A photovoltaic power station inspection method and system based on unmanned aerial vehicle intelligent scheduling

Abstract

The application discloses a photovoltaic power station inspection method and system based on unmanned aerial vehicle intelligent scheduling, belongs to the technical field of photovoltaic power station inspection, and through preprocessing and digitization processing of signals received by a radar on an unmanned aerial vehicle, setting a preset threshold, judging target existence or signal shadow based on the preset threshold, dynamically adjusting transmitting power to enhance shielding area detection and judgment, fusing multi-sensor data to complete target information in a signal shadow area, encoding the completed target information according to a radar point cloud format, merging and displaying the completed target information with original radar point clouds, developing a display interface, and directly displaying the interface by taking radar point cloud data as the bottom layer and superimposing a visual identification frame on the top layer.

Description

TECHNICAL FIELD

[0001] The application relates to photovoltaic power station inspection, in particular to a photovoltaic power station inspection method and system based on intelligent scheduling of unmanned aerial vehicles, and belongs to the technical field of photovoltaic power station inspection. BACKGROUND

[0002] In the prior art, millimeter wave radar (30GHz-300GHz) with a short wavelength is used in the inspection process, and although the diffraction ability is weak, the millimeter wave radar has a certain penetration to the sheltering objects such as vegetation, walls and component gaps, can detect areas (such as support rusting covered by shrubs, hidden cracks at the back of components) that cannot be reached by traditional visible light / infrared, and therefore has a great disadvantage in the detection of edge defects of mountain power station vegetation sheltering, distributed roof air conditioner outdoor unit shadow area and densely arranged components. Therefore, a photovoltaic power station inspection method and system based on intelligent scheduling of unmanned aerial vehicles are designed to solve the above problems. SUMMARY

[0003] The main purpose of the application is to provide a photovoltaic power station inspection method and system based on intelligent scheduling of unmanned aerial vehicles.

[0004] The purpose of the application can be achieved by adopting the following technical scheme:

[0005] A photovoltaic power station inspection method based on intelligent scheduling of unmanned aerial vehicles comprises the following steps:

[0006] Step one: pre-processing and digitizing the signals received by the radar;

[0007] Step two: setting a preset threshold, and judging the existence of a target or signal shadow based on the preset threshold;

[0008] Step three: dynamically adjusting the transmission power to enhance the detection and judgment of the sheltered area;

[0009] Step four: fusing multi-sensor data to complete the target information in the signal shadow area;

[0010] Step five: encoding the completed target information in the radar point cloud format, and merging and displaying the radar original point cloud;

[0011] Step six: developing a display interface, with the radar point cloud data at the bottom and the visual recognition frame superimposed at the top to realize intuitive display.

[0012] Preferably, in step two, the dynamic threshold is calculated by a constant false alarm rate algorithm

[0013] wherein, is the reference unit signal strength;

[0014] N is the number of reference units; ​

[0015] CF is an adjustment factor;

[0016] In step two, the digitized signal strength S is compared with a dynamic threshold If S > TH, the target is marked as present and the tracking algorithm is started, otherwise the signal shadow area is determined.

[0017] Preferably, when a signal shadow is detected , the transmit power is increased by a digital attenuator or a variable gain power amplifier to a preset high power level until the signal strength exceeds the threshold or reaches the hardware upper limit.

[0018] Preferably, in step four, the signal shadow area, combined with the target detection results of the visual sensor, is converted into physical coordinates in the radar coordinate system through spatial registration: , and a virtual point cloud is generated to supplement the radar data.

[0019] Preferably, the fusion of multiple sensors in step four specifically includes the following steps:

[0020] S11: Use GPS clock or time stamp interpolation method to ensure the time consistency of radar and visual data;

[0021] S12: Splice visual features and radar features into a multi-dimensional feature vector;

[0022] S13: Optimize the fused features through linear transformation or attention mechanism.

[0023] Preferably, the dynamic adjustment of the transmit power is based on the radar equation ;

[0024] When the blockage causes the path loss to increase , the required transmit power is calculated to maintain the received signal power constant.

[0025] Preferably, the judgment of signal shadow in step two adopts the following steps:

[0026] S21: After the radar antenna receives the electromagnetic wave signal reflected by the target, the weak signal is amplified by a low noise amplifier to improve the signal strength;

[0027] S22: Filter out noise and interference outside the frequency band through a band-pass filter to improve the signal-to-noise ratio and avoid false judgments caused by noise;

[0028] S23: Use an analog-to-digital converter to convert the preprocessed analog signal into a digital signal for subsequent digital signal processing;​​

[0029] S24: ADC samples the signal at a certain sampling frequency, and quantizes the signal amplitude into digital code;

[0030] S25: A fixed value is set in advance according to the noise level of the typical application scenario of the radar;

[0031] S26: A fixed threshold is determined by actually measuring the noise power ;

[0032] S27: The average noise power of N reference units around the detection unit is calculated ;

[0033] Wherein, is the signal strength of the reference unit, multiplied by the adjustment factor CF to obtain the dynamic threshold , which is used in the uniform clutter environment.

[0034] Preferably, in step four, target detection is also performed on the continuous N frames of visual images, and the presence state of each frame of target is recorded ;

[0035] Wherein, indicates that the target is detected in the i-th frame, otherwise;

[0036] The total number of votes of the target appearing in N frames is calculated ;

[0037] If ;

[0038] Wherein, is the voting threshold, so the decision is that the target exists Therefore, the visual decision is used as a supplement.

[0039] A photovoltaic power station inspection system based on unmanned aerial vehicle intelligent scheduling, comprising a signal processing module, including a low noise amplifier, a band pass filter and an analog to digital converter, for signal preprocessing and digitization;

[0040] Threshold calculation module, configured to calculate static threshold Or dynamic threshold , and realize real-time comparison of signal strength and threshold;

[0041] Transmit power adjustment module, integrated with digital attenuator or variable gain power amplifier, dynamically adjusts the transmit power according to the signal shadow detection result ;

[0042] Multi-sensor fusion module, including visual sensor, coordinate calibration unit and feature fusion algorithm unit, for time and space registration and feature level fusion of visual-radar data;

[0043] The display and completion module merges and displays the completed virtual point cloud with the radar original point cloud, and superimposes a visual recognition frame through a special interface.

[0044] Preferably, the feature fusion algorithm unit supports principal component analysis dimension reduction, compresses the fusion feature dimension to M dimensions, and inputs into a generative adversarial network or a variational autoencoder to generate a radar target representation of a signal shadow area;

[0045] The threshold calculation module supports mean value type CFAR and ordered statistics type CFAR, and the latter estimates the noise power by ordering the reference cell data and taking the median value.

[0046] The beneficial technical effects of the present application are:

[0047] The photovoltaic power station inspection method and system based on unmanned aerial vehicle intelligent scheduling provided by the present application has certain penetration for obstructions such as vegetation, walls, and component gaps, and can automatically enhance the signal strength according to the obstruction degree in combination with a dynamic transmission power adjustment technology, effectively covering the support of the mountain power station covered with shrubs, the shadow area of the distributed roof air conditioner outdoor unit, and other traditional blind areas, and realizing the "visualization" of defects by extracting target contour features through a visual sensor, generating a virtual point cloud through spatial registration, and merging with radar original data.

[0048] Through preprocessing means such as low-noise amplification and band-pass filtering, electromagnetic interference and background noise are effectively suppressed, the signal quality is significantly improved, even in a strong interference scene, a high signal-to-noise ratio can still be maintained, and the reliable detection of weak target signals is ensured.

[0049] When a signal shadow caused by an obstruction is detected, the system automatically increases the transmission power, enhances the signal penetration capability, increases the signal sampling frequency in the high-power mode, and further improves the detection capability of the weak reflection target behind the obstruction through multi-frame data accumulation analysis, so that even if the target reflection signal is weak, reliable identification can be realized through time series data enhancement.

[0050] Through high-precision clock synchronization and coordinate calibration technology, the radar and visual data are accurately aligned in time and space, a deep learning model is used to generate a virtual point cloud of the obstruction area, the virtual point cloud is merged with the radar original point cloud in real time, and a visual recognition frame is superimposed. BRIEF DESCRIPTION OF DRAWINGS

[0051] Figure 1 A flowchart of a preferred embodiment of the photovoltaic power station inspection method and system based on unmanned aerial vehicle intelligent scheduling according to the present application. DETAILED DESCRIPTION

[0052] In order to make the technical scheme of the present application more clear and explicit to those skilled in the art, the present application will be further described in detail below in conjunction with embodiments and drawings, but the embodiments of the present application are not limited thereto.

[0053] The unmanned aerial vehicle is provided with a camera and a millimeter wave radar, and in the unmanned aerial vehicle inspection system, the millimeter wave has a short wavelength (1 millimeter-10 millimeter corresponding to 30GHz-300GHz), weak diffraction ability, and is difficult to penetrate or bypass obstacles (such as building corners, walls, and dense vegetation).

[0054] When the target is behind the shelter, the shelter area forms a "signal shadow", which causes the radar to be unable to receive the echo reflected by the target, and thus the target is missing. In the perimeter unmanned aerial vehicle inspection, a pedestrian may hide behind a wall or a tree, and the radar may not be able to detect the pedestrian.

[0055] Therefore, the problem of "signal shadow" is formed, and the present application is designed to compare the received signal strength with a preset threshold value by the radar.

[0056] When the echo signal strength is significantly lower than the normal level and lasts for a period of time, it may indicate signal attenuation or loss caused by the shelter, and the specific scheme is as follows:

[0057] After the radar antenna receives the electromagnetic wave signal reflected by the target, a low-noise amplifier is used to amplify the weak signal and improve the signal strength.

[0058] The noise and interference (such as electromagnetic noise and clutter) outside the frequency band are filtered out by a band-pass filter, the signal-to-noise ratio (SNR) of the signal is improved, and false judgments caused by noise are avoided.

[0059] An analog-to-digital converter is used to convert the preprocessed analog signal into a digital signal, which is convenient for subsequent digital signal processing.

[0060] The ADC samples the signal at a certain sampling frequency (such as millions of samples per second), and quantizes the signal amplitude to digital code.

[0061] A fixed value is set in advance according to the noise level of the typical application scenario of the radar.

[0062] In a laboratory environment, a fixed threshold value is determined by actually measuring the noise power .

[0063] The average noise power of N reference units around the detection unit is calculated ;

[0064] The signal strength of the reference unit is multiplied by an adjustment factor CF to obtain a dynamic threshold , which is used in a uniform clutter environment.

[0065] The reference cell data is sorted, and the median value is selected as the noise estimation, which is suitable for non-uniform clutter environment and avoids threshold deviation caused by strong clutter interference.

[0066] Only the comparator circuit is used to compare the digitized signal amplitude with the threshold value in real time. If the signal strength S exceeds the dynamic threshold value , the comparator outputs a high level or triggers an interrupt, directly indicating the target detection result, which is suitable for scenarios with high real-time requirements (such as missile guidance radar).

[0067] In the digital signal processing unit (such as FPGA, DSP), the algorithm is compared point by point.

[0068] Verilog code is written in FPGA, or C language is used in DSP to realize loop judgment. If , the target exists, which is convenient for flexible adjustment of processing logic.

[0069] If , start the target tracking algorithm (such as Kalman filter) to continuously monitor the target motion state;

[0070] If , continue to receive new echo signals and repeat the above process.

[0071] Mean CFAR threshold calculation: ;

[0072] Where N is the number of reference cells, is the signal strength of the ith reference cell, and CF is the adjustment factor (set according to scene experience).

[0073] Signal judgment logic: target exists ;

[0074] Where S is the signal strength of the current detection cell, is the preset threshold (static or dynamic CFAR threshold);

[0075] Through the above technical solutions and formulas, the radar can accurately compare the signal strength with the threshold value in different environments to determine whether the target exists.

[0076] Further, the present application integrates a digital attenuator or a variable gain power amplifier in the radar transmitting module, and adjusts the transmitting power in real time through software instructions .

[0077] Low power mode (normal scene): , reduce power consumption;

[0078] High power mode (occlusion scene): , enhance signal penetration ability.

[0079] When radar determines that there is signal shadow in a certain area through signal strength comparison , the following process is triggered:

[0080] Send instructions to the transmitting module to increase to a preset high power gear;

[0081] Continuously monitor the signal strength in this area, if maintain the current power, otherwise further increase until the threshold or reach the hardware upper limit.

[0082] In combination with the "multi-frame signal accumulation detection" logic, the number of signal sampling frames is increased (such as from 10 frames / second to 20 frames / second) in high power mode, and the signal-to-noise ratio is improved through time dimension accumulation.

[0083] According to the radar equation, the relationship between the received signal power and the transmitted power is: ;

[0084] is the transmit / receive antenna gain;

[0085] is the target radar cross section;

[0086] R is the target distance;

[0087] L is the path loss caused by shielding.

[0088] Suppose the original transmitted power is , the shielding causes the path loss to increase (such as vegetation shielding ), in order to keep the received signal power unchanged, the transmitted power needs to be increased to ;

[0089] According to the dynamic threshold formula , when the transmitted power increases, the reference cell average signal strength also increases, so:

[0090] If the detection sensitivity remains unchanged (the false alarm rate is constant), the CF needs to be maintained unchanged, at this time increases linearly with ;

[0091] If you want to enhance the detection probability, you can appropriately reduce the CF in high power mode (such as from the default value of 3 to 2.5), so that the threshold growth is less than the signal strength increase, thereby improving the probability of .

[0092] Table 1 Key parameters and implementation effects

[0093]

[0094] In combination with the "multi-sensor fusion" technology, if the radar still cannot detect the target in the high-power mode (such as extreme occlusion), the visual sensor is triggered to complete (such as the camera recognizing the target behind the wall, and the radar data is fused to display a virtual point cloud).

[0095] In combination with the "multi-sensor fusion" technology, the following solutions are included:

[0096] Using a hardware clock (such as a GPS synchronous clock) or a software algorithm (such as a timestamp-based interpolation method), the time instants of data collection of the millimeter wave radar and the visual sensor (camera, thermal imager, etc.) are ensured to be consistent, avoiding target position deviation caused by time difference.

[0097] Through a calibration board or known scene features, the coordinate transformation relationship (such as a rotation matrix and a translation vector) between the visual sensor and the millimeter wave radar is established, the visual target coordinates are mapped to the radar coordinate system, and spatial unification is achieved.

[0098] A deep learning model is used to detect targets in camera images, extract target features such as contours, colors, and textures, etc. For example, the visual contour features of a pedestrian behind a wall are detected.

[0099] The millimeter wave radar data is preprocessed (such as denoising and filtering), and distance and speed information is extracted.

[0100] For signal shadow areas, the spatial positions of signal missing areas are marked.

[0101] The visual features (such as target contour vectors) and radar features (such as distance information) are combined into a multi-dimensional feature vector. The visual contour coordinates of a pedestrian are combined with the surrounding environment distance information measured by the radar to form complementary features. The following technical solutions are used:

[0102] Through a GPS clock or a special synchronization module (such as IEEE1588), the sampling time error of the radar and the camera is ensured to be less than 1ms, avoiding feature mismatch caused by time misalignment.

[0103] For asynchronously collected data, linear interpolation or nearest neighbor interpolation method is used based on timestamps for alignment, and the formula is:

[0104] ;

[0105] Wherein, , is the adjacent frame timestamp, is the interpolated feature value.

[0106] Solve the extrinsic parameters (rotation matrix R, translation vector t) of camera and radar using Zhang's calibration method or 3D calibration board.

[0107] Convert the pixel coordinates of visual targets to three-dimensional coordinates in the radar coordinate system through the camera intrinsic parameter K.

[0108] ;

[0109] where, is the inverse matrix of the camera intrinsic parameter, used to convert pixel coordinates to normalized camera coordinates.

[0110] Use a deep learning model (such as Mask R-CNN) to segment the target and extract the contour key point coordinates (unit: pixels), and convert them to physical coordinates in the radar coordinate system through spatial registration .

[0111] Extract the one-hot encoding of the target category (such as "pedestrian" "vehicle") through CNN (such as ResNet) (C is the number of categories).

[0112] Input vector: (N is the number of contour key points);

[0113] Extract the distance d, azimuth angle , and elevation angle of the target from the radar point cloud or RD map.

[0114] For obstacles within the radar scanning range, extract their distance distribution characteristics (such as the distance of the nearest obstacle , average distance ).

[0115] Output vector: ;

[0116] Concatenate the visual features and radar features by dimension to form a complementary feature vector: ;

[0117] Eliminate the scale difference between features through linear transformation or nonlinear mapping, formula: ;

[0118] , : learnable weight matrix;

[0119] is the activation function (such as ReLU);

[0120] Suitable for scenarios where the fused features are used as input to a neural network (e.g. a Transformer for object detection).

[0121] Introduce attention weight to adaptively adjust feature importance:

[0122] ;

[0123] ;

[0124] is the attention weight vector, learned through training.

[0125] is the element-wise product, used to enhance key modality features (e.g. visual dominant in occluded scenarios, radar dominant in non-occluded scenarios).

[0126] Standardize features with different dimensions (e.g. pixel coordinates and distances): ;

[0127] , is the mean and standard deviation of the feature, obtained through training data statistics.

[0128] If the dimension of the fused feature is too high (e.g. >100 dimensions), compress it to M dimensions through PCA: ;

[0129] U is the projection matrix composed of the first M principal components of the feature covariance matrix of the training data.

[0130] Example 1: Contour coordinates of a pedestrian behind a wall (estimated through camera perspective transformation);

[0131] The signal strength of the wall region is lower than the threshold , marked as signal shadow.

[0132] The visual contour coordinates are converted to physical positions (X, Y) in the radar coordinate system through spatial registration;

[0133] The radar provides environmental distance information (e.g. wall distance ) for this position;

[0134] The fusion vector is , with category encoding "pedestrian";

[0135] Input the classifier to determine whether there is an occluded target at this position (e.g. through SVM or neural network output probability value).

[0136] The GAN or VAE is trained by the above scheme to generate the radar target representation of the signal shadow area after the feature-level fusion, inputting the visual features and the non-shadow area data of the radar, for example, the model learns the feature correlation of the "pedestrian behind the wall" in the visual and radar normal areas to generate the corresponding radar point cloud features.

[0137] The visual sensor independently identifies the target (such as determining that there is a pedestrian behind the wall), the radar determines that the signal in the area is missing, the visual decision is supplemented by a rule engine or a voting mechanism, and the corresponding target information (such as a virtual point cloud) is directly added to the radar data.

[0138] Target detection is performed on continuous N frames of visual images, and the presence state of each frame of target is recorded the target is detected in the i-th frame, otherwise.

[0139] The total number of votes of the target appearing in N frames is calculated .

[0140] If ( is the voting threshold), the decision is that the target exists, and the visual decision is supplemented.

[0141] The completed target information (such as a virtual point cloud) is encoded according to the radar point cloud format and merged with the original radar point cloud for display. For example, in the radar monitoring interface, the pedestrian behind the wall is presented as a point cloud of a specific color, which is displayed uniformly with the point cloud directly detected by the radar.

[0142] A special display interface is developed, the bottom layer is radar point cloud data, and the upper layer is superimposed with a visual recognition box (such as a BoundingBox). In the signal shadow area, the target is labeled by the visual recognition box, and a virtual point cloud is dynamically generated in the missing area of the radar point cloud to realize intuitive fusion display.

[0143] The present application further provides an embodiment, but the protection scope of the present application is not limited thereto, any skilled person in the art can make equivalent replacement or change according to the technical scheme and concept of the present application within the scope disclosed by the present application, which belongs to the protection scope of the present application.​

Claims

1. A photovoltaic power plant inspection method based on unmanned aerial vehicle (UAV) intelligent scheduling, characterized in that: Includes the following steps: Step 1: Preprocess and digitize the signals received by the radar; Step 2: Set a preset threshold, and determine whether the target exists or the signal is shadowed based on the preset threshold; Step 3: Dynamically adjust the transmission power to enhance the detection and judgment of obstructed areas; Step 4: Fuse multi-sensor data to complete target information in the signal shadow area; Step 5: Encode the completed target information according to the radar point cloud format and merge it with the original radar point cloud for display; Step Six: Develop the display interface, with radar point cloud data at the bottom layer and visual recognition boxes overlaid on the top layer to achieve intuitive display; The dynamic adjustment of the transmit power is based on radar equations. When occlusion causes an increase in path loss ΔL, through P t2 =P t1 ·10 ΔL / 10 Calculate the required transmit power to maintain the received signal power P. r Constant; G t G r For transmit / receive antenna gain; σ is the target radar cross-section; R is the target distance; L represents the path loss caused by occlusion; P r For received signal power; P t This refers to the transmission power. P t1 This is the original transmission power.

2. The photovoltaic power station inspection method based on UAV intelligent scheduling according to claim 1, characterized in that: In step two, the dynamic threshold is calculated using the constant false alarm rate algorithm. Where xi is the signal strength of the reference cell; N is the number of reference units; CF is the adjustment factor; In step two, the digitized signal strength S is compared with the dynamic threshold T. CFAR In comparison, if S>T CFAR If the target is marked as present, the tracking algorithm is started; otherwise, it is determined to be a signal shadow area.

3. The photovoltaic power station inspection method based on UAV intelligent scheduling according to claim 2, characterized in that: When a signal shadow is detected, S≤T CFAR At the same time, the transmit power P is increased by a digitally controlled attenuator or a variable gain power amplifier. t Upgrade to the preset high power level until the signal strength exceeds the threshold or reaches the hardware limit.

4. The photovoltaic power station inspection method based on UAV intelligent scheduling according to claim 1, characterized in that: In step four, within the signal shadow area, combined with the target detection results from the visual sensor, the visual coordinates (u, v) are converted to physical coordinates (X, Y, Z) in the radar coordinate system through spatial registration. And generate a virtual point cloud to supplement the radar data; Where R is the rotation matrix; t is the translation vector; K is the camera's intrinsic parameter K.

5. The photovoltaic power station inspection method based on UAV intelligent scheduling according to claim 1, characterized in that: Step four, which involves fusing multi-sensor completion, specifically includes the following steps: S11: Use GPS clock or timestamp interpolation to ensure time consistency between radar and visual data; S12: Concatenate visual features and radar features into a multi-dimensional feature vector; S13: Optimize fused features through linear transformation or attention mechanism.

6. The photovoltaic power station inspection method based on UAV intelligent scheduling according to claim 1, characterized in that: The determination of signal shadows in step two is carried out using the following steps: S21: After the radar antenna receives the electromagnetic wave signal reflected by the target, it uses a low-noise amplifier to amplify the weak signal and improve the signal strength. S22: By using a bandpass filter to filter out noise and interference outside the frequency band, the signal-to-noise ratio of the signal is improved, and misjudgment caused by noise is avoided; S23: Use an analog-to-digital converter to convert the preprocessed analog signal into a digital signal, which facilitates subsequent digital signal processing; S24: The ADC samples the signal at a certain sampling frequency and quantizes the signal amplitude into digital code; S25: A fixed value is preset based on the noise level of typical radar application scenarios; S26: Determine a fixed threshold T by measuring the noise power. static ; S27: Calculate the average noise power of N reference units surrounding the detection unit. Where, x i The dynamic threshold is obtained by multiplying the reference cell signal strength by the adjustment factor CF. For use in uniform clutter environments.

7. The photovoltaic power station inspection method based on UAV intelligent scheduling according to claim 1, characterized in that: Step four also includes performing target detection on N consecutive frames of visual images and recording the presence state v of the target in each frame. i ; Among them, v i =1 indicates that a target was detected in the i-th frame, v i =0, conversely; Calculate the total number of votes for the target's appearance within N frames. If V≥V th ; Among them, V th If the voting threshold is set, then the decision is based on the existence of the objective, and therefore visual decision-making is used as a supplement.

8. A photovoltaic power station inspection system based on UAV intelligent scheduling, based on the photovoltaic power station inspection method based on UAV intelligent scheduling according to any one of claims 1-7, characterized in that: include: The signal processing module, which includes a low-noise amplifier, a bandpass filter, and an analog-to-digital converter, is used for signal preprocessing and digitization. The threshold calculation module is configured to calculate the static threshold T. static Or dynamic threshold T CFAR And achieve real-time comparison between signal strength and threshold; The transmit power adjustment module integrates a digitally controlled attenuator or a variable gain power amplifier to dynamically adjust the transmit power P based on the signal shadow detection results. t ; The multi-sensor fusion module includes a visual sensor, a coordinate calibration unit, and a feature fusion algorithm unit, which are used for spatiotemporal registration and feature-level fusion of visual-radar data; The display and completion module merges the completed virtual point cloud with the original radar point cloud for display, and overlays a visual recognition box through a dedicated interface.

9. A photovoltaic power station inspection system based on UAV intelligent scheduling according to claim 8, characterized in that: The feature fusion algorithm unit supports principal component analysis for dimensionality reduction, compressing the fused feature dimension to M dimensions, and inputting it into a generative adversarial network or variational autoencoder to generate a radar target representation of the signal shadow region; The threshold calculation module supports mean-based CFAR and ordered statistical CFAR, the latter estimating noise power by sorting the reference cell data and taking the median.

Images