Aerial moving target detection method and device, equipment and storage medium

Abstract

The invention relates to an air moving target detection method and device, equipment and a storage medium. According to the main technical scheme, the method comprises the following steps: acquiring single-view complex data, and preprocessing the single-view complex data to obtain preprocessed data; sub-aperture segmentation is carried out on the preprocessed data to obtain sub-aperture segmentation data, and the sub-aperture segmentation data comprise multi-scene same-angle sub-aperture data and single-scene different-angle sub-aperture data; performing low-rank decomposition and constant false alarm rate detection on the multi-scene same-angle sub-aperture data to obtain multi-temporal detection data; performing low-rank decomposition and constant false alarm rate detection on the single-scene different-angle sub-aperture data to obtain single-time-phase detection data; and fusing the multi-temporal detection data and the single-temporal detection data to obtain air moving target detection data. The method not only considers the aerial moving targets continuously appearing at multiple imaging moments, but also considers the targets appearing only at a few moments or sparsely sampled in time sequence, so that the effect of improving the extraction capability of the aerial moving targets is achieved.

Description

Technical Field

[0001] This application relates to the field of target detection technology, and in particular to a method, apparatus, equipment and storage medium for detecting moving targets in the air. Background Technology

[0002] As is well known, Synthetic Aperture Radar (SAR) is an active microwave imaging radar system. Mounted on a moving platform (such as a satellite, aircraft, or UAV), it transmits electromagnetic pulses and receives echo signals reflected from ground objects. Using advanced signal processing techniques, it synthesizes an equivalent "ultra-long antenna" to obtain high-resolution two-dimensional images (range and azimuth). Its greatest advantage is its all-weather, all-day operation capability, able to penetrate clouds, fog, rain, snow, and a certain degree of surface vegetation, making it an important supplement to optical remote sensing.

[0003] In traditional implementations, moving target detection using single-channel SAR data primarily focuses on ground or sea scenes, and most methods are based on processing full-aperture imaging results. Common techniques include threshold segmentation of single-scene images, multi-temporal amplitude difference analysis, and simple change detection. These methods typically treat each image as an observation of a static scene and identify anomalous regions by comparing amplitude differences or local statistical characteristics of images at different times.

[0004] However, the aforementioned methods do not specifically address the motion characteristics of moving targets in the azimuth imaging process. Moving targets in the azimuth are prone to defocusing in full-aperture imaging, leading to a reduced signal-to-noise ratio. Furthermore, when thresholding or multi-temporal differential imaging is applied on top of this, the contrast between the moving target and the complex background is insufficient, making it easily obscured by strongly scattered backgrounds. In other words, while these methods can be effective in situations with relatively simple backgrounds or significant target changes, for moving targets in spaceborne SAR images, due to their small size, weak signal, and complex motion, relying solely on full-aperture single-scene images or simple multi-temporal differential imaging often fails to achieve ideal detection results. Summary of the Invention

[0005] Based on this, this application provides a method, apparatus, device, and storage medium for detecting moving targets in the air. By acquiring single-view complex data, and performing preprocessing, sub-aperture segmentation, low-rank decomposition, constant false alarm rate (CFAR) detection, and single-multi-temporal data fusion on the single-view complex data, the detection result of moving targets in the air is obtained. This approach considers both moving targets that continuously appear at multiple imaging moments and targets that appear only at a few moments or have sparse temporal sampling, thereby improving the ability to extract moving targets in the air.

[0006] Firstly, a method for detecting moving targets in the air is provided, the method comprising: Obtain single-view complex data, preprocess the single-view complex data to obtain preprocessed data; The preprocessed data is segmented into sub-aperture data, which includes sub-aperture data from multiple scenes at the same angle and sub-aperture data from a single scene at different angles. Low-rank decomposition and constant false alarm rate detection are performed on multiple scenes of sub-aperture data at the same angle to obtain multi-temporal detection data; Low-rank decomposition and constant false alarm rate detection were performed on sub-aperture data at different angles of a single scene to obtain single-phase detection data. Multi-temporal detection data and single-temporal detection data are fused to obtain airborne moving target detection data.

[0007] According to one feasible method in an embodiment of this application, preprocessing of single-view complex data yields preprocessed data, including: Logarithmic transformation is performed on the single-view complex data to obtain logarithmically transformed data; The logarithmically transformed data is then subjected to mean filtering to obtain mean-filtered data. Radiometric correction is applied to the mean-filtered data to obtain radiometrically corrected data. Image registration is performed on the radiometrically corrected data to obtain preprocessed data.

[0008] According to one feasible method in an embodiment of this application, sub-aperture segmentation is performed on the preprocessed data to obtain sub-aperture segmented data, including: Perform a Fourier transform on the preprocessed data to obtain Doppler data; The Doppler data is divided into frequency bands, and a frequency domain weighting function is constructed. The Doppler data is weighted in the frequency domain using a frequency domain weighting function, and then subjected to inverse Fourier transform to obtain sub-aperture segmentation data.

[0009] According to one feasible method in an embodiment of this application, low-rank decomposition and constant false alarm rate detection are performed on multiple scenes of sub-aperture data at the same angle to obtain multi-temporal detection data, including: The sub-aperture data of multiple scenes with the same angle are expanded into column vectors according to pixel position and sorted according to time order to obtain the matrix data of multiple scenes with the same angle. Low-rank decomposition is performed on multiple scenes of sub-aperture matrix data with the same angle to obtain multiple sparse sequence data. Constant false alarm rate detection is performed on multiple sparse sequence data to obtain multi-temporal detection data.

[0010] According to one feasible method in an embodiment of this application, low-rank decomposition is performed on multiple scenes of sub-aperture matrix data at the same angle to obtain multiple scenes of sparse sequence data, including: Based on the low-rank sparse decomposition model, the sub-aperture matrix data of multiple scenes with the same angle is decomposed into a low-rank part and a sparse part. The low-rank part and the sparse part are solved by the preset low-rank constraint parameters and the preset sparsity constraint parameters to obtain multiple sparse term data. The sparse data from multiple scenes are rearranged column by column to obtain a sparse sequence of multiple scenes.

[0011] According to one feasible method in the embodiments of this application, constant false alarm rate (CFAR) detection is performed on sparse sequence data to obtain multi-temporal detection data, including: The pixel data to be detected in the sparse sequence data is set as the target area, and the background area is determined based on the target area; Calculate the probability density function of the background area, and obtain the detection threshold value based on the preset false alarm probability value and probability density function; Pixel detection data is obtained based on the pixel data to be detected and the detection threshold value; Pixel detection data are merged according to spatial location to obtain multi-temporal detection data.

[0012] According to one feasible method in an embodiment of this application, low-rank decomposition and constant false alarm rate detection are performed on sub-aperture data at different angles of a single scene to obtain single-temporal detection data, including: The sub-aperture data of different angles in a single scene are expanded into column vectors according to pixel positions and sorted according to the sub-aperture index order to obtain the sub-aperture matrix data of different angles in a single scene; Low-rank decomposition was performed on the sub-aperture matrix data of a single scene at different angles to obtain sparse sequence data of a single scene; Constant false alarm rate detection is performed on single-scene sparse sequence data to obtain single-phase detection data.

[0013] Secondly, an aerial moving target detection device is provided, the device comprising: The preprocessing unit is used to acquire single-view complex data, preprocess the single-view complex data, and obtain preprocessed data. The sub-aperture segmentation unit is used to segment the preprocessed data into sub-aperture data, which includes sub-aperture data from multiple scenes at the same angle and sub-aperture data from a single scene at different angles. The multi-temporal detection unit is used to perform low-rank decomposition and constant false alarm rate detection on multiple scenes of sub-aperture data at the same angle to obtain multi-temporal detection data. The single-phase detection unit is used to perform low-rank decomposition and constant false alarm rate detection on sub-aperture data at different angles of a single scene to obtain single-phase detection data. The fusion result unit is used to fuse multi-temporal detection data and single-temporal detection data to obtain airborne moving target detection data.

[0014] Thirdly, a computer device is provided, comprising: At least one processor; and A memory that is communicatively connected to at least one processor; wherein, The memory stores computer instructions that can be executed by at least one processor to enable the at least one processor to perform the methods involved in the first aspect above.

[0015] Fourthly, a computer-readable storage medium is provided, having stored thereon computer instructions, characterized in that the computer instructions are used to cause a computer to perform the methods involved in the first aspect above.

[0016] According to the technical content provided in the embodiments of this application, single-view complex data is acquired, preprocessed to obtain preprocessed data, and then sub-aperture segmentation is performed on the preprocessed data to obtain sub-aperture segmentation data. The sub-aperture segmentation data includes sub-aperture data from multiple scenes at the same angle and sub-aperture data from a single scene at different angles. Low-rank decomposition and constant false alarm rate (CFAR) detection are performed on the sub-aperture data from multiple scenes at the same angle to obtain multi-temporal detection data. Low-rank decomposition and CFAR detection are performed on the sub-aperture data from a single scene at different angles to obtain single-temporal detection data. The multi-temporal detection data and the single-temporal detection data are then fused to obtain aerial moving target detection data. This application obtains aerial moving target detection results by acquiring single-view complex data, performing preprocessing, sub-aperture segmentation, low-rank decomposition, CFAR detection, and fusion of single and multi-temporal detection data. This achieves the effect of improving the extraction capability of aerial moving targets, considering both aerial moving targets that continuously appear at multiple imaging moments and targets that only appear at a few moments or have sparse temporal sampling. Attached Figure Description

[0017] Figure 1 This is a system application diagram of an aerial moving target detection method in one embodiment; Figure 2 This is a flowchart illustrating an aerial moving target detection method in one embodiment; Figure 3 This is a flowchart of an aerial moving target detection method in one embodiment; Figure 4 This is a structural block diagram of an aerial moving target detection device in one embodiment; Figure 5 This is a schematic structural diagram of a computer device in one embodiment. Detailed Implementation

[0018] The present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the scope of the present application.

[0019] For ease of understanding, the system to which this application applies will first be described. The aerial moving target detection method provided in this application can be applied to, for example... Figure 1 The system architecture shown includes: a space platform 101, a network, and a server 103. The space platform 101 and the server 103 communicate with each other via a network. The space platform 101 is equipped with a spaceborne synthetic aperture radar (SAR). The spaceborne SAR can detect moving targets in the air in real time and obtain single-look complex (SLC) data. The server can acquire this single-look complex data for further processing. Specifically, the server acquires SLC data detected by spaceborne SAR, preprocesses the SLC data to obtain preprocessed data, performs sub-aperture segmentation on the preprocessed data to obtain sub-aperture segmented data, which includes multiple scenes of sub-aperture data at the same angle and single scenes of sub-aperture data at different angles, performs low-rank decomposition and constant false alarm rate (CFAR) detection on the multiple scenes of sub-aperture data at the same angle to obtain multi-temporal detection data, performs low-rank decomposition and CFAR detection on the single scene of sub-aperture data at different angles to obtain single-temporal detection data, and fuses the multi-temporal detection data and single-temporal detection data to obtain airborne moving target detection data. The space platform can be a satellite platform, and the server 103 can be implemented using a standalone server or a server cluster consisting of multiple servers.

[0020] Figure 2 This is a flowchart illustrating an aerial moving target detection method provided in an embodiment of this application. The method can be performed by, for example... Figure 1 Server 103 in the system shown is executing. For example... Figure 2 , Figure 3 As shown, the method may include the following steps: Step S201: Obtain single-view complex data, preprocess the single-view complex data to obtain preprocessed data.

[0021] Among them, single-view complex data, or SLC data, is a complex image after focusing, with each pixel representing a complex pixel value. Here, single-view refers to observation from a single perspective, a single image, or a single moment, which preserves the highest spatial resolution; complex refers to the amplitude and phase information contained in each pixel, with the amplitude related to the backscattering intensity of the ground object, reflecting the characteristics of the detected target (such as roughness and humidity); the phase contains distance information to the satellite, and because the wavelength is very short (on the order of centimeters), the phase is extremely sensitive to small changes in distance (such as millimeter-level deformation), which is the basis of temporal interferometry.

[0022] Specifically, SLC data of the same region are acquired from a spaceborne SAR satellite at multiple imaging times. Assume the complex pixel at the t-th imaging time can be represented as: ; Where r represents the range pixel index (image row number); a represents the azimuth pixel index (image column number), both of which can be directly obtained from the SLC data, i.e., the dimensions of the SLC image matrix; T represents the number of time-series images, determined by the number of time-series SLC data scenes acquired. Preprocessing operations are performed on the SLC data to obtain time-series images with radiometric and geometric consistency, i.e., preprocessed data, to further ensure accurate detection of moving targets in the air later.

[0023] Step S203: Perform sub-aperture segmentation on the preprocessed data to obtain sub-aperture segmentation data.

[0024] Sub-aperture segmentation is a SAR processing technique based on the azimuth Doppler domain; sub-aperture segmentation data includes multiple scenes of sub-aperture data at the same angle and single scenes of sub-aperture data at different angles.

[0025] Here, since the preprocessed data includes multiple preprocessed images, a discrete Fourier transform can be performed on each preprocessed image to obtain full-aperture Doppler data. Sub-aperture segmentation is then performed on the full-aperture Doppler data, that is, the frequency band of the full-aperture Doppler data is divided into several sub-frequency bands and the corresponding sub-aperture images are reconstructed to obtain sub-aperture segmentation data. That is, both multi-scene sub-aperture data at the same angle and single-scene sub-aperture data at different angles can be obtained.

[0026] Step S205: Perform low-rank decomposition and constant false alarm rate detection on multiple scenes of sub-aperture data at the same angle to obtain multi-temporal detection data.

[0027] Low-rank decomposition refers to approximating a complex, high-dimensional matrix that may contain a lot of noise or redundant information by decomposing it into the product of two (or more) smaller, simpler matrices. The rank of this approximate matrix is ​​much smaller than that of the original matrix. Constant false alarm rate (CFAR) detection is a detection method that adaptively determines a detection threshold using statistical information of the environmental background under a pre-given false alarm probability, and uses this threshold to make a decision on the unit to be inspected.

[0028] Specifically, for multiple sub-aperture data at the same angle, based on the sub-aperture images, low-rank decomposition is used from the temporal dimension to distinguish the static background that changes little over time from the moving target component that changes much over time, thus obtaining the temporal component that highlights the response of the moving target. On this basis, further image transformation, namely constant false alarm rate detection, is performed to obtain multi-temporal detection data.

[0029] Step S207: Perform low-rank decomposition and constant false alarm rate detection on the sub-aperture data of different angles in a single scene to obtain single-phase detection data.

[0030] Specifically, for single-scene sub-aperture data from different angles, based on the sub-aperture images, low-rank decomposition is used in the spatial dimension to distinguish the background component with small changes between different sub-apertures from the moving target component with significant changes between different sub-apertures, thus obtaining the single-temporal moving target component that highlights the moving target response. In other words, under single-scene conditions, the difference information between different sub-aperture images at the same time is used to construct the detection of single-temporal aerial moving targets. On this basis, further image transformation, i.e., constant false alarm rate detection, is performed to obtain single-temporal detection data.

[0031] Step S209: Fuse the multi-temporal detection data and the single-temporal detection data to obtain the airborne moving target detection data.

[0032] Specifically, after obtaining multi-phase detection data and single-phase detection data, the multi-phase detection data and single-phase detection data are fused spatially. Assuming that the multi-phase detection data can be represented by Dmulti(r,a) and the single-phase detection data can be represented by Dsingle(r,a), the fusion is performed using a bitwise OR operation, and the specific expression can be expressed as follows: ; After fusion, the airborne moving target detection data can be obtained, which is the final moving target detection result.

[0033] As can be seen, the embodiments of this application acquire single-view complex data, preprocess the single-view complex data to obtain preprocessed data; perform sub-aperture segmentation on the preprocessed data to obtain sub-aperture segmentation data, which includes sub-aperture data of multiple scenes at the same angle and sub-aperture data of a single scene at different angles; perform low-rank decomposition and constant false alarm rate (CFAR) detection on the sub-aperture data of multiple scenes at the same angle to obtain multi-temporal detection data; perform low-rank decomposition and CFAR detection on the sub-aperture data of a single scene at different angles to obtain single-temporal detection data; and fuse the multi-temporal detection data and the single-temporal detection data to obtain aerial moving target detection data. The aforementioned operations, through acquiring single-view complex data, performing preprocessing, sub-aperture segmentation, low-rank decomposition, CFAR detection, and fusion of single- and multi-temporal detection data, yield aerial moving target detection results. This achieves the effect of considering both aerial moving targets that continuously appear at multiple imaging moments and targets that only appear at a few moments or have sparse temporal sampling, thereby improving the ability to extract aerial moving targets.

[0034] The steps in the above method flow are described in detail below. First, the step 201, "preprocessing the single-view complex data to obtain preprocessed data," will be described in detail with reference to the embodiment.

[0035] Logarithmic transformation is performed on the single-view complex data to obtain logarithmically transformed data; mean filtering is performed on the logarithmically transformed data to obtain mean-filtered data; radiometric correction is performed on the mean-filtered data to obtain radiometrically corrected data; image registration is performed on the radiometrically corrected data to obtain preprocessed data.

[0036] Specifically, a logarithmic transformation is performed on the single-view complex data to obtain logarithmically transformed data, thereby compressing the dynamic range of the image. The logarithmic amplitude image at the t-th imaging time can be represented by I... t If (r, a) represents a variable, then the specific expression can be represented as follows: ; Among them, A t (r,a) represents the amplitude graph; |·| represents the complex modulus; I t (r,a) represents the logarithmic magnitude image.

[0037] The logarithmically transformed data is then subjected to mean filtering to obtain mean-filtered data. Taking time t as an example again, that is, for I... t (r,a) employs local mean filtering, averaging the pixels using an m×n sliding window to obtain the mean-filtered data. Here, m and n are the window radii, which can be preset according to the resolution and target scale. It should be noted that the purpose of this mean filtering is to reduce local fluctuations caused by speckle noise, thereby improving the stability of subsequent threshold decisions.

[0038] Radiometric correction is performed on the mean-filtered data to obtain radiometrically corrected data, ensuring that the same background area has a consistent brightness level at different imaging times. Here, radiometric correction of the mean-filtered data involves performing radiometric correction on the mean-filtered images at different times. An amplitude normalization method is applied to each image to make the mean and standard deviation of each image equal to a reference value. The reference value is the mean and standard deviation of T images.

[0039] Image registration is performed on the radiometrically corrected data to obtain preprocessed data, thus ensuring that the same pixel location in the time-series images corresponds to the same land cover region. Here, after the radiometric correction operation, the image at a certain imaging time is used as the reference image. Feature points are extracted between the reference image and the image to be registered for feature matching, such as scale-invariant feature points. Robust estimation is used to eliminate mismatched points, resulting in a geometric transformation model, which can be a translation or affine model. Then, interpolation resampling is used to map the image to be registered onto the coordinate system of the reference image, thus obtaining the preprocessed time-series image sequence. The preprocessed image at time t can then be represented as follows: .

[0040] The above operations, through preprocessing operations such as logarithmic transformation, mean filtering, radiometric correction, and image registration on single-view complex data, aim to obtain a time-series image that is radiometrically and geometrically consistent.

[0041] The following describes step S203, namely "segmenting the preprocessed data into sub-apertures to obtain sub-aperture segmented data", in detail with reference to the embodiments.

[0042] The preprocessed data is subjected to Fourier transform to obtain Doppler data; the Doppler data is divided into frequency bands and a frequency domain weighting function is constructed; the Doppler data is then subjected to frequency domain weighting based on the frequency domain weighting function, and an inverse Fourier transform is performed to obtain sub-aperture segmentation data.

[0043] Specifically, a Fourier transform is performed on each preprocessed image in the preprocessed data to obtain Doppler data, thereby mapping the spatial sampling in the azimuth direction to the Doppler frequency domain, which facilitates subsequent sub-band division. Here, we take... For example, after performing a discrete Fourier transform in the azimuth direction, the Doppler image at time T is obtained, and the specific expression can be represented as follows: ; Among them, f a Indicates the azimuth Doppler frequency; F a {·} represents the Fourier transform along the azimuth direction.

[0044] The Doppler data is divided into frequency bands, and a frequency domain weighting function is constructed. Here, based on a preset number of sub-images, the Doppler data, i.e., the full-aperture Doppler frequency band, is divided into K non-overlapping sub-bands, where the k-th sub-band can be represented as [f k ,min, f k Construct the corresponding frequency domain weighting function W[max]. k (f a The specific expression can be represented as follows: ; Among them, W k (f a The amplitude value of N is obtained by averaging the amplitude values ​​along the distance; r This indicates the number of distance units participating in the average.

[0045] The Doppler data is weighted in the frequency domain using a frequency domain weighting function, and then subjected to inverse Fourier transform to obtain sub-aperture segmentation data. Specifically, for each sub-band of the Doppler data, frequency domain weighting and azimuth-directed inverse Fourier transform are performed to obtain the sub-aperture segmentation data. The expression for the k-th sub-aperture image of the t-th scene in the sub-aperture segmentation data can be represented as follows: ; in, Indicates the inverse Fourier transform along the azimuth direction; I t (k) This represents the imaging data of the t-th scene and the k-th sub-aperture.

[0046] The above operation, before time-series processing, performs sub-aperture segmentation on the images at each time point in the azimuth Doppler domain, decomposing the full aperture imaging process into multiple imaging processes with corresponding shorter synthetic aperture times. That is, the full aperture is divided into several sub-bands according to the Doppler frequency. This can effectively reduce the defocusing of moving targets in the azimuth direction, make the energy of the moving target echo more concentrated in the sub-aperture image, and improve the signal-to-noise ratio of the moving target in the sub-aperture image.

[0047] The following describes in detail step S205, namely, "performing low-rank decomposition and constant false alarm rate detection on multiple scenes of sub-aperture data at the same angle to obtain multi-temporal detection data," with reference to the embodiments.

[0048] Multiple scenes of the same angle sub-aperture data are expanded into column vectors according to pixel position and sorted in chronological order to obtain multi-scene same angle sub-aperture matrix data; low-rank decomposition is performed on the multi-scene same angle sub-aperture matrix data to obtain multi-scene sparse sequence data; constant false alarm rate detection is performed on the multi-scene sparse sequence data to obtain multi-temporal detection data.

[0049] Here, multiple sub-aperture images at the same angle are expanded into column vectors by pixel position and sorted chronologically to obtain a matrix of sub-aperture images at the same angle. Specifically, a temporal data matrix of a sub-aperture at multiple imaging times is constructed, using a sub-aperture index k as the unit. That is, for each sub-aperture image... Expanded by pixel position into a column vector x t (k) Arrange the column vectors of all times in chronological order to obtain sub-aperture matrix data for multiple scenes at the same angle, which can be represented as follows: ; Each sub-aperture image is an N r ×N a The matrix is ​​linearized into a column-major linear form of length N. r N a The vector is such that the same pixel position corresponds to the same row in the vector at different times, so that the "rows" of the matrix correspond to spatial pixels and the "columns" correspond to temporal pixels.

[0050] Low-rank decomposition is performed on multiple scenes of the same angular sub-aperture matrix data to obtain multiple scenes of sparse sequence data. In one feasible approach, the multiple scenes of the same angular sub-aperture matrix data are decomposed into a low-rank part and a sparse part according to a low-rank sparse decomposition model; the low-rank part and the sparse part are solved according to preset low-rank constraint parameters and preset sparsity constraint parameters to obtain multiple scenes of sparse term data; the multiple scenes of sparse term data are rearranged column-wise to obtain multiple scenes of sparse sequence data.

[0051] Low-rank decomposition refers to utilizing the temporal correlation of the background to concentrate background energy into low-rank terms L. (k) And concentrate the local, intermittent abnormal responses into the sparse term S. (k) .

[0052] Here, based on the low-rank sparse decomposition model, the sub-aperture matrix data of multiple scenes with the same angle is decomposed into a low-rank part and a sparse part. Taking a certain sub-aperture index k as an example, the specific expression can be represented as follows: ; Among them, X (k) Represents multiple scenes with the same sub-aperture matrix data; L (k) The lower-rank part is used to represent a static background that changes little with time; S (k) The sparsity part is used to represent the scattering of moving targets in the air and other anomalous scattering that varies significantly over time.

[0053] Solving for the low-rank and sparse parts using preset low-rank and sparsity constraints yields multiple sparse term data. Specifically, the solution is obtained by minimizing the residuals while simultaneously constraining both low-rank and sparsity. Taking a specific sub-aperture index k as an example, the specific expression can be represented as follows: ; Among them, ∥·∥ F Let represent the Frobenius norm; r0 represents the preset low-rank constraint parameter, and s0 represents the preset sparsity constraint parameter. Both r0 and s0 can be preset in advance according to the data scale and target characteristics. Through the above calculation, multiple sparse term data are obtained. It should be noted that this application does not limit the specific solution method, and an iterative numerical method can be used to terminate under the condition of error threshold or maximum number of iterations.

[0054] The multi-scene sparse data is rearranged column-wise to obtain a multi-scene sparse sequence data. That is, the multi-scene sparse data S with index k. (k) Rearrange the columns back into image form to obtain the temporally sparse image sequence S. (k) (r,a) represents multi-scene sparse sequence data. Where S... (k) The tth column s t (k) The sparse vector corresponding to time t is rearranged back to N according to the same rules as the expansion. r ×N a This will give us the sparse sequence data S at time t. t (k) (r,a).

[0055] Constant false alarm rate (CFAR) detection is performed on sparse sequence data to obtain multi-temporal detection data. In one feasible approach, the pixel data to be detected in the sparse sequence data is set as the target area, and a background area is determined based on the target area; the probability density function of the background area is calculated, and a detection threshold is obtained based on a preset false alarm probability value and the probability density function; pixel detection data is obtained based on the pixel data to be detected and the detection threshold; the pixel detection data is merged according to spatial location to obtain multi-temporal detection data.

[0056] In constant false alarm rate (CFAR) detection, the target pixel is generally used as the target area, and a protection zone and a background zone are set around it. The size of these zones can be set based on the characteristics of the background and target points in the image. Since the setting of the protection zone and the background zone are existing technologies, they will not be elaborated on here.

[0057] The probability density function of the background region is calculated, and the detection threshold is obtained based on the preset false alarm probability value and the probability density function. Here, after determining the background region, a histogram of the pixels within the background region can be obtained using MATLAB software. Based on the obtained histogram, the corresponding probability distribution model is automatically selected, thus obtaining the probability density function of the background region, which can be expressed as p(x). It should be noted that the probability distribution model here is generally a normal distribution, and the mean and variance of the normal distribution can be directly obtained from the pixels within the background region.

[0058] After obtaining the probability density function, since the false alarm probability value is preset, the detection threshold value can be obtained based on the preset false alarm probability value and the probability density function. The specific expression can be represented as follows: ; Among them, P fa τ represents the preset false alarm probability value; p(x) represents the probability density function; x represents the pixel to be detected; τ represents the detection threshold value. The detection threshold value τ can be obtained through the above expression.

[0059] Pixel detection data is obtained based on the pixel data to be detected and the detection threshold. That is, each pixel x in the pixel data to be detected is compared with the detection threshold τ. If the amplitude of the pixel x is greater than the detection threshold, it is determined to be a target pixel; otherwise, it is determined to be a background pixel. The specific determination criteria can be expressed as follows: ; Through the above operations, pixel detection data can be obtained, and then pixel detection results corresponding to multiple sub-apertures can be obtained. The multi-temporal detection data corresponding to the k-th sub-aperture can be represented as shown in Figure D. (k) multi (r,a). The multi-scene sparse data is rearranged column-wise to obtain multi-scene sparse sequence data. That is, the detection results of all sub-apertures are merged according to their spatial location to obtain multi-temporal detection results, i.e., multi-scene sparse sequence data, which can be represented as D. multi (r,a).

[0060] The above operation introduces a low-rank decomposition model into the sub-aperture image. From the temporal dimension, the static background that changes little over time is modeled as a low-rank component, while the moving aerial targets and anomalous scattering that change much over time are modeled as sparse components, thus distinguishing between the background and the moving targets. At the same time, the target pixels are determined by constant false alarm rate detection technology, thereby improving the detection accuracy of moving aerial targets.

[0061] Finally, the above step S207, namely "performing low-rank decomposition and constant false alarm rate detection on sub-aperture data at different angles of a single scene to obtain single-phase detection data", will be described in detail with reference to the embodiments.

[0062] The sub-aperture data of different angles in a single scene are expanded into column vectors according to pixel positions and sorted according to the sub-aperture index order to obtain the sub-aperture matrix data of different angles in a single scene; the sub-aperture matrix data of different angles in a single scene is decomposed into low-rank to obtain the sparse sequence data of a single scene; constant false alarm rate detection is performed on the sparse sequence data of a single scene to obtain the single-phase detection data.

[0063] Here, the sub-aperture data at different angles of a single scene are expanded into column vectors by pixel position and sorted according to the sub-aperture index order to obtain the sub-aperture matrix data at different angles of a single scene. Specifically, for a certain imaging time t0, all its sub-aperture images are taken. The images of different sub-apertures are organized into a data matrix along the pixel dimension. For example, each sub-aperture image is expanded into a column vector and arranged by sub-aperture index to obtain sub-aperture matrix data for different angles of a single scene. The specific expression can be represented as follows: ; in, Let represent the column vector of the k-th sub-aperture image after it has been flattened at time t0.

[0064] Low-rank decomposition is performed on the sub-aperture matrix data of a single scene at different angles to obtain sparse sequence data of the single scene. The specific expression can be represented as follows: ; Among them, L t0 The lower rank component represents the background component that varies little across different sub-apertures; S t0 This represents the sparsity component, corresponding to the moving target component that varies significantly across different sub-apertures. It should be noted that for L... t0 and S t0 The specific solution process and L (k) Low-rank part and S (k) The principle of sparse parts is exactly the same, so it will not be repeated here.

[0065] After obtaining S t0 Then, the sparse part S t0 Rearranged into an image format, a single-phase sparse image R is obtained. single , t0 (r,a) represents a single-scene sparse sequence of data.

[0066] Constant false alarm rate (CFAR) detection is performed on single-scene sparse sequence data to obtain single-temporal detection data. Specifically, in R... single , t0 A constant false alarm rate detector is applied to (r,a), and a detection threshold is set based on the statistical characteristics of the local background to obtain the single-phase detection result D at time t0. single , t0(r,a). The single-phase detection results at all times t are merged according to time to obtain the single-phase detection result D. single (r,a) represents the single-phase detection data. It should be noted that the specific process of constant false alarm rate (CFAR) detection is exactly the same as the detection process for multi-scene sparse sequence data described above; therefore, it will not be repeated here.

[0067] The above operation improves the detection capability for moving targets in the air with short appearance times or sparse temporal sampling by constructing a single-phase detection process at the single-scene image level using the difference information between different sub-apertures at the same time.

[0068] It should be understood that, although Figure 2 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated in this application, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Furthermore, Figure 2 , Figure 3 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0069] Figure 4 This is a schematic diagram of the structure of an aerial moving target detection device provided in an embodiment of this application. The device can be installed in... Figure 1 The server in the system shown is used to perform, for example... Figure 2 The method flow is shown below. Figure 4 As shown, the device may include: a preprocessing unit 401, a sub-aperture segmentation unit 403, a multi-temporal phase detection unit 405, a single-temporal phase detection unit 407, and a fusion result unit 409. The main functions of each component module are as follows: Preprocessing unit 401 is used to acquire single-view complex data, preprocess the single-view complex data, and obtain preprocessed data; The sub-aperture segmentation unit 403 is used to segment the preprocessed data into sub-aperture data to obtain sub-aperture segmentation data. The sub-aperture segmentation data includes sub-aperture data of multiple scenes at the same angle and sub-aperture data of a single scene at different angles. The multi-temporal detection unit 405 is used to perform low-rank decomposition and constant false alarm rate detection on multiple scenes of sub-aperture data at the same angle to obtain multi-temporal detection data. The single-phase detection unit 407 is used to perform low-rank decomposition and constant false alarm rate detection on sub-aperture data at different angles of a single scene to obtain single-phase detection data. The fusion result unit 409 is used to fuse multi-temporal detection data and single-temporal detection data to obtain airborne moving target detection data.

[0070] In one embodiment, the preprocessing unit 401 is further configured to: Logarithmic transformation is performed on the single-view complex data to obtain logarithmically transformed data; The logarithmically transformed data is then subjected to mean filtering to obtain mean-filtered data. Radiometric correction is applied to the mean-filtered data to obtain radiometrically corrected data. Image registration is performed on the radiometrically corrected data to obtain preprocessed data.

[0071] In one embodiment, the sub-aperture segmentation unit 403 is further configured to: Perform a Fourier transform on the preprocessed data to obtain Doppler data; The Doppler data is divided into frequency bands, and a frequency domain weighting function is constructed. The Doppler data is weighted in the frequency domain using a frequency domain weighting function, and then subjected to inverse Fourier transform to obtain sub-aperture segmentation data.

[0072] In one embodiment, the multi-phase detection unit 405 is further configured to: The sub-aperture data of multiple scenes with the same angle are expanded into column vectors according to pixel position and sorted according to time order to obtain the matrix data of multiple scenes with the same angle. Low-rank decomposition is performed on multiple scenes of sub-aperture matrix data with the same angle to obtain multiple sparse sequence data. Constant false alarm rate detection is performed on multiple sparse sequence data to obtain multi-temporal detection data.

[0073] In one embodiment, the multi-phase detection unit 405 is further configured to: Based on the low-rank sparse decomposition model, the sub-aperture matrix data of multiple scenes with the same angle is decomposed into a low-rank part and a sparse part. The low-rank part and the sparse part are solved by the preset low-rank constraint parameters and the preset sparsity constraint parameters to obtain multiple sparse term data. The sparse data from multiple scenes are rearranged column by column to obtain a sparse sequence of multiple scenes.

[0074] In one embodiment, the multi-phase detection unit 405 is further configured to: The pixel data to be detected in the sparse sequence data is set as the target area, and the background area is determined based on the target area; Calculate the probability density function of the background area, and obtain the detection threshold value based on the preset false alarm probability value and probability density function; Pixel detection data is obtained based on the pixel data to be detected and the detection threshold value; Pixel detection data are merged according to spatial location to obtain multi-temporal detection data.

[0075] In one embodiment, the single-phase detection unit 407 is further configured to: The sub-aperture data of different angles in a single scene are expanded into column vectors according to pixel positions and sorted according to the sub-aperture index order to obtain the sub-aperture matrix data of different angles in a single scene; Low-rank decomposition was performed on the sub-aperture matrix data of a single scene at different angles to obtain sparse sequence data of a single scene; Constant false alarm rate detection is performed on single-scene sparse sequence data to obtain single-phase detection data.

[0076] The same or similar parts among the above embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the device embodiments are basically similar to the method embodiments, so the description is relatively simple, and the relevant parts can be referred to the description of the method embodiments.

[0077] It should be noted that the embodiments of this application may involve the use of user data. In practical applications, user-specific personal data may be used in the scheme described herein within the scope permitted by applicable laws and regulations, provided that it complies with the applicable laws and regulations of the country (e.g., explicit consent from the user, actual notification to the user, explicit authorization from the user, etc.).

[0078] According to embodiments of this application, this application also provides a computer device and a computer-readable storage medium.

[0079] like Figure 5 The diagram shown is a block diagram of a computer device according to an embodiment of this application. The term "computer device" is intended to represent various forms of digital computers or mobile devices. The digital computer may include a desktop computer, a portable computer, a workbench, a personal digital assistant, a server, a mainframe computer, and other suitable computers. The mobile device may include a tablet computer, a smartphone, a wearable device, etc.

[0080] like Figure 5 As shown, device 500 includes a computing unit 501, a ROM 502, a RAM 503, a bus 504, and an input / output (I / O) interface 505. The computing unit 501, ROM 502, and RAM 503 are interconnected via the bus 504. The input / output (I / O) interface 505 is also connected to the bus 504.

[0081] The computing unit 501 can execute various processes in the method embodiments of this application according to computer instructions stored in the read-only memory (ROM) 502 or computer instructions loaded from the storage unit 508 into the random access memory (RAM) 503. The computing unit 501 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. The computing unit 501 can include, but is not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. In some embodiments, the methods provided in the embodiments of this application can be implemented as computer software programs, which are tangibly contained in a computer-readable storage medium, such as the storage unit 508.

[0082] RAM 503 can also store various programs and data required for the operation of device 500. Part or all of the computer program can be loaded and / or installed on device 500 via ROM 502 and / or communication unit 509.

[0083] The input unit 506, output unit 507, storage unit 508, and communication unit 509 in device 500 can be connected to I / O interface 505. The input unit 506 can be, for example, a keyboard, mouse, touchscreen, or microphone; the output unit 507 can be, for example, a display, speaker, or indicator light. Device 500 can exchange information and data with other devices through the communication unit 509.

[0084] It should be noted that the device may also include other components necessary for normal operation. It may also include only the components necessary for implementing the solution of this application, without necessarily including all the components shown in the figures.

[0085] Various implementations of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SOCs), payload programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof.

[0086] The computer instructions used to implement the methods of this application may be written in any combination of one or more programming languages. These computer instructions may be provided to the computing unit 501 such that when executed by the computing unit 501, such as a processor, the computer instructions cause the execution of the steps involved in the embodiments of the methods of this application.

[0087] The computer-readable storage medium provided in this application can be a tangible medium that can contain or store computer instructions for performing the steps involved in the method embodiments of this application. The computer-readable storage medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, and other forms of storage media.

[0088] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for detecting moving targets in the air, characterized in that, The method includes: Acquire single-view complex data, and preprocess the single-view complex data to obtain preprocessed data; The preprocessed data is segmented into sub-aperture data to obtain sub-aperture segmentation data, which includes sub-aperture data of multiple scenes at the same angle and sub-aperture data of a single scene at different angles. Low-rank decomposition and constant false alarm rate detection are performed on the multiple scenes of sub-aperture data at the same angle to obtain multi-temporal detection data; Low-rank decomposition and constant false alarm rate detection are performed on the sub-aperture data at different angles of the single scene to obtain single-phase detection data; The multi-temporal detection data and the single-temporal detection data are fused to obtain airborne moving target detection data.

2. The method according to claim 1, characterized in that, The preprocessing of the single-view complex data to obtain preprocessed data includes: Perform a logarithmic transformation on the single-view complex data to obtain logarithmically transformed data; The logarithmically transformed data is then subjected to mean filtering to obtain mean-filtered data. Radiometric correction is performed on the mean-filtered data to obtain radiometrically corrected data; Image registration is performed on the radiation correction data to obtain preprocessed data.

3. The method according to claim 2, characterized in that, The preprocessed data is segmented into sub-aperture data to obtain sub-aperture segmentation data, including: The preprocessed data is subjected to Fourier transform to obtain Doppler data; The Doppler data is divided into frequency bands, and a frequency domain weighting function is constructed. The Doppler data is weighted in the frequency domain according to the frequency domain weighting function, and then subjected to inverse Fourier transform to obtain sub-aperture segmentation data.

4. The method according to claim 1, characterized in that, The process of performing low-rank decomposition and constant false alarm rate detection on the multiple scenes of sub-aperture data at the same angle yields multi-temporal detection data, including: The multi-scene sub-aperture data with the same angle is expanded into column vectors according to pixel position and sorted according to time order to obtain multi-scene sub-aperture matrix data with the same angle. The multiple scenes of the same angle sub-aperture matrix data are subjected to low-rank decomposition to obtain multiple scenes of sparse sequence data. Constant false alarm rate detection is performed on the multi-scene sparse sequence data to obtain multi-temporal detection data.

5. The method according to claim 4, characterized in that, The low-rank decomposition of the multiple scenes of sub-aperture matrix data at the same angle yields multiple scenes of sparse sequence data, including: According to the low-rank sparse decomposition model, the multi-scene sub-aperture matrix data with the same angle is decomposed into a low-rank part and a sparse part. The low-rank part and the sparse part are solved according to the preset low-rank constraint parameters and the preset sparsity constraint parameters to obtain multi-scene sparse term data. The multi-scene sparse data items are rearranged column by column to obtain multi-scene sparse sequence data.

6. The method according to claim 5, characterized in that, The constant false alarm rate (CFAR) detection of the sparse sequence data to obtain multi-temporal detection data includes: The pixel data to be detected in the sparse sequence data is set as the target area, and the background area is determined based on the target area; Calculate the probability density function of the background region, and obtain the detection threshold value based on the preset false alarm probability value and the probability density function; Pixel detection data is obtained based on the pixel data to be detected and the detection threshold value; The pixel detection data are merged according to spatial location to obtain multi-temporal detection data.

7. The method according to claim 6, characterized in that, The process of performing low-rank decomposition and constant false alarm rate detection on the sub-aperture data at different angles of the single scene to obtain single-temporal detection data includes: The sub-aperture data of different angles in a single scene is expanded into column vectors according to pixel positions and sorted according to the sub-aperture index order to obtain the sub-aperture matrix data of different angles in a single scene. The sub-aperture matrix data of different angles of a single scene are decomposed into low-rank decomposition to obtain sparse sequence data of a single scene. Constant false alarm rate detection is performed on the single-scene sparse sequence data to obtain single-phase detection data.

8. An aerial moving target detection device, characterized in that, The device includes: The preprocessing unit is used to acquire single-view complex data, preprocess the single-view complex data, and obtain preprocessed data. The sub-aperture segmentation unit is used to perform sub-aperture segmentation on the preprocessed data to obtain sub-aperture segmentation data. The sub-aperture segmentation data includes sub-aperture data of multiple scenes at the same angle and sub-aperture data of a single scene at different angles. The multi-temporal detection unit is used to perform low-rank decomposition and constant false alarm rate detection on the multiple scenes of sub-aperture data at the same angle to obtain multi-temporal detection data. A single-phase detection unit is used to perform low-rank decomposition and constant false alarm rate detection on the single scene's sub-aperture data at different angles to obtain single-phase detection data. The fusion result unit is used to fuse the multi-temporal detection data and the single-temporal detection data to obtain airborne moving target detection data.

9. A computer device, comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores computer instructions executable by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the method of any one of claims 1-7.

10. A computer-readable storage medium storing computer instructions thereon, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1 to 7.

Images