to obtain the fusion feature, and to determine the recognition result of the spatial object to be recognized based on the motion feature set, the RCS statistical feature set and the fusion feature.

[0086] Based on the above solution, the present invention can be further improved as follows.

[0087] Furthermore, the structured dataset includes at least: time information, distance information, azimuth information, elevation information, and RCS sequence information.

[0088] Furthermore, the characteristics of the running data within each sliding data block are calculated, specifically including: Based on the distance, azimuth, and elevation information in the structured dataset, the GEO coordinate system of the spatial object to be identified is transformed to the ECEF coordinate system to obtain the ECEF coordinate sequence. The ECEF coordinate sequence within the sliding data block is subjected to multiple polynomial fittings to obtain a polynomial model. The midpoint of all observation times within the sliding data block is used as the fitting time, and the position and velocity information of the spatial object to be identified at the fitting time are calculated using a polynomial model. The position and velocity information are transformed from the ECEF coordinate system to the ECI coordinate system to obtain the position vector and velocity vector in the ECI coordinate system. Based on position and velocity vectors, the characteristics of the running data are calculated, and a set of motion features is constructed.

[0089] Furthermore, the ECEF coordinate sequence within the sliding data block is subjected to multiple polynomial fittings to obtain a polynomial model, specifically including: Multiple polynomial fittings were performed on the ECEF coordinate sequence within the sliding data block to identify and remove abnormal observation points whose residuals exceeded a preset threshold, thus obtaining valid observation points. Using valid observation points, perform polynomial fitting multiple times, and iterate at least once to perform polynomial fitting and identification and elimination operations. Based on the obtained valid observation points, a new fitting is performed to obtain a polynomial model.

[0090] Furthermore, the operational data characteristics include at least: semi-major axis, eccentricity, orbital inclination, right ascension of the ascending node, argument of perigee, true anomaly, altitude of apogee, and altitude of perigee.

[0091] Furthermore, the characteristics of RCS sequence data include at least the maximum, minimum, mean, median, coefficient of variation, range, kurtosis coefficient, skewness coefficient, variance, and standard deviation of the sequence within each sliding data block.

[0092] Furthermore, the recognition result of the spatial object to be identified is determined based on the motion feature set, the RCS statistical feature set, and the fused features, specifically including: Input the set of motion features into the first branch of the decision tree model to obtain the first classification score; Input the RCS statistical feature set into the second branch of the decision tree model to obtain the second classification score; The fused features are input into the third branch of the decision tree model to obtain the third classification score; The decision fusion classification score is determined based on the first classification score, the second classification score, and the third classification score. The decision fusion classification score is converted into a probability distribution, and the category with the highest probability value is output as the recognition result of the spatial object to be identified.

[0093] It should be noted that the beneficial effects of the multi-branch decision network identification system based on narrowband radar features of space objects provided in the above embodiments are the same as those of the multi-branch decision network identification method based on narrowband radar features of space objects, and will not be repeated here. Furthermore, the system provided in the above embodiments is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the system can be divided into different functional modules according to the actual situation to complete all or part of the functions described above. In addition, the system and method embodiments provided in the above embodiments belong to the same concept, and their specific implementation process is detailed in the method embodiments, and will not be repeated here.

[0094] like Figure 3 As shown, an embodiment of the present invention provides a computer device 300, which includes a processor 320 coupled to a memory 310. The memory 310 stores at least one computer program 330, which is loaded and executed by the processor 320 to enable the computer device 300 to implement any of the above-described methods. Specifically: The computer device 300 can vary considerably due to differences in configuration or performance. It may include one or more processors 320 (Central Processing Units, CPUs) and one or more memories 310. The one or more memories 310 store at least one computer program 330, which is loaded and executed by the one or more processors 320 to enable the computer device 300 to implement the multi-branch decision network identification method based on the narrowband radar characteristics of space objects provided in the above embodiments. Of course, the computer device 300 may also have wired or wireless network interfaces, a keyboard, and input / output interfaces for input and output. The computer device 300 may also include other components for implementing device functions, which will not be elaborated upon here.

[0095] An embodiment of the present invention provides a computer-readable storage medium storing at least one computer program, which is loaded and executed by a processor to enable a computer to implement any of the above-described methods.

[0096] Alternatively, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, a floppy disk, and an optical data storage device, etc.

[0097] In an exemplary embodiment, a computer program product or computer program is also provided, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform any of the above-described multi-branch decision network identification methods based on narrowband radar features of space objects.

[0098] It should be noted that the terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and represent a limitation on a specific order or sequence. Where appropriate, the order of use for similar objects can be interchanged so that the embodiments of this application described herein can be implemented in an order other than that shown or described.

[0099] Those skilled in the art will recognize that this invention can be implemented as a system, method, or computer program product. Therefore, this disclosure can be specifically implemented in the following forms: it can be entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software, generally referred to herein as a "circuit," "module," or "system." Furthermore, in some embodiments, the invention can also be implemented as a computer program product contained in one or more computer-readable media, which includes computer-readable program code.

[0100] Any combination of one or more computer-readable media may be used. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device.

[0101] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A multi-branch decision network identification method based on the narrowband radar features of space objects, characterized in that, include: S1, acquire narrowband radar data of the spatial object to be identified, and determine the structured data set of the spatial object to be identified based on the narrowband radar data; S2, the structured data set is divided into multiple sliding data blocks by a sliding window; Calculate the running data characteristics within each sliding data block to obtain a set of motion features; S3, calculate the RCS sequence data features within each sliding data block to obtain the RCS statistical feature set; S4, the motion feature set and the RCS statistical feature set are concatenated to obtain the fusion feature, and the recognition result of the spatial object to be identified is determined based on the motion feature set, the RCS statistical feature set and the fusion feature.

2. The multi-branch decision network identification method based on the narrowband radar features of space objects according to claim 1, characterized in that, The structured data set includes at least: time information, distance information, azimuth information, elevation information, and RCS sequence information.

3. The multi-branch decision network identification method based on the narrowband radar features of space objects according to claim 1, characterized in that, The calculation of the running data characteristics within each sliding data block specifically includes: Based on the distance, azimuth, and pitch information in the structured dataset, the GEO coordinate system of the spatial object to be identified is transformed to the ECEF coordinate system to obtain the ECEF coordinate sequence. The ECEF coordinate sequence within the sliding data block is subjected to multiple polynomial fittings to obtain a polynomial model. The midpoint of all observation times within the sliding data block is used as the fitting time, and the position and velocity information of the spatial object to be identified at the fitting time are calculated using the polynomial model. The position information and velocity information are transformed from the ECEF coordinate system to the ECI coordinate system to obtain the position vector and velocity vector in the ECI coordinate system. Based on the position vector and the velocity vector, the running data features are calculated, and a set of motion features is constructed.

4. The multi-branch decision network identification method based on the narrowband radar features of space objects according to claim 3, characterized in that, The process of performing multiple polynomial fittings on the ECEF coordinate sequence within the sliding data block to obtain a polynomial model specifically includes: Multiple polynomial fittings were performed on the ECEF coordinate sequence within the sliding data block to identify and remove abnormal observation points whose residuals exceeded a preset threshold, thus obtaining valid observation points. Using the effective observation points, polynomial fitting is performed multiple times, and polynomial fitting and identification and elimination operations are performed at least once in a loop. The polynomial model is obtained by refitting based on the obtained effective observation points.

5. A multi-branch decision network identification method based on narrowband radar features of space objects according to any one of claims 1 or 3, characterized in that, The operational data features include at least: semi-major axis, eccentricity, orbital inclination, right ascension of the ascending node, argument of perigee, true anomaly, altitude of apogee, and altitude of perigee.

6. The multi-branch decision network identification method based on the narrowband radar features of space objects according to claim 1, characterized in that, The RCS sequence data features include at least the maximum, minimum, mean, median, coefficient of variation, range, kurtosis coefficient, skewness coefficient, variance, and standard deviation of the sequence within each sliding data block.

7. The multi-branch decision network identification method based on the narrowband radar features of space objects according to claim 1, characterized in that, The determination of the recognition result of the spatial object to be identified based on the motion feature set, the RCS statistical feature set, and the fused features specifically includes: The motion feature set is input into the first branch of the decision tree model to obtain the first classification score; The RCS statistical feature set is input into the second branch of the decision tree model to obtain the second classification score; The fused features are input into the third branch of the decision tree model to obtain the third classification score; A decision fusion classification score is determined based on the first classification score, the second classification score, and the third classification score. The decision fusion classification score is converted into a probability distribution, and the category with the highest probability value is output as the recognition result of the spatial object to be identified.

8. A multi-branch decision network identification system based on the narrowband radar characteristics of space objects, characterized in that, include: The module includes a data acquisition module, a motion feature module, an RCS feature module, and a fusion recognition module. The data acquisition module is used to acquire narrowband radar data of the spatial object to be identified, and to determine the structured data set of the spatial object to be identified based on the narrowband radar data. The motion feature module is used to divide the structured data set into sliding window segments to obtain multiple sliding data blocks; Calculate the running data characteristics within each sliding data block to obtain a set of motion features; The RCS feature module is used to calculate the RCS sequence data features within each sliding data block to obtain a set of RCS statistical features; The fusion recognition module is used to concatenate the motion feature set and the RCS statistical feature set to obtain fusion features, and to determine the recognition result of the spatial object to be recognized based on the motion feature set, the RCS statistical feature set and the fusion features.

9. A computer device, characterized in that, The computer device includes a processor coupled to a memory storing at least one computer program, which is loaded and executed by the processor to enable the computer device to implement a multi-branch decision network identification method based on the narrowband radar features of a space object as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one computer program, which is loaded and executed by a processor to enable the computer to implement a multi-branch decision network identification method based on the narrowband radar features of space objects as described in any one of claims 1 to 7.

Images