A background registration and component reverse lookup radar foreign matter detection method

Abstract

The application discloses a kind of background registration and component reverse search radar foreign matter detection methods, comprising the following steps: obtaining the first background radar data under the state of no foreign matter and the first target radar data of the target scene to be detected, generate first background distance-angle amplitude chart and first target distance-angle amplitude chart;Get the first enhanced background cumulative chart after registration;Get the first background suppression difference chart;Get multiple candidate target components;Calculate component-level background residual characteristics, and filter background residual candidates according to component-level background residual characteristics, output foreign matter detection results.The application introduces background structure self-registration and component-level background reverse search mechanism, effectively solves the small-angle misplacement problem between target scene and background chart caused by starting angle deviation of rotating scanning radar, acquisition time difference, motion speed fluctuation or radio frequency link state change in actual application.

Description

Technical Field

[0001] This invention relates to the technical fields of radar signal processing, millimeter-wave radar foreign object detection, airport runway and road foreign object detection, and small target detection under strong static background, and particularly to a background registration and component reverse lookup radar foreign object detection method. Background Technology

[0002] When rotating scanning radar performs foreign object detection, it typically needs to identify small targets from a strong static background such as runways, roads, or other fixed scenes. Existing methods often employ background subtraction, clutter maps, CFAR detection, and connected component analysis. Among these, CFAR (Constant False Alarm Rate) detection is a method that uses reference cells around the detected cell to estimate the local background level and adaptively sets the detection threshold accordingly to control the false alarm probability. Clutter maps are typically background intensity maps formed from targetless or historical scan data. They describe stable reflection, clutter, or strong scattering structures in the scene and serve as a basis for background suppression in subsequent detection.

[0003] In recent years, technologies related to foreign object detection (FOD) in specific scenarios such as airport runways have been continuously evolving. For example, patent CN113064162B, "Detection Method and Device for Airport Runway FOD Radar System," proposes a detection scheme based on region of interest (ROI) extraction. This involves continuously scanning a target area free of FOD to establish initial background data, utilizing the discontinuous characteristics of the target area's edges and the geometric relationship of the radar equipment layout to remove redundant edge lines, extracting the ROI, and then using a clutter map built based on the initial background data to perform constant false alarm rate (CFAR) detection on real-time data. This type of method improves the detection probability in multi-target, long-range environments to a certain extent, ensuring the real-time and continuous requirements of airport FOD detection on runways.

[0004] When the background data and target scene data are strictly aligned and the background changes are minimal, the above method can achieve a certain level of target extraction effectiveness. By constructing a stable clutter map and combining it with CFAR detection, the system can effectively detect anomalies that deviate from the statistical characteristics of the background while controlling the false alarm rate. Furthermore, connected component analysis can be used to aggregate spatially adjacent detection points into candidate target components, improving the structural rationality of the detection results.

[0005] However, in practical rotating scanning radar systems, it is difficult to guarantee complete consistency between each scan and the modeled background. Due to factors such as initial angle deviations in the mechanical turntable, slight differences in radar acquisition timing, fluctuations in scanning speed, and changes in the RF link status due to temperature drift and aging, a non-negligible small angular misalignment occurs between the target scan image and the pre-acquired background image. When directly using the background image or clutter image for subtraction, strong background edges and sidelobe regions are prone to residual responses due to this misalignment, which may be misidentified as targets after CFAR detection. For small, distant foreign objects, the response amplitude is low; increasing the detection threshold easily leads to missed detections, while decreasing the threshold introduces more background residue false alarms. Therefore, there is an urgent need for a radar foreign object detection method that combines background structure self-registration, background suppression differential, and candidate component-level verification to reduce false targets caused by background misalignment and improve the stability of small, weak foreign object detection against strong static backgrounds. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art and provide a background registration and component reverse radar foreign object detection method to solve the problem of background residual false alarms caused by small angular misalignment between background and target data, and improve the reliability and stability of detecting small foreign objects in strong static scenes.

[0007] The objective of this invention is achieved through the following technical solution: A background registration and component reverse radar foreign object detection method includes the following steps: Acquire the first background radar data in the absence of foreign objects and the first target radar data of the scene to be detected, and perform basic preprocessing on the first background radar data and the first target radar data respectively to generate the first background range-angle amplitude map and the first target range-angle amplitude map; The first background distance-angle amplitude map is subjected to background enhancement processing to identify strong scattering regions and locally expand them in the distance and angle directions to obtain a first enhanced background map; the first target distance-angle amplitude map is cumulatively processed along the angle direction to obtain a first target cumulative map, and the first enhanced background map is cumulatively processed along the angle direction to obtain a first enhanced background cumulative map; The angle offset is estimated based on the background structure features of the first target cumulative image and the first enhanced background cumulative image, and the angle is registered based on the angle offset to obtain the registered first enhanced background cumulative image. Based on the first target cumulative map and the registered first enhanced background cumulative map, a positive difference is calculated and a background suppression weight is generated, thereby obtaining the first background suppression difference map; The first background suppression difference map is subjected to constant false alarm rate detection and candidate map post-processing to obtain multiple candidate target components; The first target cumulative image, the registered first enhanced background cumulative image, and the positive difference are reviewed back on the candidate target components to calculate the component-level background residue features. The background residue candidates are then filtered based on the component-level background residue features, and the foreign object detection results are output.

[0008] This invention proposes a background registration and component-based reverse-lookup radar foreign object detection method, aiming to solve the problem of misalignment between background and target images caused by small deviations in scanning angle in practical radar systems, resulting in a large number of false alarms. The method first acquires background radar data in a foreign object-free state and simultaneously acquires target radar data of the scene to be detected. After basic preprocessing, corresponding range-angle amplitude maps, i.e., two-dimensional radar images, are generated. To enhance the stability of the background structure, strong scattering regions (such as runway edges and fixed structures) in the background image are identified and locally expanded in the range and angle directions to form an "enhanced background image," making it more suitable for subsequent registration requirements. Next, the target image and the enhanced background image are cumulatively processed on the same range-angle grid to obtain a target cumulative image and an enhanced background cumulative image. This processing is used to enhance background structures and target responses that stably appear across multiple scan cycles or slow time dimensions, while reducing the impact of random cusps in a single frame on subsequent registration and detection results. Using these two cumulative images, the angular offset between them is estimated by comparing the background structure (such as peak position and contour shape). Based on this, the background cumulative image is angularly registered to align it spatially with the target image. Then, a background suppression difference image is generated using a positive difference method (retaining only the portion of the target image stronger than the background) combined with adaptive weights. This effectively suppresses known static background and highlights potential foreign object signals. Subsequently, constant false alarm rate (CFAR) detection and connected component analysis are performed on this difference image to initially screen out several candidate target components. Crucially, this method does not stop at point-level detection but, for each candidate component, backtracks to the original target cumulative image, the registered background cumulative image, and the difference results to calculate the background residual features (such as background matching degree) within the component region, thereby determining whether it is caused by background misalignment. If it is determined to be background residual, it is removed; otherwise, it is retained as a real foreign object. This two-tiered mechanism of "first coarse inspection, then fine screening," especially with the introduction of component-level reverse verification, significantly reduces false alarms caused by mechanical vibration, temperature drift, or asynchronous scanning, and improves the reliability and stability of detecting small foreign objects under strong static conditions.

[0009] As a preferred embodiment, the background enhancement processing includes: identifying strong scattering regions in the first background distance-angle amplitude map, locally expanding the strong scattering regions in both the distance and angle directions, and fusing the expansion results with the first background distance-angle amplitude map to expand the suppression range of stable strong background structures, enabling subsequent differential processing to cover slight drift, sidelobes, and edge remnants near the strong background; the first enhanced background map can be represented as: ;in, This represents the first enhanced background image. This represents the first background distance angle amplitude map. This represents a local enhancement term generated based on the background region with strong scattering.

[0010] As a preferred embodiment, the estimation of the angle offset includes: extracting the outline of a first target background structure from the first target cumulative image, extracting the outline of a first enhanced background structure from the first enhanced background cumulative image, calculating the similarity between the first target background structure outline and the first enhanced background structure outline within a preset candidate angle offset range, and determining the candidate angle offset with the highest similarity as the angle offset; the angle offset estimation can be expressed as: ;in, This indicates the operation of finding the maximum value. Indicates the candidate angle offset. Represents the set of candidate angle offsets. This represents the background structure outline extracted from the first target cumulative image. This represents the background structure contour extracted from the first enhanced background cumulative image. Sim(·) represents the similarity calculation function, and Shift(·) represents the angle translation function. This represents the estimated optimal angle offset.

[0011] As a preferred embodiment, the generation of the first background suppression difference map includes: calculating a first positive difference between the first target cumulative map and the registered first enhanced background cumulative map: ;in, This represents the first positive difference. This represents the cumulative graph of the first target. This represents the first enhanced background cumulative image after registration. This represents the function that takes the maximum value. Indicates distance index, Indicates the angle index.

[0012] As a preferred embodiment, the generation of the first background suppression difference map further includes: calculating the background percentage. ;in, Indicates the proportion of the background. To avoid dividing by zero for extremely small positive numbers.

[0013] The generation of the first background suppression difference map further includes: generating background suppression weights based on the background proportion. ;in, This represents the background suppression weight. This represents a clipping function that restricts values ​​to a specified range. and These are the preset lower and upper weight limits, respectively.

[0014] As a preferred embodiment, the generation of the first background suppression difference map further includes: weighting the first positive difference using the background suppression weights to obtain the first background suppression difference map. ;in, This represents the first background suppression difference map.

[0015] As a preferred embodiment, before performing constant false alarm rate (CFAR) detection on the first background suppression difference map, the method further includes robust normalization processing of the first background suppression difference map by angle column, the expression of which is: ;in, The graph represents the robustly normalized result. Median(·) denotes the median function, and MAD(·) denotes the absolute median difference function. This represents the values ​​corresponding to all distance cells in the a-th angle column of the background suppression difference plot.

[0016] As a preferred method, the component-level background reverse lookup includes: calculating the target response features within the component, the background response features within the component, the newly added differential response features within the component, and the component morphological features for each candidate target component; when the candidate target component is also significant in the background and the newly added response is insufficient, the candidate target component is filtered as a background residual candidate; when the candidate target component has a newly added response relative to the background and meets the morphological requirements, the candidate target component is retained as a foreign object target candidate.

[0017] As a preferred approach, the features used for determination in the component-level background reverse lookup include at least: a first ratio of the cumulative background intensity within the component to the cumulative target intensity. ; And the second ratio of the positive differential intensity within the component to the cumulative background intensity: ;in, Let represent a candidate target component, and Mean(·) represent a function to calculate the mean, indicating that in the candidate target component The value of the registered enhanced background cumulative map within the range, Indicated in the candidate target component The cumulative value of the target within the range, Indicated in the candidate target component The values ​​of the positive difference plot within the range.

[0018] As a preferred approach, the first target cumulative image and the first enhanced background cumulative image are subjected to intensity compression or robust normalization to reduce the influence of local strong points on structural matching, and then the first target background structure contour and the first enhanced background structure contour are extracted.

[0019] The present invention has at least the following beneficial effects: This invention effectively solves the problem of small-angle misalignment between the target scene and the background image caused by turntable starting angle deviation, acquisition time difference, motion speed fluctuation, or RF link status change in practical applications of rotating scanning radar by introducing background structure self-registration and component-level background back-checking mechanisms. Specifically, by matching the background structure features of the first target cumulative image and the first enhanced background cumulative image, the angular offset is estimated and corrected, achieving accurate alignment between background data and target data, significantly reducing the residual response caused by misalignment in strong background edges and sidelobe regions. Simultaneously, after differential detection, the original cumulative image and differential image are further checked back on a component-by-component basis to calculate component-level background residual features, and background residual candidates are filtered accordingly. This significantly suppresses false alarms caused by background misalignment without sacrificing the detection rate of weak foreign objects. Compared with existing methods that rely solely on background subtraction, clutter maps, and CFAR detection, this invention significantly improves the reliability and stability of detection results while maintaining high sensitivity to weak foreign objects at long distances. Attached Figure Description

[0020] To clarify the technical details of the embodiments of the present invention, the relevant drawings will be briefly described below. It should be emphasized that the drawings shown are only for illustrative purposes to present some embodiments of the present invention and do not constitute a limitation on the scope of protection of the invention. Those skilled in the art can reasonably deduce or derive other related illustrative content based on these drawings without creative effort.

[0021] Figure 1 A flowchart of a background registration and component reverse radar foreign object detection method; Figure 2 The flowchart shows the process of feature determination and filtering for candidate target components. Detailed Implementation

[0022] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings, but the scope of protection of the present invention is not limited to the following description.

[0023] It should be noted that the specific details provided below are intended to help readers gain a more comprehensive understanding of the example embodiments, but are not all the conditions necessary to implement these embodiments. Those skilled in the art should understand that the example embodiments can still be effectively implemented even with these specific details omitted. For example, the system may be illustrated using a block diagram to avoid excessive technical details interfering with the understanding of the overall architecture and process. Furthermore, in some cases, to highlight the core ideas of the example and maintain clarity of expression, some well-known processing steps, conventional structures, or mature technologies may be intentionally omitted, as they have no substantial impact on the implementation of this solution.

[0024] like Figure 1 As shown, a background registration and component reverse lookup radar foreign object detection method includes the following steps: Acquire the first background radar data in the absence of foreign objects and the first target radar data of the scene to be detected, and perform basic preprocessing on the first background radar data and the first target radar data respectively to generate the first background range-angle amplitude map and the first target range-angle amplitude map; The first background distance-angle amplitude map is subjected to background enhancement processing to identify strong scattering regions and locally expand them in the distance and angle directions to obtain a first enhanced background map; the first target distance-angle amplitude map is cumulatively processed along the angle direction to obtain a first target cumulative map, and the first enhanced background map is cumulatively processed along the angle direction to obtain a first enhanced background cumulative map; The angle offset is estimated based on the background structure features of the first target cumulative image and the first enhanced background cumulative image, and the angle is registered based on the angle offset to obtain the registered first enhanced background cumulative image. Based on the first target cumulative map and the registered first enhanced background cumulative map, a positive difference is calculated and a background suppression weight is generated, thereby obtaining the first background suppression difference map; The first background suppression difference map is subjected to constant false alarm rate detection and candidate map post-processing to obtain multiple candidate target components; The first target cumulative image, the registered first enhanced background cumulative image, and the positive difference are reviewed back on the candidate target components to calculate the component-level background residue features. The background residue candidates are then filtered based on the component-level background residue features, and the foreign object detection results are output.

[0025] This invention acquires radar data of a background-free target and the target object, generating corresponding range-angle amplitude maps. The background map undergoes strong scattering region identification and local expansion to enhance its structure. Accumulation is performed on the same range-angle grid to obtain a target cumulative map and an enhanced background cumulative map. Background structure features from the two cumulative maps are used to estimate and correct angular offsets, achieving background registration. Subsequently, a positive difference is calculated based on the registered background and target cumulative maps, combined with background suppression weights, to generate a differential map that suppresses static background. Constant false alarm rate (CFAR) detection and post-processing are performed on this differential map to obtain candidate target components. Finally, for each candidate component, the original cumulative map and differential results are reviewed to calculate its component-level background residual features. Based on these features, false candidates caused by background misalignment are eliminated, and the true foreign object detection result is output.

[0026] In one embodiment, basic preprocessing includes raw data shape normalization, removal of DC components, windowing, frequency domain transformation, selection of effective frequency bands, distance axis mapping, and generation of amplitude or energy maps.

[0027] Basic preprocessing is a crucial first step in radar or sensor signal processing, aiming to transform the raw acquired data into a clearer and more easily analyzable form. The entire process typically begins with raw data shape normalization, adjusting data from different channels or frames to a uniform dimension and format to ensure consistency in subsequent processing. Next, DC component removal is performed, primarily to eliminate fixed offsets introduced by the sensor or hardware (such as strong background signals near zero frequency), preventing them from masking genuine, weak target echoes. Following this, windowing is applied to the signal, multiplying it in the time domain by a smooth window function (such as a Hanning window) to reduce spectral leakage and improve the accuracy of subsequent frequency domain analysis. After windowing, the system performs a frequency domain transformation (usually a Fast Fourier Transform, FFT) to convert the time-domain signal to the frequency domain, revealing the different frequency components contained within the signal. Since not all frequencies carry useful information, the next step is to select effective frequency bands, filtering out noise-dominated or irrelevant bands based on the application scenario, retaining the frequency range that may contain target characteristics. Next, distance axis mapping is performed using frequency domain information, converting frequency values ​​into corresponding distance units according to physical relationships, forming distance-dimensional data with practical spatial meaning. Finally, by calculating the amplitude or energy of each distance-time (or distance-angle) unit, amplitude or energy maps are generated, serving as intuitive inputs for subsequent target detection, clustering, or feature extraction. This series of interconnected steps gradually refines the original chaotic signal into a clearly structured and physically meaningful graphical representation, laying a solid foundation for advanced algorithms.

[0028] In one embodiment, constant false alarm rate (CFAR) detection is performed for each angle column along the distance direction, or two-dimensional CFAR detection or other CFAR detection methods may be used as needed.

[0029] Constant False Alarm Rate (CFAR) detection is an adaptive threshold setting method that detects real targets as reliably as possible while maintaining a constant false alarm probability. Instead of using a fixed threshold, it dynamically adjusts the detection threshold based on the background noise or clutter intensity of the "neighborhood area" surrounding the target cell. In radar or sensing systems, one-dimensional CFAR detection is typically performed separately for each angle column along the range direction. That is, at a fixed angle, each range cell is traversed sequentially, and several adjacent cells before and after the current cell are selected as a reference window. The average noise level of the current local environment is estimated, and then multiplied by an empirical coefficient to obtain an adaptive threshold. If the signal strength of the central cell exceeds this threshold, a target is considered to be present. This method effectively addresses the problem of varying clutter intensity at different ranges. Of course, if the scene is more complex (e.g., dense targets, non-uniform clutter), two-dimensional CFAR detection can also be used, which considers neighborhood information in both range and angle dimensions to estimate the background, thus achieving more robust detection performance. Furthermore, various CFAR variants (such as CA-CFAR, OS-CFAR, etc.) can be selected according to actual needs to balance detection sensitivity and anti-interference capability. In summary, the core advantage of CFAR lies in its adaptability, which can automatically adjust sensitivity in unknown or time-varying background environments to ensure a stable false alarm rate and avoid missed detections or false alarms caused by fixed thresholds.

[0030] In a preferred embodiment, the background enhancement process includes: identifying strong scattering regions in the first background distance-angle amplitude map, locally expanding the strong scattering regions in both the distance and angle directions, and fusing the expansion results with the first background distance-angle amplitude map to expand the suppression range of stable strong background structures, enabling subsequent differential processing to cover slight drift, sidelobes, and edge remnants near the strong background; the first enhanced background map can be represented as: ;in, This represents the first enhanced background image. This represents the first background distance angle amplitude map. This represents a local enhancement term generated based on the background region with strong scattering.

[0031] In a preferred embodiment, the angular direction accumulation processing involves accumulating data from multiple scan cycles along the slow-time direction to enhance the response of real targets that appear consecutively in the angular direction and reduce the impact of random cusps in a single frame on the detection results; the accumulation process can be expressed as: ; ;in, This represents the cumulative graph of the first target. Indicates the first Target distance-angle amplitude diagram for each scan cycle, This represents the first enhanced background cumulative image. Indicates the first Enhanced background image for each scan cycle, This indicates the cumulative number of scan cycles.

[0032] In a preferred embodiment, the estimation of the angle offset includes: extracting a first target background structure contour from the first target cumulative image, extracting a first enhanced background structure contour from the first enhanced background cumulative image, calculating the similarity between the first target background structure contour and the first enhanced background structure contour within a preset candidate angle offset range, and determining the candidate angle offset with the highest similarity as the angle offset; the angle offset estimation can be expressed as: ;in, This indicates the operation of finding the maximum value. Indicates the candidate angle offset. Represents the set of candidate angle offsets. This represents the background structure outline extracted from the first target cumulative image. This represents the background structure contour extracted from the first enhanced background cumulative image. Sim(·) represents the similarity calculation function, and Shift(·) represents the angle translation function. This represents the estimated optimal angle offset.

[0033] In one embodiment, the first target cumulative image and the first enhanced background cumulative image are subjected to intensity compression or robust normalization to reduce the influence of local strong points on structural matching, and then the first target background structural contour and the first enhanced background structural contour are extracted.

[0034] In a preferred embodiment, it should be noted that the aforementioned single-scale background structure contour extraction method may have limitations when handling complex scenes with multi-scale features. In actual radar scenarios, background structures exhibit different scale characteristics at different distances: nearby buildings, equipment, etc., are rich in detail and require high-resolution contour extraction; distant mountains, large structures, etc., are larger in scale and require low-resolution contour extraction to avoid noise interference. To address this issue, this embodiment introduces a multi-scale pyramid decomposition method to extract background structure contours at different scales, and then performs weighted fusion matching, thereby improving the stability of background structures at different scales participating in angle registration.

[0035] In a preferred embodiment, the background structure contour extraction can employ a multi-scale pyramid decomposition method to adapt to scale variations in the background structure at different distances. Specifically, Gaussian pyramids are constructed for the first target cumulative image and the first enhanced background cumulative image, respectively. Then, the background structure contour is extracted at each pyramid level, and finally, the contour matching results at each level are weighted and fused.

[0036] The construction process of the Gaussian pyramid can be represented as follows: ;in, Indicates the first Layered pyramid image (unit: intensity value); Indicates the first Layered pyramid image (unit: intensity value); Indicates the first The standard deviation of the layer is Two-dimensional Gaussian kernel function (satisfying) ; Indicates distance index; Indicates the first Layered pyramid image at location The intensity value at that location; Indicates angle index; This represents the offset index of the convolution kernel in the distance direction; This indicates the offset index of the convolution kernel in the angular direction; Represents the radius of the Gaussian kernel in the distance direction; This represents the radius of the Gaussian kernel in the angular direction; Represents a pyramid-level index; This represents the total number of layers in the pyramid (typically 1). ), Indicates the first Gaussian kernel standard deviation of the layer The benchmark standard deviation (typically 1) is the standard deviation of the benchmark. ).

[0037] Extract the background structure outline at each pyramid level: ; ;in, Indicates the first Layer target background structure outline (binary image, value is 0 or 1); This represents the outline of the enhanced background structure in the layer (binary image, with values ​​of 0 or 1); Indicates the first Layered target cumulative map pyramid (unit: intensity value); Indicates the first Layer-enhanced background cumulative image pyramid (unit: intensity value); EdgeDetect(·) represents the edge detection function.

[0038] The correlation calculation for contour matching at each level uses normalized cross-correlation: ;in, Indicates the first Layer at offset Relevance (range) ); Indicates the candidate angle offset; This represents the summation over all distances and angular positions; Indicates the first Layered enhancement of background structure outline translation in angular direction The value after; Considering the contribution weights of different pyramid levels to the final registration, an adaptive weight based on the level resolution is introduced: ;in, Indicates the first The normalized weight coefficients of the layer (satisfying) ); This represents the weight decay factor (typically 2-3); This represents the exponential decay factor that increases with the level. This indicates weight adjustment based on the decay factor; The final optimal angle offset is determined by weighted fusion of matching results from each level: ;in, This represents the final estimated optimal angle offset; Indicates the set of candidate offsets Find the operation that maximizes the weighted correlation. This represents the set of candidate angle offsets.

[0039] In a preferred embodiment, the registered first enhanced background cumulative image is obtained by angularly shifting the first enhanced background cumulative image, and its expression is: ;in, This represents the first enhanced background cumulative image after registration. Indicates distance index, Indicates angle index, This represents the first enhanced background cumulative image. This represents the optimal angle offset. This indicates that the original first enhanced background cumulative image is translated upwards and to the right (or in the direction of increasing angle) at the angular index. After each unit, in the position The value taken at the location.

[0040] In a preferred embodiment, generating the first background suppression difference map includes: calculating a first positive difference between the first target cumulative map and the registered first enhanced background cumulative map. ;in, This represents the first positive difference. This represents the cumulative graph of the first target. This represents the first enhanced background cumulative image after registration. This represents the function that takes the maximum value. Indicates distance index, Indicates the angle index.

[0041] In a preferred embodiment, the generation of the first background suppression difference map further includes: calculating the background percentage. ;in, Indicates the proportion of the background. To avoid dividing by zero for extremely small positive numbers.

[0042] The generation of the first background suppression difference map further includes: generating background suppression weights based on the background proportion. ;in, This represents the background suppression weight. This represents a clipping function that restricts values ​​to a specified range. and These are the preset lower and upper weight limits, respectively.

[0043] In a preferred embodiment, the generation of the first background suppression difference map further includes: weighting the first positive difference using the background suppression weights to obtain the first background suppression difference map. ;in, This represents the first background suppression difference map.

[0044] In a preferred embodiment, before performing constant false alarm rate (CFAR) detection on the first background suppression difference map, the method further includes robustly normalizing the first background suppression difference map by angle column, the expression of which is: ;in, The graph represents the robustly normalized result. Median(·) denotes the median function, and MAD(·) denotes the absolute median difference function. In the first background suppression difference map, the first... All distance index values ​​for each angle column.

[0045] In a preferred embodiment, the candidate image post-processing includes at least one of angular continuity filtering, binary closing operation, and connected component extraction; the angular continuity filtering is used to delete isolated detection points with insufficient continuous width in the angular direction, the binary closing operation is used to connect small broken regions inside the same target, and the connected component extraction is used to extract the candidate target components from the post-processed binary image.

[0046] In a millimeter-wave radar target detection system, the original detection results generate a binary candidate image containing response points of multiple suspected targets. The post-processing stage employs at least one of the following operations: angle continuity filtering, binary closing operations, and connected component extraction. First, angle continuity filtering removes isolated points with insufficient continuous width along the angular direction. For example, a detection point may only respond in a single angular unit, with no response in adjacent angles, and its continuous width is less than a set threshold (e.g., 3 angular units), thus being identified as noise and deleted. Next, binary closing operations (dilation followed by erosion) connect small fracture regions within the same target caused by signal attenuation or interference—for example, a real foreign object may break at a mid-angle position, forming two separate blocks, which are merged into a continuous region through closing operations. Finally, connected component extraction performs 8-connectivity labeling on the processed binary image, outputting each connected region as a complete candidate target component for use by subsequent component-level background lookup modules. These three post-processing methods work synergistically to effectively suppress isolated noise, repair target structures, and accurately extract candidate components with reasonable morphology.

[0047] In a preferred embodiment, see Figure 2 The component-level background reverse lookup includes: calculating the target response features within the component, the background response features within the component, the newly added differential response features within the component, and the component morphological features for each candidate target component; when the candidate target component is also significant in the background and the newly added response is insufficient, the candidate target component is filtered as a background residual candidate; when the candidate target component has a newly added response relative to the background and meets the morphological requirements, the candidate target component is retained as a foreign object target candidate.

[0048] Component-level background detection is a refined mechanism for distinguishing between real and fake targets, primarily used to differentiate between genuine foreign objects and strong clutter or artifacts remaining in the background. It considers not only whether a candidate target is "significant," but also whether it exhibits "new, reasonable response features relative to the background." Specifically, the system first extracts multi-dimensional features from each candidate target component, including target response features within the component (such as signal strength and energy concentration), background response features of the component's location (i.e., the average clutter level of the surrounding environment), newly added differential response features (reflecting signal changes in the current frame compared to historical or reference backgrounds), and morphological features of the component (such as size, shape regularity, and edge continuity). During analysis, if a candidate component is found to have a strong signal but a similarly strong response in the background, and the newly added differential response is weak—meaning it is likely a long-standing static clutter or system noise residue—then the component will be classified as a "background residue candidate" and filtered out. Conversely, if a candidate component is not only significantly higher than the local background but also exhibits a marked new response (indicating its emergence) while its morphological characteristics fall within a reasonable range for a real target (e.g., not excessively broken or abnormally stretched), then it will be retained as a genuine "foreign object target candidate" and enter the subsequent processing flow. This mechanism effectively avoids misclassifying fixed clutter as a target, improving the robustness and accuracy of the detection system.

[0049] In practice, component-level background lookup typically follows these steps: First, the system obtains several candidate target components through preliminary detection (such as CFAR). Each component is a set of connected pixels or units in a distance-angle map or image domain. Next, for each candidate component, four types of features are calculated. The target response feature within a component typically uses the statistical values ​​of the signal strength of all units within that component, such as the mean, maximum, or total energy, to measure its "significance." The background response feature within a component obtains the original background intensity level of the area by querying the historical background map of the corresponding location (such as a long-term average or a clutter map obtained through modeling), used to determine whether the current response is "abnormal." The newly added differential response feature within a component calculates the difference (or ratio) between the current frame signal and the background reference map, then sums or averages the results within the component region. This reflects whether the component is newly appearing content. If the difference is close to zero, it indicates only background reproduction; if it is significantly positive, it may be a newly added target. The morphological characteristics of the components are analyzed from a geometric perspective, such as calculating the area, aspect ratio, compactness, or edge smoothness of the components, in order to eliminate noise spots or irregular clutter.

[0050] For example, suppose a candidate component covers five adjacent range-angle cells. The system first calculates the average current echo amplitude of these five cells (target response), and then reads the average clutter intensity at the same location in the background model (background response). If the two are close, the component may simply be background undulation; if the current value is significantly higher, the system further calculates the sum of the differences between the current value and the background (new response). Simultaneously, the system checks whether the component is elongated (potentially sidelobe interference) or overly dispersed (potentially noise), retaining only components with moderate area and compact shape. Ultimately, only components with significant new responses and reasonable morphology are identified as genuine foreign object candidates, while the rest are discarded as background remnants. This process effectively improves the reliability of the detection results, especially suitable for complex or dynamically changing background environments.

[0051] In a preferred embodiment, while the statistical feature-based reverse lookup method described above can effectively filter out most residual background candidates, it may still pose a risk of misjudgment for some complex background interferences with statistical characteristics similar to real foreign objects (such as metal structure sidelobes, multipath reflections, and edge effects of strong scatterers). To address this issue, this embodiment further introduces a verification method based on radar physical scattering characteristics. This method utilizes the inherent physical characteristic that real foreign targets have a specific scattering cross-section distribution pattern in the angular dimension to perform secondary verification of candidate targets, thereby significantly reducing the false alarm rate.

[0052] Component-level background lookup also includes candidate target authenticity verification based on radar cross-section angle dependence. Specifically, for each candidate target component, its intensity distribution characteristics in the angular dimension are analyzed, the rate of change of the cross-section angle is calculated, and the matching degree with the physical scattering model of a typical foreign object is evaluated. This verification mechanism is based on the following physical principles: Real foreign objects (such as metal fragments and rocks) typically exhibit a certain continuity or local concentration in echo intensity as the observation angle changes. In contrast, candidate responses formed by background misalignment, strong background edge remnants, sidelobes, or random noise tend to exhibit isolated spikes, abnormally flat distributions, or response patterns highly consistent with the background structure. Therefore, the angular intensity profile, angular change rate, breadth range, and degree of matching with a preset reference model of candidate target components can be used as auxiliary criteria to further distinguish between real foreign objects and background remnant candidates.

[0053] First, extract the normalized intensity profile of the candidate target component in the angular dimension: ;in, Indicates candidate components At angle Normalized cumulative strength at (satisfying) ); Indicates the location of the target cumulative graph. The intensity value at that location; Indicates candidate components A set of indices along the distance dimension; Indicates candidate components A set of indices along the angular dimension; Indicates distance index; Indicates angle index, It is a dummy variable used in the denominator for summation, used to iterate through all angular positions belonging to this component. To calculate the total energy. It is related to the energy in the molecule. They are different roles. It is the specific angular position from which the normalized value is to be calculated, and It's just a temporary variable name used for summation; This indicates that the cumulative target graph is at a distance index of Angle index is The intensity value at that location.

[0054] Calculate the rate of change of the scattering cross-section angle (i.e., the first-order difference of the normalized intensity profile): ;in, Indicates candidate components At angle The rate of change of the scattering cross-section angle at that location; Represents the set of angle indices after removing boundaries ( ),Right now Indexing the collection from the original perspective The set of angle indices remaining after removing the minimum and maximum values; Indicates candidate components At angle Normalized intensity at; Indicates candidate components At angle The normalized strength at that point.

[0055] Define a matching index for the scattering characteristics of candidate targets, using a reference Gaussian scattering model. ;in, Indicates candidate components Scattering characteristics matching degree (range) ); This represents the rate of change of the theoretical scattering cross-section angle of the reference foreign object target; Indicates the angle of the reference scattering center; Indicates the reference scattering angle broadening parameter (typical value) ).

[0056] The reference scattering model uses a normalized intensity profile with a Gaussian distribution: ;in, Represents the natural exponential function; This represents the squared difference between the angle index and the angle of the reference scattering center.

[0057] Its corresponding rate of change of angle is: ;in, This represents the normalized intensity of the reference model at the angle; Indicates the reference model at the angle The normalized strength at that point.

[0058] When the scattering characteristic matching degree is lower than a preset threshold, the candidate target is determined to be a false target: ;in, This represents the threshold for matching scattering characteristics (typical value 0.6~0.8). Indicates the filtering candidate component The operation.

[0059] Furthermore, to enhance the stability of the verification, this embodiment also introduces a physical constraint mechanism for the broadening of the scattering angle. The broadening of the scattering angle of a real foreign object target should be within a reasonable range; too narrow an angle may correspond to a single strong scattering point (such as background clutter), while too wide an angle may correspond to a large-area distribution (such as ground clutter).

[0060] Calculate the scattering angle broadening of candidate targets (i.e., the standard deviation of the normalized intensity profile): ;in, Indicates candidate components The scattering angle is broadened; Indicates candidate components Intensity-weighted average angle;

[0061] For real foreign objects, the broadening of their scattering angle should satisfy physical constraints: ;in, This represents the minimum scattering angle broadening threshold (typical value). ); This represents the maximum scattering angle broadening threshold (typical value). ).

[0062] Candidate targets outside this range are identified as false targets. The aforementioned verification method based on physical scattering characteristics complements the aforementioned reverse lookup method based on statistical features. The statistical feature reverse lookup mainly performs preliminary screening based on dimensions such as intensity ratio and morphological features, while the physical scattering characteristic verification performs in-depth verification based on the physical nature of radar scattering. The combination of the two can effectively distinguish between real foreign targets and complex background interference, significantly improving the reliability of the detection system.

[0063] In a preferred embodiment, the features used for determination in the component-level background lookup include at least: a first ratio of the cumulative background intensity within the component to the cumulative target intensity. ; And the second ratio of the positive differential intensity within the component to the cumulative background intensity: ;in, This represents the candidate target component, and Mean(·) represents the function to calculate the mean. This represents the value of the registered enhanced background cumulative map within the range of the candidate target components. Indicated in the candidate target component The cumulative value of the target within the range, Indicated in the candidate target component The values ​​of the positive difference plot within the range.

[0064] In a preferred embodiment, the first target cumulative image and the first enhanced background cumulative image are subjected to intensity compression or robust normalization to reduce the influence of local strong points on structural matching, and then the first target background structural contour and the first enhanced background structural contour are extracted.

[0065] A background registration and component-level reverse lookup radar foreign object detection system, applied to a background registration and component-level reverse lookup radar foreign object detection method, includes a first radar data acquisition module, a first basic preprocessing module, a first background enhancement module, a first angle accumulation module, a first background self-registration module, a first background suppression differential module, a first constant false alarm rate detection module, a first candidate image post-processing module, a first component-level reverse lookup module, and a first result output module, wherein: The first radar data acquisition module is used to acquire the first background radar data in the absence of foreign objects and the first target radar data of the scene to be detected; The first basic preprocessing module is used to perform basic preprocessing on the first background radar data and the first target radar data respectively, to generate a first background range-angle amplitude map and a first target range-angle amplitude map; The first background enhancement module is used to perform background enhancement processing on the first background distance-angle amplitude map to obtain a first enhanced background map; The first angle accumulation module is used to accumulate the first target distance-angle magnitude map along the angle direction to obtain a first target accumulation map, and to accumulate the first enhanced background map along the angle direction to obtain a first enhanced background accumulation map; The first background self-registration module is used to estimate the angle offset based on the background structure features of the first target cumulative image and the first enhanced background cumulative image, and to perform angle registration on the first enhanced background cumulative image based on the angle offset to obtain the registered first enhanced background cumulative image. The first background suppression difference module is used to calculate a positive difference based on the first target cumulative map and the registered first enhanced background cumulative map and generate a background suppression weight, thereby obtaining a first background suppression difference map; The first constant false alarm rate (CFAR) detection module is used to perform CFAR detection on the first background suppression difference map. The first candidate image post-processing module is used to perform candidate image post-processing on the constant false alarm rate detection results to obtain multiple candidate target components; The first component-level background back lookup module is used to back-look up the first target cumulative image, the registered first enhanced background cumulative image and the positive difference in units of the candidate target components, calculate the component-level background residual features, and filter background residual candidates according to the component-level background residual features; The first result output module is used to output the foreign object detection results.

[0066] In one embodiment, the estimation of the angular offset does not depend on manually specified target distance or target angle, but is based on adaptive registration based on the general background structure in the scene.

[0067] In one embodiment, angular direction cumulative processing is used to enhance the response of real targets that appear continuously in the angular direction and reduce the impact of random cusps in a single frame on the detection results.

[0068] In one embodiment, the first basic preprocessing module performs DC removal, windowing, frequency domain transformation, and amplitude map generation on the background data and target scene data; the first background enhancement module locally expands and enhances the background strong scattering region; the first angle accumulation module accumulates the target image and the enhanced background image along the angular or slow-time direction; the first background self-registration module estimates the angular offset based on the background structure and enhanced background structure in the target scene and registers the enhanced background image; the first background suppression difference module generates a background suppression difference image based on the target accumulated image and the registered enhanced background image; the first robust normalization module generates a constant false alarm rate (CFAR) detection input image; the first CFAR detection module generates a candidate point binary image; the first candidate image post-processing module performs angular continuity filtering, binary closing operation, and connected component extraction; and the first component-level background reverse lookup module calculates component-level background residual features for candidate target components and filters background residual candidates.

[0069] In a preferred embodiment, the first background enhancement module includes a first strong scattering region identification unit, a first local expansion unit, and a first fusion unit. The first strong scattering region identification unit is used to identify strong scattering regions in the first background distance-angle amplitude map. The first local expansion unit is used to locally expand the strong scattering regions in the distance direction and the angle direction to obtain a first local enhancement term. The first fusion unit is used to fuse the first local enhancement term with the first background distance-angle amplitude map to obtain the first enhanced background map.

[0070] In a preferred embodiment, the first background self-registration module includes a first background structure contour extraction unit, a first similarity calculation unit, and a first angle registration unit. The first background structure contour extraction unit is used to extract a first target background structure contour from the first target cumulative image and extract a first enhanced background structure contour from the first enhanced background cumulative image. The first similarity calculation unit is used to calculate the similarity between the first target background structure contour and the first enhanced background structure contour after translation within a preset candidate angle offset range. The first angle registration unit is used to perform angle registration on the first enhanced background cumulative image based on the candidate angle offset with the highest similarity.

[0071] In a preferred embodiment, the first background suppression difference module includes a first positive difference calculation unit, a first background suppression weight generation unit, and a first weighting unit. The first positive difference calculation unit is used to calculate the first positive difference between the first target cumulative map and the registered first enhanced background cumulative map. The first background suppression weight generation unit is used to generate a first background suppression weight based on the proportion of the registered first enhanced background cumulative map in the first target cumulative map. The first weighting unit is used to weight the first positive difference using the first background suppression weight to obtain the first background suppression difference map.

[0072] In a preferred embodiment, the first constant false alarm rate (CFAR) detection module is further included before the first robust normalization module. The first robust normalization module is used to perform robust normalization processing on the first background suppression difference map according to the angle column. The processing method of the first robust normalization module is: calculate the median and absolute median difference for each angle column, and generate a normalized map as the input map for constant false alarm rate detection.

[0073] In a preferred embodiment, the first candidate image post-processing module includes at least one of a first angle continuity filtering unit, a first binary closing operation unit, and a first connected component extraction unit. The first angle continuity filtering unit is used to delete isolated detection points with insufficient continuous width in the angle direction. The first binary closing operation unit is used to connect small broken regions inside the same target. The first connected component extraction unit is used to extract the candidate target components from the post-processed binary image.

[0074] In a preferred embodiment, the first component-level background reverse lookup module includes a first component feature calculation unit and a first candidate filtering unit. The first component feature calculation unit is used to calculate the target response features within the component, the background response features within the component, the newly added differential response features within the component, and the component morphology features for each candidate target component. The first candidate filtering unit is used to filter background residual candidates based on the component-level background residual features.

[0075] In one embodiment, the features used for determination in the component-level background reverse lookup also include: the average or peak value of the cumulative intensity of the target within the component, the average or peak value of the cumulative intensity of the background within the component, the average or peak value of the positive differential intensity within the component, the intensity ratio of the background to the target, the intensity ratio of the positive differential to the background, and the component area, number of points, fill rate, angle width, and distance width.

[0076] During component-level background lookup, the system integrates multiple features to determine whether a candidate target component is a real target, rather than clutter or noise. These features are mainly based on three dimensions: intensity, morphology, and spatial distribution, forming a multi-faceted discrimination criterion. First, regarding intensity, the system not only focuses on the cumulative mean or peak intensity of the target signal within the component but also assesses the intensity level of the surrounding background region. By comparing the two (e.g., calculating the intensity ratio of the target to the background), it measures whether the target significantly "stands out" of the environment. Furthermore, the system introduces positive differential intensity, i.e., the degree of change of the target signal relative to the background. Its mean or peak value and its ratio to the background intensity more sensitively reflect the dynamic characteristics or edge sharpness of the target, helping to distinguish between residual background responses and genuine newly added foreign object responses.

[0077] Building upon this foundation, the system further incorporates the geometric and structural features of components for auxiliary judgment. For example, the pixel area occupied by a component, the number of points it contains, its fill rate (i.e., the ratio of actual points to the total number of pixels within the bounding box), and its broadening in the angular and distance dimensions (i.e., angular width and distance width) collectively characterize the spatial distribution of the target. Real targets typically possess a reasonable size range and a compact structure, while noise or interference often manifests as point-like, sparse, or anomalously shaped fragments. By organically combining intensity and morphological features, the background detection mechanism can more robustly distinguish between valid targets and false responses, thereby improving the accuracy and robustness of the overall detection system.

[0078] In a preferred embodiment, the features calculated by the first component feature calculation unit include at least: a first ratio of the cumulative background intensity to the cumulative target intensity within the component. The second ratio of the positive differential intensity within the component to the cumulative intensity of the background. .

[0079] In a preferred embodiment, the first radar data acquisition module, the first basic preprocessing module, the first background enhancement module, the first angle accumulation module, the first background self-registration module, the first background suppression differential module, the first constant false alarm rate detection module, the first candidate image postprocessing module, the first component-level background reverse lookup module, and the first result output module are implemented by software programs, hardware circuits, or a combination of software and hardware, and are integrated in the same processing device or distributed in the radar front end, edge computing device, or host computer according to the radar system architecture.

[0080] In a preferred embodiment, the detection results output by the first result output module include at least one of target distance, target angle, candidate box, target intensity, detection level, or confidence information. The function of the first result output module is to present the target information finally identified by the detection algorithm in a structured form, facilitating subsequent analysis or application system use. Its output typically includes at least one of target distance, target angle, candidate box, target intensity, and detection level or confidence information. Target distance and angle reflect the target's position in space, determined by distance axis mapping and angle estimation algorithms in signal processing, respectively. The candidate box describes the boundary of the area occupied by the target in the distance-angle map (or similar representation), often used for multi-target separation or visualization annotation. Target intensity represents the strength of the target's echo, generally derived from normalized amplitude or energy values, and can indirectly reflect the target's size, material, or reflectivity. The detection level or confidence information is used to measure the reliability of the detection result, typically based on a comprehensive judgment of the contrast between the signal and background noise, CFAR detection margin, or the output probability of the classification model. These pieces of information are not isolated but interconnected, forming a complete detection output. For example, a high-confidence target typically corresponds to clear angular localization, reasonable distance estimation, and a significant intensity response. By integrating these elements, the first output module not only tells the system "whether there is a target," but also provides key information such as "where the target is," "how strong it is," and "how credible this judgment is," providing a solid foundation for subsequent tracking, classification, or decision-making. This invention can reduce subtraction residue caused by background misalignment, reduce false targets at strong background edges, and improve the detection stability of distant, weak foreign objects against strong static backgrounds.

[0081] Although preferred embodiments of the present invention have been described in detail herein, those skilled in the art, upon understanding its core inventive concept, can still make various modifications and adjustments. Therefore, the appended claims should be understood to cover not only the above-described preferred embodiments but also all equivalent variations, modifications, and alternatives falling within the scope of the present invention. It should be emphasized that the above content is merely an example of preferred embodiments of the present invention and is not intended to limit the scope of protection of the present invention; any modifications, equivalent substitutions, optimizations, or improvements made in accordance with the spirit and principles of the present invention should be considered as included within the patent protection scope of the present invention.

Claims

1. A method for background registration and component reverse radar foreign object detection, characterized in that, Includes the following steps: Acquire first background radar data in a non-object-free state and first target radar data of the target scene to be detected, and perform basic preprocessing on the first background radar data and the first target radar data respectively to generate a first background range-angle amplitude map and a first target range-angle amplitude map; The first background distance-angle amplitude map is subjected to background enhancement processing to identify strong scattering regions and locally expand them in the distance and angle directions to obtain a first enhanced background map; the first target distance-angle amplitude map is cumulatively processed along the angle direction to obtain a first target cumulative map, and the first enhanced background map is cumulatively processed along the angle direction to obtain a first enhanced background cumulative map; The angle offset is estimated based on the background structure features of the first target cumulative image and the first enhanced background cumulative image, and the angle is registered based on the angle offset to obtain the registered first enhanced background cumulative image. Based on the first target cumulative map and the registered first enhanced background cumulative map, a positive difference is calculated and a background suppression weight is generated, thereby obtaining the first background suppression difference map; The first background suppression difference map is subjected to constant false alarm rate detection and candidate map post-processing to obtain multiple candidate target components; The first target cumulative image, the registered first enhanced background cumulative image, and the positive difference are retrieved back for each candidate target component to calculate the component-level background residue features. Based on the component-level background residue features, background residue candidates are filtered and the foreign object detection results are output.

2. The background registration and component reverse lookup radar foreign object detection method according to claim 1, characterized in that, The background enhancement process includes: identifying strong scattering regions in the first background distance-angle amplitude map, locally expanding the strong scattering regions in both the distance and angle directions, and fusing the expansion results with the first background distance-angle amplitude map to expand the suppression range of stable strong background structures, enabling subsequent differential processing to cover slight drift, sidelobes, and edge remnants near the strong background; the first enhanced background map is represented as: ； in, This represents the first enhanced background image. This represents the first background distance-angle amplitude map. This represents a local enhancement term generated based on the background region with strong scattering.

3. The background registration and component reverse lookup radar foreign object detection method according to claim 1, characterized in that, The estimation of the angle offset includes: extracting the first target background structure contour from the first target cumulative image, extracting the first enhanced background structure contour from the first enhanced background cumulative image, calculating the similarity between the first target background structure contour and the first enhanced background structure contour within a preset candidate angle offset range, and determining the candidate angle offset with the highest similarity as the angle offset; the angle offset estimation is expressed as: ； in, This indicates the operation of finding the maximum value. Indicates the candidate angle offset. This represents the set of candidate angle offsets. This represents the background structure outline extracted from the first target cumulative image. This represents the background structure outline extracted from the first enhanced background cumulative image. Sim(·) represents the similarity calculation function, and Shift(·) represents the angle translation function. This represents the estimated optimal angle offset.

4. The background registration and component reverse lookup radar foreign object detection method according to claim 1, characterized in that, The generation of the first background suppression difference map includes: calculating the first positive difference between the first target cumulative map and the registered first enhanced background cumulative map. ； in, This represents the first positive difference. This represents the cumulative graph of the first target. This represents the first enhanced background cumulative image after registration. This represents the function that takes the maximum value. Indicates distance index, Indicates the angle index.

5. The background registration and component reverse lookup radar foreign object detection method according to claim 4, characterized in that, The generation of the first background suppression difference map also includes: calculating the background percentage. ； in, Indicates the proportion of the background. To avoid extremely small positive numbers divided by zero, the generation of the first background suppression difference map also includes: generating background suppression weights based on the background proportion. ； in, This represents the background suppression weight. This represents a clipping function that restricts values ​​to a specified range. and These are the preset lower and upper weight limits, respectively.

6. The background registration and component reverse lookup radar foreign object detection method according to claim 5, characterized in that, The generation of the first background suppression difference map further includes: weighting the first positive difference using the background suppression weights to obtain the first background suppression difference map. ； in, This represents the first background suppression difference map.

7. The background registration and component reverse lookup radar foreign object detection method according to claim 1, characterized in that, Before performing constant false alarm rate (CFAR) detection on the first background suppression difference map, the method further includes robust normalization of the first background suppression difference map by angle column, the expression of which is: ； in, The graph represents the robustly normalized result. Median(·) denotes the median function, and MAD(·) denotes the absolute median difference function. In the first background suppression difference map, the first... All distance index values ​​for each angle column.

8. The background registration and component reverse lookup radar foreign object detection method according to claim 1, characterized in that, The component-level background reverse lookup includes: calculating the target response features within the component, the background response features within the component, the newly added differential response features within the component, and the component morphological features for each candidate target component; when the candidate target component is also significant in the background and the newly added response is insufficient, the candidate target component is filtered as a background residual candidate; when the candidate target component has a newly added response relative to the background and meets the morphological requirements, the candidate target component is retained as a foreign object target candidate.

9. The background registration and component reverse lookup radar foreign object detection method according to claim 8, characterized in that, The features used for determination in the component-level background reverse lookup include at least: a first ratio of the cumulative intensity of the background within the component to the cumulative intensity of the target. ； And the second ratio of the positive differential intensity within the component to the cumulative background intensity: ； in, This represents the candidate target component, and Mean(·) represents the function to calculate the mean. Indicated in the candidate target component The value of the registered enhanced background cumulative map within the range, Indicated in the candidate target component The cumulative value of the target within the range, Indicated in the candidate target component The values ​​of the positive difference plot within the range.

10. The background registration and component reverse lookup radar foreign object detection method according to claim 3, characterized in that, The first target cumulative image and the first enhanced background cumulative image are subjected to intensity compression or robust normalization to reduce the influence of local strong points on structural matching, and then the first target background structure contour and the first enhanced background structure contour are extracted.

Images