Space-based observation strong clutter characterization and suppression method

Abstract

The application discloses a space-based observation strong clutter characterization and suppression method, and specifically comprises the following steps: step 1, a parameterized space-time model is established based on the slow variation characteristics of the clutter in a short time window, and the clutter is modeled and characterized; step 2, an iterative least square estimation method is adopted to simultaneously realize optimal estimation of the jitter offset and the clutter model parameters through a closed iterative loop; step 3, an index for quantitatively evaluating the clutter suppression effect is proposed, and according to the evaluation index, an optimal filter is adaptively selected from a filter library containing Fourier, wavelet and polynomial space-time filters, the original image is filtered, effective suppression of the clutter is realized, the strong clutter is suppressed to the sensor noise level or lower, and the sea-air weak and small target monitoring performance in a complex background is significantly improved.

Description

Technical Field

[0001] This invention belongs to the field of space-based infrared target detection technology and relates to a clutter suppression method, specifically a method for characterizing and suppressing strong clutter in space-based observation. Background Technology

[0002] Space-based infrared detection has the characteristics of a large field of view and continuous operation around the clock. It has broad application prospects in many fields such as temperature measurement, night vision monitoring, agricultural supervision and road monitoring. Therefore, space-based infrared detection is an effective technical means to achieve the monitoring of sea and air targets.

[0003] However, during detection, the signals received by the sensor are far from pure target radiation. Strong background clutter, coupled with long-distance transmission causing sea and air targets to appear weak in infrared staring camera imaging, easily leads to false alarms and erroneous trajectories during target monitoring, affecting the effectiveness of the infrared detection system. Background clutter mainly originates from the scattering of sunlight through clouds, aerosols, and the Earth's surface, or its own thermal radiation—its intensity is typically tens or even hundreds of times greater than the sensor's own noise and the weak target signal being detected. For space-based infrared staring sensors, platform vibration (shaking) causing line-of-sight sway further exacerbates the non-stationary changes in background clutter on the image plane. Especially for small, weakly radiating sea and air targets, their signals are easily overwhelmed by strong background clutter; effective clutter suppression technology is crucial for improving their detection performance.

[0004] Existing clutter suppression techniques can be mainly categorized into spatial domain, temporal domain, and spatiotemporal joint processing methods. Spatial filtering methods, including median filtering, matrix filtering, and Fourier filtering, separate targets and backgrounds by utilizing the spatial differences in their spatial characteristics within a single frame. These methods are effective for weak clutter with strong spatial correlation, but their fundamental limitation lies in their inability to utilize the temporal information of target motion, and their performance degrades drastically when platform jitter causes scene non-stationarity. The classic temporal domain method is the inter-frame difference method, which suppresses stationary or slowly changing backgrounds by registering and subtracting consecutive frames. This method has a good suppression effect on stationary backgrounds, but it relies on subpixel-level precise image registration and is extremely sensitive to platform jitter. Simple differencing can double the sensor noise power, further degrading the detection capability for small targets in the air and sea. While advanced methods such as spatiotemporal adaptive processing combine spatiotemporal information, their performance often depends on a large number of independent and identically distributed training samples to accurately estimate statistical characteristics. In the complex and variable non-uniform clutter environment of space-based infrared scenes, it is difficult to obtain sufficient samples that meet the conditions, resulting in degraded performance. Summary of the Invention

[0005] To address the issues of platform jitter and clutter nonstationarity faced by space-based infrared sensors in strong clutter backgrounds, this invention provides a method for characterizing and suppressing strong clutter in space-based observations.

[0006] The objective of this invention is achieved through the following technical solution:

[0007] A method for characterizing and suppressing strong clutter in space-based observations includes the following steps:

[0008] Step 1: Establish a parameterized spatiotemporal model based on the slow-varying characteristics of clutter within a short time window to model and characterize the clutter;

[0009] Step 2: The iterative least squares estimation method is adopted, and the optimal estimation of jitter offset and clutter model parameters is achieved simultaneously through a closed iterative loop.

[0010] Step 3: Propose quantitative evaluation indicators for clutter suppression effect, and based on the evaluation indicators, adaptively select the optimal filter from a filter library containing various spatiotemporal filters such as Fourier, wavelet, polynomial, and dual-resolution spline filters to filter the original image, thereby achieving effective clutter suppression and reducing strong clutter to the sensor noise level or lower, significantly improving the monitoring performance of small sea and air targets in complex backgrounds.

[0011] Compared with the prior art, the present invention has the following advantages:

[0012] 1. To address the problem of extremely low signal-to-clutter-to-noise ratio of targets and the failure of traditional filtering methods under strong clutter backgrounds, this invention establishes a parameterized spatiotemporal clutter model and uses the characteristics of high clutter intensity and short-term correlation for iterative estimation, thereby achieving "whitening" processing that suppresses the background to below the sensor noise level, significantly improving the detection floor for weak targets in the sea and air.

[0013] 2. To address the problem of image instability caused by high-frequency platform jitter and the need for expensive external stabilization devices in traditional methods, this invention proposes an iterative jitter compensation algorithm embedded in the filtering process. It innovatively uses strong clutter as a "stabilization reference source" and estimates and corrects sub-pixel-level inter-frame offset in real time through minimum distance matching, achieving adaptive scene stabilization without relying on sophisticated hardware.

[0014] 3. To address the problem that clutter spatial structures are complex and variable in actual observations and that a single fixed model is difficult to achieve global optimization, this invention constructs a reconfigurable filter library containing multi-principle filters and designs an adaptive selection mechanism based on real-time performance indicators (G factor, Q factor) to ensure that the algorithm can automatically maintain the optimal suppression performance under the current conditions when facing various scenarios, from uniform backgrounds to highly structured cloud edges. Attached Figure Description

[0015] Figure 1This is a flowchart of a method for characterizing and suppressing strong clutter in space-based observations. Detailed Implementation

[0016] The technical solution of the present invention will be further described below with reference to the accompanying drawings, but it is not limited thereto. Any modifications or equivalent substitutions to the technical solution of the present invention that do not depart from the spirit and scope of the technical solution of the present invention should be covered within the protection scope of the present invention.

[0017] This invention provides a method for characterizing and suppressing strong clutter in space-based observations. Addressing the difficulty in characterizing and modeling clutter, this method constructs a parameterized spatiotemporal clutter characterization model based on the characteristics of clutter intensity being much higher than noise and targets, and exhibiting spatiotemporal correlation within a short time window. It assumes that clutter is a spatially arbitrarily varying but temporally slowly varying function, with its inter-frame variations primarily stemming from overall translation caused by platform jitter. Using the observed image sequence, the clutter background is modeled as a linear combination of a selected set of basis functions, with their coefficients serving as parameters to be estimated. To address the image mismatch caused by platform jitter and the sharp performance degradation of traditional methods, a joint iterative spatiotemporal filtering and jitter compensation algorithm is proposed. Utilizing the strong clutter signal itself as a stable reference source, self-stabilization and clutter estimation are achieved through an iterative method. Based on the background model estimated in the previous frame, minimum distance matching is performed with the current frame to accurately estimate the inter-frame jitter offset. Using historical data from multiple frames in a calibrated coordinate system, the clutter model parameters are re-estimated using the least squares method. The updated model is used to predict and subtract the clutter from the current frame, outputting the processed residual image. To address the challenge of obtaining sufficient samples to meet the requirements in complex and ever-changing real-world observation environments, a reconfigurable filter bank and automatic selection mechanism are designed and integrated. This filter bank includes various spatiotemporal filters constructed based on different basis function principles, as well as classical spatial filters and temporal difference filters serving as performance benchmarks. The system automatically selects the optimal filter and its parameter configuration for the current scene by evaluating the performance metrics of each filter's output in real time, thereby ensuring the algorithm's robustness and optimal performance. Figure 1 As shown, the specific steps include the following:

[0018] Step 1: Establish a parameterized spatiotemporal model based on the slow-varying characteristics of clutter within a short time window to model and characterize the clutter. Specific steps are as follows:

[0019] Step 1-1: Assume the algorithm's input is a series of consecutive frames of 2D images that have been registered with infrared sensors, with a pixel size of [missing information]. ,image Represented as:

[0020]

[0021]

[0022] in, Sensor noise; Background clutter; For target features associated with the sensor's point spread function (PSF), the coordinates are... Maximum strength is ; It is the first Number of unknown targets in the frame; This refers to the unknown planar displacement caused by jitter; The coordinates in the image are Pixels; , .

[0023] Steps 1-2: Limit the scope of application to the following types of residual filtering algorithms:

[0024]

[0025] in, It is the output of the filter; Based on the Frame Image To the Frame Image In a size of The estimated value of the sliding window is used to estimate the current value of the first sliding window. Pixels in a frame Clutter components at the location .

[0026] Steps 1-3: Slide a window over a relatively short time period Intraframe, background clutter Changes caused by physical factors (such as wind and light variations) are very slow and negligible; the changes they exhibit in image sequences mainly originate from... The resulting overall translation. Parameters The value of determines the time span during which the clutter signal remains relatively stable. This parameter, determined experimentally, corresponds to the length of the time window suitable for time-domain filtering. Therefore, the clutter background can be modeled as a parameterized spatial function. Specifically, assuming a non-inertial coordinate system, clutter can be represented by a set of selected basis functions. Linear combination:

[0027]

[0028] in, It is an unknown parameter. The model order is... These are the basis functions selected from the best-fit criterion. Choice and its dimension The allowable approximation error is determined by the real clutter, which may have very high spatial variation. This coordinate system corresponds to the nominal translation δ.

[0029] make Indicates a length of time window Internally, based on data of An estimate in which ={ (i,j)}, , For any nominal translation , the background is at the . The frame prediction estimate has the form:

[0030]

[0031] Step 2: Employ the iterative least squares estimation method, using a closed iterative loop to simultaneously achieve optimal estimates of both jitter offset and clutter model parameters. The specific steps are as follows:

[0032] Step 2-1, assuming for all frames Translation was estimated. ,make These offset estimates are represented as follows:

[0033]

[0034] in, Indicates within the time window The total squared error within the parameter is used to measure the total squared error. and the set of known offset estimates The degree of fit to the observed data, Indicates the first The image of the frame, Indicates the first One parameter.

[0035] Step 2-2, Estimated value This is derived using the following least squares method:

[0036]

[0037] Steps 2-3: Perform jitter compensation and background estimation in the parameterized model class using an iterative process. The iterative steps are as follows:

[0038] Step 2-3-1, Initialization of the iterative algorithm: using the previous... Frame data is used to obtain initial jitter estimation sequences and initial clutter parameter estimates through a simple autonomous registration algorithm. This step takes approximately [time period missing]. The result of the initialization phase is the nth observation. The parameter in the first... Frame estimate ( ), offset estimation set ( ), = , No. Frame in pixel The clutter estimate at the location.

[0039] Step 2-3-2, Jitter estimation (for the first...) frame, > ): Using the clutter background estimation obtained from the previous frame iteration As a stable reference, it is matched with the current frame, and the offset that minimizes the difference between the two is calculated. For the Frame jitter estimation uses minimum distance estimation. That is, the solution to the following nonlinear minimization problem:

[0040]

[0041] When the noise follows a Gaussian distribution, this estimate corresponds to the maximum likelihood estimator. For the maximum allowable offset, For the first The predicted value of the frame background estimation. is the offset vector to be searched.

[0042] Step 2-3-3, Parameter Estimation: Obtain jitter estimates for the current frame and historical frames within the window ( After that, all frames are aligned to the corrected coordinate system. Then, the clutter parameters at the current time are re-estimated by solving a least-squares problem:

[0043]

[0044] Steps 2-3-4, Clutter Suppression and Output: Using the latest estimated parameters and jitter, reconstruct the prediction of clutter for the current frame. And output the filtered residual image:

[0045]

[0046] In this residual image, strong background clutter has been greatly suppressed, leaving mainly noise and potential target signals.

[0047] Step 3: Propose quantitative evaluation indicators for clutter suppression effectiveness. Based on these indicators, adaptively select the optimal filter from a filter library containing various spatiotemporal filters such as Fourier, wavelet, and polynomial filters to filter the original image, achieving effective clutter suppression. This reduces strong clutter to sensor noise levels or lower, significantly improving the performance of air and sea target monitoring in complex environments. Specific steps are as follows:

[0048] Step 3-1: A good filtering algorithm must preserve the signal as much as possible while suppressing clutter as much as possible. Therefore, an index is proposed to evaluate the clutter suppression effect:

[0049] (1) The factor defines the relative value of the residual clutter standard deviation:

[0050]

[0051] in, It is the relative value of the residual clutter standard deviation. It is the variance of the sensor noise. It is the variance of the output frames. Clearly, ≥1, the ideal value of this factor is 1, which means complete suppression of clutter.

[0052] (2) The factor defines the signal attenuation coefficient:

[0053] in, It is the signal attenuation coefficient. For input signal, For the output signal, under ideal conditions, =0, which means accurate signal recovery (no signal degradation).

[0054] (3) The factor defines the relative value of the effective S(C+N)R:

[0055]

[0056] in, It is the relative value of the effective S(C+N)R, due to signal attenuation. , The ideal value of the factor is 1, which means that clutter is completely suppressed and the signal is not degraded.

[0057] Step 3-2: For the basis function library, select the following parameter models and corresponding spatiotemporal filters to form the filter library:

[0058] (1) A two-dimensional Fourier series "Fourier" with two Nyquist laws: function Represented by Fourier series, the spatial frequency is and , , , , Corresponding to and Frequency components in the direction, Indicates the image in and Number of pixels in the direction, and Indicates the pixel in and Effective size of direction, function The spectrum can be recovered in a wider band than the Nyquist band because the platform vibration allows the value of the function b(x, y) to be observed at the midpoint and its spectrum to be recovered in any band.

[0059] (2) Two-dimensional wavelet sequence "wavelet": This method is similar to the previous one, but uses a wavelet basis instead of a Fourier basis. The main advantage of using a wavelet basis in the model is that, unlike the Fourier basis, wavelets have compact support properties. When describing functions containing a large number of jumps, edges, and other local "sharp" features, wavelets can efficiently capture these features with "poor frequency responses".

[0060] (3) Local polynomial approximation of Pol: This model utilizes functions A first-order polynomial surface approximation for each pixel. Specifically, for each pixel... The frame that has been shifted relative to the reference frame can be approximated as:

[0061]

[0062]

[0063] in, Represents the coordinates after offset The clutter (background) intensity value at that location, Represent a A basis function vector of dimension, = ( ) is the first coefficients of the polynomial of degree dimensional vector, The reference frame is the first component. , , , This indicates transpose.

[0064] Step 3-3: Set up a central performance monitoring and selection module to analyze the output of each filter channel in real time and receive feedback from subsequent target detection stages. The central performance monitoring and selection module optimizes according to predefined criteria (such as maximizing...). Factor, Minimization The system automatically switches or fuses the optimal filter configuration for the current specific clutter scene (such as high-structure clouds or uniform sea-sky background) to ensure that the system always works in the best state.

Claims

1. A method for characterizing and suppressing strong clutter in space-based observations, characterized in that... The method includes the following steps: Step 1: Establish a parameterized spatiotemporal model based on the slow-varying characteristics of clutter within a short time window to model and characterize the clutter; Step 2: The iterative least squares estimation method is adopted, and the optimal estimation of jitter offset and clutter model parameters is achieved simultaneously through a closed iterative loop. Step 3: Propose a quantitative evaluation index for the clutter suppression effect, and based on the evaluation index, adaptively select the optimal filter from a filter library containing various spatiotemporal filters such as Fourier, wavelet, and polynomial filters to filter the original image and achieve effective clutter suppression.

2. The method for characterizing and suppressing strong clutter in space-based observations according to claim 1, characterized in that... The specific steps of step 1 are as follows: Step 1-1: Assume the input is a series of consecutive frames of 2D images that have been registered with an infrared sensor, with a pixel size of [missing information]. ,image Represented as: in, This is sensor noise; Background clutter; The target features are related to the point spread function of the sensor, with coordinates as follows: Maximum strength is ; It is the first Number of unknown targets in the frame; This refers to the unknown planar displacement caused by jitter; The coordinates in the image are Pixels; , ; Steps 1-2: Limit the scope of application to the following types of residual filtering algorithms: in, It is the output of the filter; Based on the Frame Image To the Frame Image In a size of The estimated value of the sliding window is used to estimate the current value of the first sliding window. Pixels in a frame Clutter components at the location ; Steps 1-3: Assume that in a non-inertial coordinate system, clutter is represented by a set of selected basis functions. Linear combination: in, It is an unknown parameter. The model order is... These are the basis functions selected from the best-fit criteria; make Indicates a length of time window Internally, based on data of An estimate in which ={ (i,j)}, , For any nominal translation The background is in The frame prediction estimate has the following form: 。 3. The method for characterizing and suppressing strong clutter in space-based observations according to claim 1, characterized in that... The specific steps of step 2 are as follows: Step 2-1, assuming for all frames Translation was estimated. ,make These offset estimates are represented as follows: in, Indicates within the time window The total squared error within the parameter is used to measure the total squared error. and the set of known offset estimates The degree of fit to the observed data, Indicates the first The image of the frame, Indicates the first One parameter; Step 2-2, Estimated Value This is derived using the following least squares method: Steps 2-3: Perform jitter compensation and background estimation in the parameterized model class using an iterative process. The iterative steps are as follows: Step 2-3-1, Initialization of the iterative algorithm: using the previous... Frame data, through an autonomous registration algorithm, yields initial jitter estimation sequences and initial clutter parameter estimates. The result of the initialization phase is the first... The parameter in the first... Frame estimate ( ), offset estimation set ( ), = , No. Frame in pixel Clutter estimate at the location; Step 2-3-2, Jitter Estimation: Utilize the clutter background estimate obtained from the previous frame iteration. As a stable reference, it is matched with the current frame, and the offset that minimizes the difference between the two is calculated. For the Frame jitter estimation uses minimum distance estimation. That is, the solution to the following nonlinear minimization problem: For the maximum allowable offset, For the first The predicted value of the frame background estimation. Let be the offset vector to be searched. When the noise is Gaussian distributed, this estimate corresponds to the maximum likelihood estimator. Step 2-3-3, Parameter Estimation: Obtain jitter estimates for the current frame and historical frames within the window ( After that, all frames are aligned to the corrected coordinate system, and the clutter parameters at the current time are re-estimated by solving a least-squares problem. Steps 2-3-4, Clutter Suppression and Output: Using the latest estimated parameters and jitter, reconstruct the prediction of clutter for the current frame. And output the filtered residual image: 。 4. The method for characterizing and suppressing strong clutter in space-based observations according to claim 1, characterized in that... In step 3, the evaluation indicators include the relative value of the residual clutter standard deviation, the signal attenuation coefficient, and the relative value of the effective S(C+N)R.

5. The method for characterizing and suppressing strong clutter in space-based observations according to claim 4, characterized in that... The relative value of the residual clutter standard deviation is expressed as: in, It is the relative value of the residual clutter standard deviation. It is the variance of the sensor noise. It is the variance of the output frames.

6. The method for characterizing and suppressing strong clutter in space-based observations according to claim 4, characterized in that... The signal attenuation coefficient is expressed as: in, It is the signal attenuation coefficient. For input signal, This is the output signal.

7. The method for characterizing and suppressing strong clutter in space-based observations according to claim 4, characterized in that... The relative value of the effective S(C+N)R is expressed as: in, It is the relative value of the effective S(C+N)R. For input signal, For output signal, It is the variance of the sensor noise. It is the variance of the output frames.

Images