A low-altitude unmanned aerial vehicle target identification method and device based on heterogeneous information fusion

Abstract

The application discloses a low-altitude unmanned aerial vehicle target identification method based on heterogeneous information fusion, and comprises the following steps: obtaining unmanned aerial vehicle heterogeneous monitoring information set facing a low-altitude environment; the unmanned aerial vehicle heterogeneous monitoring information set comprises radar tracking result information, video monitoring result information and ADS-B device receiving result information; the radar tracking result information comprises radar echo signals containing unmanned aerial vehicle targets and target motion speed information; the video monitoring result information comprises images containing unmanned aerial vehicle targets and target motion speed information; the ADS-B device receiving result information comprises receiving signals, target speed information and timestamp information; the unmanned aerial vehicle heterogeneous monitoring information set is preprocessed to obtain a preprocessing information set; the preprocessing information set is subjected to fusion identification processing to obtain identification result information of the low-altitude unmanned aerial vehicle target.

Description

Technical Field

[0001] This invention relates to the fields of anti-drone technology, industrial data processing, and artificial intelligence technology, specifically to a method and apparatus for low-altitude drone target identification based on heterogeneous information fusion. Background Technology

[0002] With the rapid development of the low-altitude economy, the popularity of civilian drones continues to increase, and they are widely used in fields such as aerial photography, logistics, surveying and mapping, and emergency rescue. However, at the same time, problems such as illegal flights, unauthorized flights, and air traffic disruptions are becoming increasingly prominent, posing a serious threat to low-altitude airspace safety, civil aviation flight safety, and security of important locations. Therefore, achieving accurate and efficient low-altitude drone target identification has become a core technical requirement in the field of low-altitude security and airspace management.

[0003] Currently, most existing low-altitude UAV target recognition technologies rely on a single monitoring data source, mainly divided into three single modes: radar monitoring, video surveillance, and ADS-B equipment monitoring. Each of these technologies has obvious limitations and is difficult to adapt to the recognition needs in complex low-altitude environments. Among these technologies, radar monitoring has strong anti-interference capabilities and can acquire dynamic information such as the target's speed and distance. It is well-suited to adverse weather conditions and complex lighting environments. However, for low-altitude, small, and slow-moving drones, radar echo signals are easily affected by terrain and buildings, resulting in significant signal noise. Relying solely on radar makes it difficult to accurately distinguish drones from birds and other low-altitude floating objects, leading to frequent misjudgments. Video surveillance technology can intuitively acquire images of the target's appearance and distinguish its shape. However, it is highly susceptible to environmental factors such as lighting, rain, snow, fog, and obstructions. Image clarity drops significantly at night or in adverse weather conditions, causing a sharp decrease in target detection accuracy. Furthermore, it is difficult to accurately acquire long-distance motion parameters of the target, resulting in insufficient recognition stability. ADS-B device monitoring technology can directly acquire standardized information such as the target's speed and timestamp, with high data transmission efficiency. However, this technology can only monitor compliant drones equipped with and normally activated ADS-B devices. It cannot effectively monitor unauthorized drones that are not equipped with the device, have their devices turned off, or have their signals blocked, severely limiting its applicability.

[0004] Furthermore, existing identification methods that attempt to fuse multi-source data generally suffer from inadequate data preprocessing and unreasonable heterogeneous information fusion logic. On the one hand, multi-source monitoring data suffers from inconsistent formats, spatiotemporal asynchrony, data gaps, noise interference, and numerous outliers. The lack of targeted cleaning and spatiotemporal registration processes leads to ineffective alignment of information from different data sources, resulting in information conflicts, redundancy, or missing data during fusion, significantly reducing identification accuracy. On the other hand, most existing fusion methods only perform simple data overlay, failing to address the characteristic differences among radar, video, and ADS-B heterogeneous information sources through refined feature extraction and optimization. This fails to fully leverage the complementary advantages of different data sources. Additionally, the lack of scientific matching and weighting mechanisms in target type identification and motion parameter fusion estimation results in low target type identification accuracy and large motion speed estimation errors, making it difficult to meet the actual control needs of real-time, accurate identification of low-altitude UAVs. In summary, existing technologies cannot simultaneously achieve identification accuracy, stability, and comprehensiveness in complex low-altitude environments, making it difficult to achieve full coverage and high-precision identification of compliant and non-compliant UAVs, thus hindering the further development of low-altitude security control technology. Summary of the Invention

[0005] This invention primarily addresses the challenge of balancing recognition accuracy, stability, and comprehensiveness in complex low-altitude environments, making it difficult to achieve full coverage and high-precision identification of both compliant and non-compliant drones. This invention discloses a low-altitude drone target recognition method and device based on heterogeneous information fusion.

[0006] In a first aspect, this invention discloses a method for low-altitude unmanned aerial vehicle (UAV) target recognition based on heterogeneous information fusion, comprising: S1, Obtain a set of heterogeneous monitoring information for UAVs in low-altitude environments; the set of heterogeneous monitoring information includes radar tracking results, video surveillance results, and ADS-B device reception results; the radar tracking results include radar echo signals containing UAV targets and target speed information; the video surveillance results include images containing UAV targets and target speed information; the ADS-B device reception results include received signals, target speed information, and timestamp information; S2, preprocess the heterogeneous monitoring information set of the UAV to obtain a preprocessed information set; S3, perform fusion recognition processing on the preprocessed information set to obtain the recognition result information of the low-altitude UAV target.

[0007] The preprocessing of the heterogeneous monitoring information set of the UAV to obtain a preprocessed information set includes: S21, perform data category cleaning processing on the heterogeneous monitoring information set of the UAV to obtain the first information set; S22, perform registration processing on the first information set to obtain a preprocessed information set.

[0008] The process of fusing and recognizing the preprocessed information set to obtain the recognition result information of the low-altitude UAV target includes: S31, Perform classification feature extraction processing on the preprocessed information set to obtain a classification feature set; S32, Based on the preset target standard feature set, perform type recognition processing on the classification feature set to obtain the target type recognition result; S33, based on the preset target standard feature set and target type identification results, the target motion speed information in the preprocessed information set is fused and estimated to obtain the target speed estimation result; S34. Based on the target type identification result and the target velocity estimation result, the identification result information of the low-altitude UAV target is constructed.

[0009] The step of performing classification feature extraction on the preprocessed information set to obtain a classification feature set includes: S311, Perform echo feature extraction processing on the radar echo signal containing the UAV target in the radar tracking result information in the preprocessed information set to obtain the radar feature vector; S312, perform image feature extraction processing on the images containing drone targets in the video surveillance result information of the preprocessed information set to obtain image feature vectors; S313, Perform signal feature extraction processing on the received signal in the ADS-B device received result information in the preprocessed information set to obtain a signal vector; S314, using the radar feature vector, image feature vector and signal vector, a classification feature set is constructed.

[0010] The process of extracting echo features from radar echo signals containing UAV targets in the preprocessed information set to obtain radar feature vectors includes: S3111, Empirical mode decomposition is performed on the radar echo signal containing the UAV target in the radar tracking result information of the preprocessed information set to obtain a IMF component signals and a margin signal. S3112, the a IMF component signals are filtered to obtain s high-frequency components and d low-frequency components; S3113, calculate the mean of each of the d low-frequency components to obtain d first mean components; S3114, perform wavelet threshold denoising on the s high-frequency components to obtain the denoised s high-frequency components; S3115, calculate the mean of each of the denoised s high-frequency components to obtain s second mean components; S3116, Perform mean calculation on the residual signal to obtain a third mean component; S3117, using the d first mean components, s second mean components and 1 third mean component, a radar feature vector is constructed.

[0011] The step of performing signal feature extraction processing on the received signal in the ADS-B device received result information in the preprocessed information set to obtain a signal vector includes: S3131, Perform reduced-order variational mode decomposition on the received signal in the ADS-B device received result information in the preprocessed information set to obtain the first transformation sequence; S3132, using a signal reconstruction model, the first transformed sequence is reconstructed to obtain the feature signal; the expression of the signal reconstruction model is: In the formula, For the preset first in the signal reconstruction model One reconstruction filter, , The characteristic signal is N, and N is the total number of reconstruction filters. This is the first transformation sequence; S3133, Discretely sample the feature signal to obtain a signal vector.

[0012] The process of performing type recognition processing on the classification feature set based on the preset target standard feature set to obtain the target type recognition result includes: S321, a type recognition model is constructed based on the preset target standard feature set and the classification feature set; the preset target standard feature set includes radar feature standard vector, image feature standard vector and signal standard vector for each target type, and each target type has a corresponding sequence number; S322, Solve the type recognition model to obtain the recognition type number; S323, Determine the target type corresponding to the identification type number, which is the target type identification result; The expression for the type recognition model is: Where I is the identification type index, and i is the index of the target type in the target standard feature set. , and These are the k-th elements of the radar feature standard vector, image feature standard vector, and signal standard vector of target type i in the target standard feature set, respectively. , and These are the k-th elements of the radar feature vector, image feature vector, and signal vector in the classification feature set, respectively, where K is the total number of elements in the radar feature vector.

[0013] A second aspect of this invention discloses a low-altitude unmanned aerial vehicle (UAV) target identification device based on heterogeneous information fusion, the device comprising: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the low-altitude UAV target recognition method based on heterogeneous information fusion.

[0014] In a third aspect of this invention, a computer-readable storage medium is disclosed, wherein the computer-readable storage medium stores computer instructions, which, when invoked by a computer, are used to execute the aforementioned low-altitude unmanned aerial vehicle target recognition method based on heterogeneous information fusion.

[0015] In a fourth aspect of this invention, an information data processing terminal is disclosed, which is used to implement the low-altitude unmanned aerial vehicle target recognition method based on heterogeneous information fusion.

[0016] The beneficial effects of this invention are as follows: This invention integrates three types of heterogeneous monitoring information—radar tracking, video surveillance, and ADS-B devices—breaking the limitations of monitoring from a single data source. By fully leveraging the complementary advantages of different data sources, it achieves full-coverage monitoring of low-altitude drone targets. This solves the problems of easy misjudgment in pure radar monitoring and the significant environmental interference in video surveillance. It also compensates for the deficiency of ADS-B devices in monitoring unauthorized drones, significantly improving the comprehensiveness and applicability of low-altitude drone identification. It can adapt to various complex low-altitude environments such as daytime, nighttime, rain, snow, fog, and haze, as well as the full-scene identification needs of compliant and non-compliant drones.

[0017] This invention addresses the characteristics of heterogeneous monitoring information by establishing a dedicated data preprocessing workflow. First, invalid data is removed by data category cleaning, missing values ​​are filled, and noise and outliers are eliminated to ensure the integrity and validity of the basic data. Then, relying on the timestamp information of the ADS-B device, accurate time registration of multi-source data is completed, solving the core problems of spatiotemporal asynchrony and information conflict of heterogeneous data. This allows information from different data sources to be accurately aligned, laying a solid data foundation for subsequent fusion and recognition, effectively avoiding recognition errors caused by data disorder, and improving the stability of the overall recognition process.

[0018] This invention employs tailored, refined feature extraction methods for three different types of monitoring information. It performs empirical mode decomposition and wavelet denoising optimization on radar echo signals, target detection and two-dimensional discrete cosine transform feature extraction on video images, and reduced-order variational mode decomposition and signal reconstruction on ADS-B received signals. This fully leverages the core feature information of each data source, constructing a unified and complementary set of classification features. This avoids feature redundancy issues caused by simple data aggregation, maximizes the feature advantages of multi-source information, and significantly improves the recognizability of target features. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the implementation of the method of the present invention. Detailed Implementation

[0020] To better understand the content of this invention, an embodiment is provided here.

[0021] Figure 1 This is a flowchart illustrating the implementation of the method of the present invention.

[0022] In a first aspect, this invention discloses a method for low-altitude unmanned aerial vehicle (UAV) target recognition based on heterogeneous information fusion, comprising: S1, Obtain a set of heterogeneous monitoring information for UAVs in low-altitude environments; the set of heterogeneous monitoring information includes radar tracking results, video surveillance results, and ADS-B device reception results; the radar tracking results include radar echo signals containing UAV targets and target speed information; the video surveillance results include images containing UAV targets and target speed information; the ADS-B device reception results include received signals, target speed information, and timestamp information; S2, preprocess the heterogeneous monitoring information set of the UAV to obtain a preprocessed information set; S3, perform fusion recognition processing on the preprocessed information set to obtain the recognition result information of the low-altitude UAV target.

[0023] Automatic Dependent Surveillance-Broadcast (ADS-B) information is critical flight information such as the aircraft's position, speed, and altitude that the aircraft automatically transmits to ground stations and other aircraft.

[0024] The radar tracking results in the heterogeneous monitoring information set of the UAV are obtained by tracking the UAV target with radar, and the target's speed information is obtained by tracking the UAV with radar. The video surveillance results are obtained by tracking the drone using optical tracking equipment. The target's velocity information is obtained by applying a least-squares polynomial fitting method to the acquired images. The process is as follows: using a collinear positioning method based on an imaging model, the position coordinates of the moving target in different image sequences are calculated. Then, the least-squares polynomial fitting method is applied to obtain a curve fitting polynomial for the target's position and velocity. By differentiating the position polynomial, the target velocity at any desired time can be obtained.

[0025] The ADS-B device receives the result information by receiving the ADS-B signal emitted by the UAV. The target speed information and timestamp information are obtained by parsing the ADS-B signal.

[0026] The preprocessing of the heterogeneous monitoring information set of the UAV to obtain a preprocessed information set includes: S21, perform data category cleaning processing on the heterogeneous monitoring information set of the UAV to obtain the first information set; S22, perform registration processing on the first information set to obtain a preprocessed information set.

[0027] The data category cleaning process first involves checking the data category and deleting data whose data type does not match the preset data type. The second step is to clean the data, including filling in missing values, smoothing noisy data, and smoothing or deleting outlier points. The registration process involves performing time registration on various types of information in the heterogeneous monitoring information set of the UAV based on the timestamp information in the result information received by the ADS-B device. Both the radar echo signal and the image containing the UAV target contain time information; The process of fusing and recognizing the preprocessed information set to obtain the recognition result information of the low-altitude UAV target includes: S31, Perform classification feature extraction processing on the preprocessed information set to obtain a classification feature set; S32, Based on the preset target standard feature set, perform type recognition processing on the classification feature set to obtain the target type recognition result; S33, based on the preset target standard feature set and target type identification results, the target motion speed information in the preprocessed information set is fused and estimated to obtain the target speed estimation result; S34. Based on the target type identification result and the target velocity estimation result, the identification result information of the low-altitude UAV target is constructed.

[0028] The step of performing classification feature extraction on the preprocessed information set to obtain a classification feature set includes: S311, Perform echo feature extraction processing on the radar echo signal containing the UAV target in the radar tracking result information in the preprocessed information set to obtain the radar feature vector; S312, perform image feature extraction processing on the images containing drone targets in the video surveillance result information of the preprocessed information set to obtain image feature vectors; S313, Perform signal feature extraction processing on the received signal in the ADS-B device received result information in the preprocessed information set to obtain a signal vector; S314, using the radar feature vector, image feature vector and signal vector, a classification feature set is constructed.

[0029] The process of extracting echo features from radar echo signals containing UAV targets in the preprocessed information set to obtain radar feature vectors includes: S3111, Empirical mode decomposition is performed on the radar echo signal containing the UAV target in the radar tracking result information of the preprocessed information set to obtain a IMF component signals and a margin signal. S3112, the a IMF component signals are filtered to obtain s high-frequency components and d low-frequency components; S3113, calculate the mean of each of the d low-frequency components to obtain d first mean components; S3114, perform wavelet threshold denoising on the s high-frequency components to obtain the denoised s high-frequency components; S3115, calculate the mean of each of the denoised s high-frequency components to obtain s second mean components; S3116, Perform mean calculation on the residual signal to obtain a third mean component; S3117, using the d first mean components, s second mean components, and 1 third mean component, a radar feature vector is constructed. The elements of the radar feature vector include d first mean components, s second mean components, and 1 third mean component.

[0030] The step of filtering the a IMF component signals to obtain s high-frequency components and d low-frequency components involves using a preset high-pass filter to filter the a IMF component signals to obtain s high-frequency components, and using a preset low-pass filter to filter the a IMF component signals to obtain d low-frequency components.

[0031] The step of performing image feature extraction processing on images containing drone targets from the video surveillance results information in the preprocessed information set to obtain image feature vectors includes: S3121, Perform image target detection on the images containing drone targets in the video surveillance result information of the preprocessed information set to obtain sub-image information; the sub-image information contains drone targets; S3122, Perform a two-dimensional discrete cosine transform on the sub-image information to obtain the image feature vector.

[0032] The two-dimensional discrete cosine transform (2D-DCT) first performs a one-dimensional DCT on each row of the sub-image information, then performs a one-dimensional DCT on each column of the result to obtain two-dimensional coefficients, and finally arranges the two-dimensional coefficients in a specific order (such as Zigzag scanning) to form an image feature vector.

[0033] The image target detection can employ the sliding window method, the minimum closed rectangle extraction method, or a deep learning-based detection method.

[0034] The minimum bounding rectangle extraction involves extracting the minimum bounding rectangle from the segmented binary target region (e.g., an aircraft mask obtained through preprocessing). Common methods include the minimum coverage method, which involves rotating the target region, constructing rectangles parallel to the coordinate axes, and selecting the one with the smallest area.

[0035] The step of performing signal feature extraction processing on the received signal in the ADS-B device received result information in the preprocessed information set to obtain a signal vector includes: S3131, Perform reduced-order variational mode decomposition on the received signal in the ADS-B device received result information in the preprocessed information set to obtain the first transformation sequence; S3132, Using a signal reconstruction model, the first transform sequence is reconstructed to obtain the feature signal; S3133, Discretely sample the feature signal to obtain a signal vector.

[0036] The signal reconstruction model expression is: In the formula, For the preset first in the signal reconstruction model One reconstruction filter, , The characteristic signal is N, and N is the total number of reconstruction filters. This is the first transformation sequence.

[0037] The first The reconstruction filter can be a pre-defined bandpass filter for the i-th frequency band. The sampling frequency for discretely sampling the feature signal can be the center frequency of the N / 2-th frequency band.

[0038] The signal reconstruction model can accurately eliminate multipath interference, electromagnetic clutter, and invalid noise components caused by signal transmission loss in low-altitude environments by using multi-band reconstruction filters to reconstruct the raw monitoring signals received by ADS-B devices. This process fully preserves the core effective features representing the UAV's attributes and motion state in the ADS-B signal, avoiding feature loss problems caused by single signal processing methods. This expression, through targeted reconstruction in different frequency bands, adapts to the characteristics of ADS-B signals in low-altitude scenarios, which are susceptible to interference and have poor stability. It significantly improves the purity of subsequent signal feature extraction while ensuring the consistency between the reconstructed feature signal and the original effective signal. This solves the technical challenges of signal distortion and feature blurring in complex low-altitude environments, providing high-quality foundational data for subsequent signal vector extraction and target recognition.

[0039] The process of performing type recognition processing on the classification feature set based on the preset target standard feature set to obtain the target type recognition result includes: S321, a type recognition model is constructed based on the preset target standard feature set and the classification feature set; the preset target standard feature set includes radar feature standard vector, image feature standard vector and signal standard vector for each target type, and each target type has a corresponding sequence number; S322, Solve the type recognition model to obtain the recognition type number; S323, Determine the target type corresponding to the identification type number, which is the target type identification result.

[0040] The expression for the type recognition model is: Where I is the identification type index, and i is the index of the target type in the target standard feature set. , and These are the k-th elements of the radar feature standard vector, image feature standard vector, and signal standard vector of target type i in the target standard feature set, respectively. , and These are the k-th elements of the radar feature vector, image feature vector, and signal vector in the classification feature set, respectively, where K is the total number of elements in the radar feature vector; This represents the i corresponding to the minimum value in the expression.

[0041] The radar feature vector, image feature vector, and signal vector all have the same dimension; The type identification model is solved to obtain the identification type number, which can be achieved using a numerical optimization algorithm or a particle filter algorithm.

[0042] The described type recognition model combines the feature differences of multi-source heterogeneous monitoring data and adopts a multi-dimensional feature difference joint calculation method to achieve comprehensive matching and comparison of radar features, image features, ADS-B signal features, and standard target features. Compared with a single feature matching model, it significantly improves the anti-interference capability and recognition accuracy of low-altitude UAV target type recognition. This expression comprehensively considers three core matching indicators: feature distribution similarity, absolute value of feature amplitude difference, and feature Euclidean distance. It can effectively avoid recognition deviations caused by single monitoring source data errors and environmental interference. It is adapted to the actual situation of multi-source data with temporal deviations and amplitude fluctuations in low-altitude scenarios, and achieves complementary fusion of features from different monitoring sources. It accurately distinguishes UAVs from interfering targets such as low-altitude birds, small aircraft, and debris, narrowing the target type matching range and avoiding misidentification and omission. At the same time, it is adapted to the feature differences of various low-altitude UAV models, ensuring the universality and stability of the recognition results.

[0043] The target motion velocity information in the preprocessed information set is fused and estimated based on the preset target standard feature set and target type identification results to obtain target velocity estimation results, including: Based on the preset target standard feature set and target type recognition results, weight vector calculation is performed to obtain the weight vector. Based on the weight vector, the target motion velocity information in the preprocessed information set is weighted and summed to obtain the target velocity estimation result. The weighted summation involves multiplying the first, second, and third elements of the weight vector by the target velocity information from the radar tracking results, video surveillance results, and ADS-B device reception results, respectively, and then summing all the multiplication results to obtain the target velocity estimation result.

[0044] The expression for calculating the weight vector is: ; ; ; ; in, The cosine similarity between the standard signal vector representing the target type identification result in the target standard feature set and the signal vector in the classification feature set. This represents the i-th element of the weight vector. to These are the intermediate calculation quantities from the 1st to the 3rd respectively. and They are respectively Norm calculation and Frobenius norm calculation The i-th element of the weight vector and These are the radar feature vector and image feature vector in the classification feature set, respectively. and These are the radar feature standard vector and image feature standard vector for target type i, respectively, in the target standard feature set.

[0045] The weight vector calculation expression, based on target type recognition results and combined with radar feature matching degree, image feature matching degree, and ADS-B signal feature cosine similarity, achieves adaptive weight allocation of multi-source motion velocity data. This breaks the limitations of traditional fixed-weight fusion, making the velocity fusion results more consistent with the reliability of actual monitoring data. The expression quantifies the deviation between radar and image monitoring features and standard features through norm calculation, and characterizes the matching degree of ADS-B signal features through cosine similarity. Monitoring sources with smaller deviations and higher matching degrees are assigned higher weights, while those with larger deviations are assigned lower weights. This automatically eliminates the negative impact of monitoring source data with large errors and severe interference, fully leveraging the dominant role of highly reliable monitoring source data.

[0046] In all embodiments of the present invention, the variables involved in all computational expressions or mathematical functions have been dimensionlessized before computation.

[0047] In all embodiments of the present invention, the values ​​of the independent variables in the input of all computational expressions or mathematical functions meet the reasonable requirements of the input range of the computational expressions or mathematical functions, and can ensure that the computational expressions or mathematical functions can be calculated smoothly without violating physical laws or mathematical rules.

[0048] This invention constructs a scientific type recognition model and combines it with a standard feature set to achieve accurate target type matching. At the same time, it dynamically calculates weights based on the recognition results and performs weighted fusion estimation of multi-source motion velocity information, taking into account both the accuracy of target type recognition and the precision of motion parameter estimation. Compared with traditional single recognition methods, it effectively reduces the type misjudgment rate and velocity estimation error, and can quickly and accurately output complete UAV target recognition results. It is suitable for the real-time and high-precision recognition requirements of low-altitude security control and has stronger practical engineering application value.

[0049] The overall technical process of this invention is logically clear and highly operable. The preprocessing stage and the fusion recognition stage are closely connected. It has made targeted optimizations to address the pain points of recognizing small, slow-moving drones at low altitudes. It can be adapted to existing low-altitude monitoring equipment without complex hardware modifications, effectively improving the recognition performance of existing low-altitude monitoring systems, reducing the cost of low-altitude security management and control, and promoting the intelligent and precise development of low-altitude drone management and control technology.

[0050] A second aspect of this invention discloses a low-altitude unmanned aerial vehicle (UAV) target identification device based on heterogeneous information fusion, the device comprising: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the low-altitude UAV target recognition method based on heterogeneous information fusion.

[0051] In a third aspect of this invention, a computer-readable storage medium is disclosed, wherein the computer-readable storage medium stores computer instructions, which, when invoked by a computer, are used to execute the aforementioned low-altitude unmanned aerial vehicle target recognition method based on heterogeneous information fusion.

[0052] In a fourth aspect of this invention, an information data processing terminal is disclosed, which is used to implement the low-altitude unmanned aerial vehicle target recognition method based on heterogeneous information fusion.

[0053] The above description is merely an embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of the claims of the present invention.

Claims

1. A low-altitude unmanned aerial vehicle target identification method based on heterogeneous information fusion, characterized in that, include: S1, obtain a set of heterogeneous monitoring information for UAVs in low-altitude environments; the set of heterogeneous monitoring information for UAVs includes radar tracking results, video surveillance results, and ADS-B device reception results. The radar tracking results include radar echo signals containing UAV targets and target speed information; The video surveillance results include images containing drone targets and information on the target's movement speed; The ADS-B device receives result information, including received signal, target speed information, and timestamp information; S2, preprocess the heterogeneous monitoring information set of the UAV to obtain a preprocessed information set; S3, perform fusion recognition processing on the preprocessed information set to obtain the recognition result information of the low-altitude UAV target; The identification results include target type identification results and target velocity estimation results.

2. The low-altitude unmanned aerial vehicle target identification method based on heterogeneous information fusion according to claim 1, characterized in that, The preprocessing of the heterogeneous monitoring information set of the UAV to obtain a preprocessed information set includes: S21, perform data category cleaning processing on the heterogeneous monitoring information set of the UAV to obtain the first information set; S22, perform registration processing on the first information set to obtain a preprocessed information set.

3. The low-altitude UAV target recognition method based on heterogeneous information fusion as described in claim 1, characterized in that, The process of fusing and recognizing the preprocessed information set to obtain the recognition result information of the low-altitude UAV target includes: S31, Perform classification feature extraction processing on the preprocessed information set to obtain a classification feature set; S32, Based on the preset target standard feature set, perform type recognition processing on the classification feature set to obtain the target type recognition result; S33, based on the preset target standard feature set and target type identification results, the target motion speed information in the preprocessed information set is fused and estimated to obtain the target speed estimation result; S34. Based on the target type identification result and the target velocity estimation result, the identification result information of the low-altitude UAV target is constructed.

4. The low-altitude UAV target recognition method based on heterogeneous information fusion as described in claim 3, characterized in that, The step of performing classification feature extraction on the preprocessed information set to obtain a classification feature set includes: S311, Perform echo feature extraction processing on the radar echo signal containing the UAV target in the radar tracking result information in the preprocessed information set to obtain the radar feature vector; S312, perform image feature extraction processing on the images containing drone targets in the video surveillance result information of the preprocessed information set to obtain image feature vectors; S313, Perform signal feature extraction processing on the received signal in the ADS-B device received result information in the preprocessed information set to obtain a signal vector; S314, using the radar feature vector, image feature vector and signal vector, a classification feature set is constructed.

5. The low-altitude UAV target recognition method based on heterogeneous information fusion as described in claim 4, characterized in that, The process of extracting echo features from radar echo signals containing UAV targets in the preprocessed information set to obtain radar feature vectors includes: S3111, Empirical mode decomposition is performed on the radar echo signal containing the UAV target in the radar tracking result information of the preprocessed information set to obtain a IMF component signals and a margin signal. S3112, the a IMF component signals are filtered to obtain s high-frequency components and d low-frequency components; S3113, calculate the mean of each of the d low-frequency components to obtain d first mean components; S3114, perform wavelet threshold denoising on the s high-frequency components to obtain the denoised s high-frequency components; S3115, calculate the mean of each of the denoised s high-frequency components to obtain s second mean components; S3116, Perform mean calculation on the residual signal to obtain a third mean component; S3117, using the d first mean components, s second mean components and 1 third mean component, a radar feature vector is constructed.

6. The low-altitude UAV target recognition method based on heterogeneous information fusion as described in claim 4, characterized in that, The step of performing signal feature extraction processing on the received signal in the ADS-B device received result information in the preprocessed information set to obtain a signal vector includes: S3131, Perform reduced-order variational mode decomposition on the received signal in the ADS-B device received result information in the preprocessed information set to obtain the first transformation sequence; S3132, using a signal reconstruction model, the first transformed sequence is reconstructed to obtain the feature signal; the expression of the signal reconstruction model is: In the formula, For the preset first in the signal reconstruction model One reconstruction filter, , The characteristic signal is N, and N is the total number of reconstruction filters. This is the first transformation sequence; S3133, Discretely sample the feature signal to obtain a signal vector.

7. The low-altitude UAV target recognition method based on heterogeneous information fusion as described in claim 3, characterized in that, The process of performing type recognition processing on the classification feature set based on the preset target standard feature set to obtain the target type recognition result includes: S321, a type recognition model is constructed based on the preset target standard feature set and the classification feature set; the preset target standard feature set includes radar feature standard vector, image feature standard vector and signal standard vector for each target type, and each target type has a corresponding sequence number; S322, Solve the type recognition model to obtain the recognition type number; S323, Determine the target type corresponding to the identification type number, which is the target type identification result; The expression for the type recognition model is: Where I is the identification type index, and i is the index of the target type in the target standard feature set. , and These are the k-th elements of the radar feature standard vector, image feature standard vector, and signal standard vector of target type i in the target standard feature set, respectively. , and These are the k-th elements of the radar feature vector, image feature vector, and signal vector in the classification feature set, respectively, where K is the total number of elements in the radar feature vector.

8. A low-altitude unmanned aerial vehicle (UAV) target recognition device based on heterogeneous information fusion, characterized in that, The device includes: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the low-altitude UAV target recognition method based on heterogeneous information fusion as described in any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions, which, when invoked by a computer, are used to execute the low-altitude unmanned aerial vehicle target recognition method based on heterogeneous information fusion as described in any one of claims 1 to 7.

10. An information data processing terminal, characterized in that, The information data processing terminal is used to implement the low-altitude UAV target recognition method based on heterogeneous information fusion as described in any one of claims 1 to 7.

Images