SAR image MTF quality evaluation method and system based on natural scene edge

Abstract

The invention discloses an SAR (Synthetic Aperture Radar) image MTF (Modulation Transfer Function) quality evaluation method and system based on a natural scene edge, and the method comprises the steps: building a selection standard system for automatic edge screening of a natural scene SAR image, and determining a high-contrast linear edge according to the standard system; automatically detecting a high-contrast linear edge in the natural scene, and intercepting a region of interest for subsequent calculation based on the position of the linear edge; gray level distribution is extracted in the normal direction of the linear edge, and an edge spread function is constructed and used for describing gray level transition response of the imaging system in the edge area; carrying out normalization and smoothing processing on the edge spread function, and carrying out spatial domain first-order differential calculation to obtain a line spread function; and performing fast Fourier transform on the line spread function to obtain a modulation degree transfer function curve, and representing the detail transfer capability of the system under different spatial frequencies. According to the method, the imaging quality of the SAR image of the natural scene can be automatically, accurately and quantitatively evaluated under the condition that manual field arrangement is not needed.

Description

Technical Field

[0001] This invention relates to the field of SAR imagery, and more particularly to a method and system for evaluating the MTF quality of SAR images based on natural scene edges. Background Technology

[0002] Synthetic Aperture Radar (SAR), as an active microwave imaging system, can acquire surface information under all-weather and all-day conditions, and is widely used in fields such as land surveys, disaster monitoring, and military reconnaissance. In practical applications, the quality of SAR images directly affects the reliability of subsequent detection, identification, and interpretation tasks; therefore, rapid and accurate evaluation of SAR imaging quality is of great significance. However, imaging performance in complex natural scenes is affected by multiple factors such as system resolution, number of views, and noise suppression strategies, resulting in variations in image sharpness and detail fidelity. Therefore, a universally applicable quantitative evaluation mechanism is urgently needed.

[0003] The modulation transfer function (MTF) characterizes the contrast preservation capability of an imaging system at different spatial frequencies and is an important indicator for measuring image sharpness and detail transfer performance. Current MTF measurement methods still mainly rely on artificially deployed corner reflectors and other strong scattering point targets or specially constructed multi-stripe test ranges to derive the system's frequency characteristics through high contrast response. However, point target measurements are difficult to reflect the structural details of complex targets, and deploying test ranges is time-consuming, costly, and limited in applicable scenarios, making it difficult to directly use for quality evaluation of SAR images of natural scenes.

[0004] In recent years, researchers have gradually discovered that stable, high-contrast linear structures such as road and building edges are commonly found in natural scene features. Their grayscale transitions contain rich spatial frequency information and can replace artificial targets as feature carriers for MTF derivation. However, traditional natural scene edge MTF measurement still relies on manual selection, which is highly subjective, has low automation, and is sensitive to noise and geometric distortion, exhibiting problems such as unrobust edge detection and unstable response extraction. Furthermore, inconsistent edge quality can easily introduce curve oscillations, affecting the reliability of the evaluation.

[0005] In summary, to overcome the limitations of manual intervention and scene constraints and achieve quality assessment of SAR images of natural scenes, an MTF estimation method with automatic edge filtering, robust response modeling, and reliable frequency characteristic derivation capabilities is needed to improve the universality, automation level, and engineering usability of imaging quality assessment. Summary of the Invention

[0006] This invention provides a method and system for evaluating the MTF quality of SAR images based on natural scene edges. This invention enables automated, accurate, and quantitative evaluation of the imaging quality of SAR images of natural scenes without the need for manual scene setup. Details are described below:

[0007] Firstly, a method for evaluating the MTF quality of SAR images based on natural scene edges, the method comprising:

[0008] A selection standard system for automatic edge screening of SAR images of natural scenes is established, and high-contrast linear edges are determined based on the standard system.

[0009] Automatically detect high-contrast linear edges in natural scenes and extract regions of interest for subsequent calculations based on the position of the linear edges;

[0010] The gray-level distribution is extracted along the normal direction of the linear edge and an edge diffusion function is constructed to describe the gray-level transition response of the imaging system in the edge region. The edge diffusion function is normalized and smoothed, and the line diffusion function is obtained by calculating the first-order derivative in the spatial domain.

[0011] Perform a Fast Fourier Transform on the line spread function to obtain the modulation transfer function curve, which characterizes the system's detail transfer capability at different spatial frequencies.

[0012] The selection criteria system for the automatic edge filtering is as follows:

[0013] The outline of the main body at the edge is continuous and straight without jagged edges or irregular deformation;

[0014] The edge grayscale transition is clear and steep, and the contrast between bright and dark areas is significant, meaning the contrast between bright and dark areas meets the requirements. ≥0.20, where and These are the median brightness values ​​of the local pixel sets on both sides of the edge normal direction after linear normalization.

[0015] The backscattering characteristics of the two edge regions are uniform and consistent, but there are differences in brightness;

[0016] The background area is smooth and clean, without trees, shadows, or dense urban textures;

[0017] The edge direction is consistent with or deviates from the SAR main imaging direction within a preset range, with a deviation angle not exceeding 15°; the edge has continuity, maintaining structural continuity and integrity within at least 10 consecutive pixels, and meets the requirements of a linear index. ≥ ,in The mean square error is the result of fitting a straight line to the edge line segment using the least squares method. To preset the straightness threshold, ideally... It should be greater than 0.85.

[0018] The brightness difference is as follows:

[0019] Based on local regions of SAR images, the gray values ​​of the pixel sets on both sides of the edge normal are uniformly and linearly normalized to... Interval; calculate the median brightness of the normalized pixel sets on both sides respectively. and Brightness difference is defined as ,satisfy This ensures clear transitions in grayscale at the edges.

[0020] The automatic detection of high-contrast linear edges in natural scenes, and the extraction of regions of interest for subsequent calculations based on the positions of these linear edges, includes:

[0021] For each candidate linear edge, four indicators are calculated: length, brightness contrast, goodness of fit of the line, and orientation consistency. An edge quality scoring function is then constructed.

[0022] Sort all candidate edges according to the scoring results, and select the linear edge with the highest score that meets the threshold condition as the target edge;

[0023] Based on the position and orientation of the target linear edge, a region of interest of fixed width is automatically extracted along the normal direction of the edge, and the region of interest is rotated or affine corrected so that the edge is aligned with the image coordinate axis direction within the region of interest.

[0024] Specifically, the edge diffusion function curve is smoothed for noise suppression using a moving average or Gaussian filter; interval truncation is performed on the upper and lower platform segments outside the grayscale transition region to retain the response segments where grayscale transitions from dark to bright, in order to reduce error propagation of the edge diffusion curve.

[0025] Among them, the sample indexes in the edge diffusion function curve that are between 5% and 95% grayscale are considered as valid response segments.

[0026] In a second aspect, a SAR image MTF quality assessment device based on natural scene edges, the device comprising: a processor and a memory, the memory storing program instructions, the processor calling the program instructions stored in the memory to cause the device to perform the method described in any one of the first aspects.

[0027] Third aspect, a computer-readable storage medium storing a computer program, the computer program including program instructions that, when executed by a processor, cause the processor to perform the method described in any one of the first aspects.

[0028] The beneficial effects of the technical solution provided by this invention are:

[0029] 1. This invention eliminates the need for artificial standard targets such as corner reflectors or stripe ranges to directly acquire high-contrast linear edge structures for spatial frequency response measurement from natural scene SAR images. This method overcomes the limitations of imaging scenes and deployment conditions, and can stably carry out imaging quality evaluation in on-orbit operation and complex natural environments, thereby significantly improving the universality and engineering practical value of the method.

[0030] 2. High-quality linear edges are difficult to obtain in non-target natural scenes. Most edges in the scene do not have monotonic gray-level transition characteristics that stably reflect the modulation characteristics of the system. If they are directly used for quality evaluation, it will lead to unstable gray-level projection and incomparable frequency domain results. In particular, the coherent speckles that are common in synthetic aperture radar images further cause the edge gray-level to fluctuate randomly, making it difficult to guarantee the clarity of the boundary and the consistency of the direction. Directly applying the natural edge evaluation method in the field of optics is not only difficult to stably obtain the measurement edge that meets the requirements, but also easy to introduce errors such as instability of the spread function and frequency domain energy leakage in subsequent calculations, making it difficult to obtain comparable modulation transfer function results.

[0031] To address the aforementioned challenges, this invention, taking into account the random gray-level fluctuations caused by coherent speckles in SAR images, constructs an edge filtering rule and directional consistency constraint system for SAR images. This system enables automatic filtering of high-quality natural edges, retaining only edge regions with good straightness, stable gray-level transitions, and orientations matching the main imaging geometry. It effectively eliminates interfering background regions such as trees, shadows, and urban textures, providing reliable input for subsequent gray-level projection and frequency domain analysis. This enhances the realism and stability of edge gray-level transition characteristics and improves the measurement accuracy of MTF.

[0032] 3. This invention employs a detail transfer modeling strategy that combines ESF (Edge Spread Function) construction, LSF (Line Spread Function) extraction, and Fast Fourier Transform. It constructs a stable edge spread function through projection accumulation and median statistics, and incorporates a window function in the frequency domain calculation stage to reduce sidelobe energy interference. This approach can fully characterize the response attenuation law of the imaging system from the spatial domain to the frequency domain, and can directly obtain key indicators such as MTF50, MTF30, and MTF10, which can be used to distinguish the detail preservation ability and effective resolution of different imaging systems. The quantitative evaluation results are more scientific and comparable.

[0033] 4. This invention is highly automated and robust, and can be adapted to multiple imaging platforms, multiple resolutions and multiple observation scenarios. It can be widely used in the development, delivery and acceptance and long-term quality monitoring processes of SAR image processing systems, providing strong support for improving imaging quality and evaluating system performance. Attached Figure Description

[0034] Figure 1 The flowchart shows a high-resolution SAR image MTF quality assessment method based on natural scene edges.

[0035] Figure 2 A schematic diagram of the components of a high-resolution SAR image MTF quality assessment system based on natural scene edges;

[0036] Figure 3 This is a flowchart of edge detection and ROI extraction in an application example;

[0037] Figure 4 This is a graph showing the MTF evaluation curve for the edges of a SAR image of a natural scene. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below.

[0039] In existing technologies, when assessing the imaging quality of high-resolution SAR images, traditional MTF measurement methods are insufficiently applicable in large-scale natural scenes and on-orbit operating conditions due to their reliance on manually deployed standard targets such as corner reflectors or stripe ranges. On the one hand, the placement of artificial targets is limited by imaging timing, geographical environment, and cost, making it difficult to conduct high-frequency quality monitoring in engineering practice. On the other hand, edge measurement methods for natural scenes often rely on manual selection or simple threshold screening, resulting in problems such as tilt distortion, texture interference, and grayscale discontinuities in the extracted edges, affecting the accuracy of ESF and MTF calculations.

[0040] Example 1

[0041] To address the aforementioned issues, this invention provides a SAR image MTF quality evaluation method based on natural scene edges. It achieves high-quality automatic extraction of natural edges by introducing an edge screening standard system, constructs an ESF (Extended Sequence Filter) using grayscale projection, obtains the LSF (Local Sequence Filter) through spatial domain differentiation, and finally transforms it to the frequency domain to form an MTF curve, thereby comprehensively describing the spatial frequency response characteristics of the imaging system. This invention requires no artificial standard targets and can be directly applied to SAR images of natural scenes. It exhibits high automation, strong robustness, and wide applicability in complex natural environments and on-orbit scenarios, effectively supporting the quality assessment and long-term performance monitoring of high-resolution SAR imaging systems.

[0042] Specifically, such as Figure 1 As shown, the SAR image MTF quality assessment method based on natural scene edges provided in this embodiment may include:

[0043] S1: Establish a selection standard system for automatic edge screening of SAR images of natural scenes, and determine high-contrast linear edges based on this standard system. The automatic edge screening standard system stipulates that high-contrast linear edges must simultaneously meet the following criteria:

[0044] (1) The edge body has a good linear structure, and its outline is continuous and straight with basically no jagged edges or irregular deformation;

[0045] (2) The grayscale transition at the edge is clear and steep, the contrast between the bright and dark areas is significant, and the ideal edge width is less than one pixel;

[0046] (3) The backscattering characteristics of the two sides of the edge are uniform and consistent, and there is a sufficient brightness difference.

[0047] Ideally, based on a local region of a SAR image, the gray values ​​of the pixel sets on both sides of the edge normal are uniformly and linearly normalized to... The interval is then used to calculate the median brightness of the normalized pixel sets on both sides. and The difference in brightness is defined as Ideally, it should meet the following requirements. This ensures that the grayscale transitions at the edges are sufficiently clear.

[0048] (4) The background area is smooth and clean, without interfering targets such as trees, shadows, and dense urban textures;

[0049] (5) The edge direction is preferably consistent with the SAR main imaging direction or the deviation is within the allowable range, with an ideal deviation angle of no more than 15°, in order to improve the stability of the measurement results;

[0050] (6) The edges have sufficient continuity, that is, the structure remains continuous and complete within a range of at least 10 consecutive pixels, and there are no obvious bends or jagged edges.

[0051] S2: Automatically detects high-contrast linear edges in natural scenes and extracts regions of interest (ROIs) based on the location of the linear edges for subsequent calculations;

[0052] S3: Extract the gray-level distribution along the normal direction of the linear edge and construct the edge spread function (ESF) to describe the gray-level transition response of the imaging system in the edge region;

[0053] S4: Normalize and smooth the edge diffusion function ESF, and obtain the line diffusion function LSF by calculating the first-order differential in the spatial domain;

[0054] S5: Perform a Fast Fourier Transform (FFT) on the Line Spread Function (LSF) to obtain the Modulation Transfer Function (MTF) curve, which characterizes the system's detail transfer capability at different spatial frequencies.

[0055] Example 2

[0056] The following section, using specific examples and calculation formulas, further illustrates the scheme in Example 1. See the description below for details:

[0057] In some preferred embodiments, step S2 above, which automatically detects high-contrast linear edges in a natural scene and extracts a region of interest (ROI) based on the linear edge location for subsequent calculations, may further include:

[0058] S21: Preprocess the input SAR image, including: performing normalization, contrast enhancement and optional smoothing filtering operations on the image to enhance the saliency of edge structures and suppress speckle noise interference, thereby improving the stability of the subsequent edge detection stage.

[0059] S22: Edge detection is performed on the preprocessed image using the gradient-based Canny operator. It is preferable to use high and low thresholds dynamically set based on gray-level histogram statistics to enhance the adaptive capability of edge detection, expressed as:

[0060]

[0061] In the formula, Indicates the input image; For dynamic thresholds; This represents the initial edge detection result.

[0062] S23: Line segment detection is performed on the initial edge map based on probabilistic Hough transform to obtain a set of candidate edges with continuity and directional consistency.

[0063]

[0064] In the formula, The minimum line segment length, The maximum interruption gap is used to constrain edge integrity. The Hough Transform is used to quickly detect continuous long straight line segments by randomly sampling and accumulating the parameters in the edge map, and outputs a set of candidate edges that meet the length and continuity constraints.

[0065] S24: To improve the quality of candidate edges, four indicators are calculated for each candidate linear edge: length, brightness contrast, goodness of fit of the straight line, and orientation consistency. An edge quality scoring function is then constructed.

[0066]

[0067] All indicators are normalized to interval, The weight can be set according to the platform scenario.

[0068] Length indicator:

[0069]

[0070] Brightness contrast index:

[0071]

[0072] in, and These represent the grayscale statistical values ​​of equidistant samples on the left and right sides of the edge, respectively. This is a median operation used to sort a set of pixels and take the median value to provide more stable brightness statistics than the mean.

[0073] Linear indicator:

[0074]

[0075] in, This represents the least squares fitting error.

[0076] Directional consistency index:

[0077]

[0078] in, The reference value for the angle in the main imaging direction. The direction angle, derived from the endpoint positions of the candidate edge segments, is calculated using the following formula:

[0079]

[0080] in: and The pixel coordinates of the two endpoints of the candidate edge segment in the image coordinate system.

[0081] S25: Sort all candidate edges according to the scoring results, select the linear edge with the highest score and that meets the threshold condition as the target edge, and ensure that it has high contrast, continuity and geometric stability.

[0082] S26: Based on the position and direction of the target linear edge, automatically extract a region of interest (ROI) of fixed width along the normal direction of the edge, and perform rotation or affine correction on the ROI region to ensure that the edge is strictly aligned with the image coordinate axis direction within the ROI, thereby avoiding grayscale projection deviation caused by edge tilt.

[0083] Specifically, let the two endpoints of the target edge be:

[0084]

[0085] Then its normalized tangent vector can be defined. With normal vector for:

[0086]

[0087] in:

[0088]

[0089] Then take the midpoint of the edge. As the origin of ROI:

[0090]

[0091] in, The x-coordinate of the midpoint of the edge line segment is . The ordinate is the y-coordinate of the midpoint of the edge line segment.

[0092] by Centered on the original image, perform an affine coordinate transformation to rotate the edge direction to the horizontal axis direction. The transformation can be expressed as:

[0093]

[0094] in, The output sampling coordinates within the ROI. The coordinates in the SAR image , Let be the unit normal vector component along the edge direction. The final ROI constraint can be written as:

[0095]

[0096] in, These are the geometrically corrected image pixel values. This represents the sampling length of the ROI in the edge direction. This represents the sampling width of the ROI in the normal direction.

[0097] By using orientation correction and region clipping, the grayscale transitions in the ROI have a strictly consistent geometric reference frame, providing high-quality input for subsequent ESF construction.

[0098] In some preferred embodiments, step S3 above, which involves extracting the grayscale distribution along the normal direction of the linear edge and constructing an edge diffusion function (ESF) to describe the grayscale transition response of the imaging system in the edge region, may further include:

[0099] S31: Using the linear edge of the target in the ROI after direction correction as a reference, determine the normal direction sampling axis of the edge so that all projected samples are strictly perpendicular to the edge structure distribution, avoiding grayscale mixing errors caused by geometric skew.

[0100] S32: Map the position of each pixel in the ROI to a one-dimensional coordinate system along the normal direction, and statistically accumulate the pixel gray level based on the coordinate to obtain the gray level variation distribution along the normal direction;

[0101] Among them, the median statistical method is preferred to suppress extreme value noise interference.

[0102] S33: To improve spatial sampling accuracy, the extracted one-dimensional grayscale sequence is resampled and sorted to ensure consistent sampling intervals, thereby constructing a continuous and comparable edge spread function (ESF). The ESF can be expressed as:

[0103]

[0104] In the formula: For the first The grayscale value at each sampling location, Let be the set of pixels projected onto the same normal position. This represents the pixel grayscale value in the ROI.

[0105] S34: Normalize the constructed ESF curve to uniformly map the grayscale range to... This ensures that the ESF has a consistent dynamic range representation capability under different scenarios and brightness conditions.

[0106] S35: Perform noise-suppressing smoothing on the ESF curve, preferably using moving average or Gaussian filtering, to reduce the impact of speckle noise on the grayscale transition trend and enhance the expression of actual edge response characteristics.

[0107] In some preferred embodiments, step S4 above, which normalizes and smooths the edge diffusion function ESF and obtains the line diffusion function LSF through first-order differential calculation in the spatial domain, may further include:

[0108] S41: Perform amplitude normalization on the ESF curve, uniformly mapping the curve's grayscale changes to... The interval ensures that the ESF representation under different image conditions has a consistent brightness dynamic range, reducing the impact of imaging amplitude differences on subsequent modeling. The normalized form can be expressed as:

[0109]

[0110] in, The edge spread function (ESF) represents the maximum gray value within the sampling interval. This represents the minimum gray value of the edge spread function (ESF) within the sampling interval.

[0111] S42: Smooth the normalized ESF curve to suppress noise fluctuations. Moving average or Gaussian smoothing is preferred to make the grayscale transition smoother and enhance the curve's ability to represent the true edge response. The smoothed curve is denoted as... ;

[0112] S43: Perform interval truncation on the upper and lower platform segments outside the grayscale transition area, retaining only the effective response segments where the grayscale rapidly transitions from dark to bright, in order to reduce the propagation of edge diffusion curve errors caused by edge background fluctuations. The sample index in the ESF curve within the grayscale range of 5% to 95% can be selected as the effective response segment.

[0113] S44: Perform a first-order spatial domain differential operation on the smoothed ESF curve to obtain the line spread function (LSF), which is used to quantitatively characterize the gray-level transition velocity at the edges and the response characteristics of the imaging system to high-frequency details. The discrete gradient estimation is preferably implemented using the central difference form, as shown in the following expression:

[0114]

[0115] in, For the first Line spread function values ​​at each sampling location, These are the smoothed ESF sample values. The sampling interval is along the edge normal direction.

[0116] Through the above differential operation, the location with the largest gradient change in ESF can be effectively highlighted, so that the imaging system's ability to respond to edge details can be accurately expressed.

[0117] In some preferred embodiments, step S5 above, which involves performing a Fast Fourier Transform (FFT) on the line spread function (LSF) to obtain the modulation transfer function (MTF) curve, characterizing the system's detail transfer capability at different spatial frequencies, may further include:

[0118] S51: To reduce spectral leakage caused by the finite length of the LSF sequence, it is preferable to use a weighted window function such as the Hanning window to window the LSF, and to enhance the frequency domain sampling accuracy through zero padding in order to obtain a smooth and continuous frequency response curve.

[0119] S52: Perform a Fast Fourier Transform (FFT) on the processed LSF to obtain the amplitude response of the imaging system in the spatial frequency domain. Its mathematical expression preferably takes the following form:

[0120]

[0121] in, express points, Represents a discrete linear diffusion function sequence. Represents pixel-domain spatial frequency. This indicates that a complex number is assigned, where n is the discrete sampling index, representing the nth sampling position of the LSF sequence.

[0122] S53: Normalize the obtained amplitude-frequency response so that the zero-frequency response equals 1, in order to enable comparison between different SAR systems:

[0123]

[0124] in, This is the amplitude-frequency response value at a frequency of zero. For spatial frequency equal to The value of the time-adjustment system transfer function.

[0125] S54: Based on the normalized MTF curve, three key feature frequencies—MTF50, MTF30, and MTF10—are extracted to quantitatively reflect the spatial resolution performance and detail preservation capability of the SAR imaging system. The specific calculation method is as follows:

[0126]

[0127] in, For the MTF curve at an amplitude equal to The corresponding spatial frequency, MTF represents the MTF amplitude level. MTF50 indicates the effective resolution corresponding to perceived sharpness, MTF30 corresponds to the lower limit of recognizable detail structure, and MTF10 corresponds to the system's ability to transmit the limit of high-frequency information. The higher the characteristic frequency, the stronger the imaging system's ability to preserve details at the corresponding spatial frequency, thus quantitatively reflecting the image's edge sharpness and texture fidelity.

[0128] In summary, the high-resolution SAR image MTF quality assessment method based on natural scene edges provided by the above embodiments of the present invention overcomes the limitations of traditional imaging quality assessment methods that rely on artificial targets and ideal scene conditions. It has the following technical advantages:

[0129] For a given SAR image of a natural scene, the proportion of measurable edges that meet the requirements of good linearity, clear gray-level transitions, and direction consistent with the main imaging direction is extremely low. In particular, if the edge direction deviates from the main imaging direction, it will lead to the accumulation of errors such as unstable normal projection, gray-level transition broadening, and frequency domain leakage, making the edges unable to reflect the true modulation characteristics of the system. Based on this, the embodiments of the present invention first construct an automatic edge screening standard system for real ground conditions. The candidate edges are evaluated from multiple dimensions such as linear structure integrity, gray-level contrast, backscattering consistency, background purity, main imaging direction consistency, and spatial continuity. Interference areas caused by texture noise, ground debris, and geometric distortion are automatically eliminated, thereby ensuring that the selected edges have high quality and high measurability, laying a stable foundation for subsequent accurate MTF calculation.

[0130] SAR images contain coherent speckles, resulting in random grayscale fluctuations. Direct edge detection often obscures true physical boundaries with noise and texture, leading to the loss of meaningful linear edges and the detection of numerous false edges. To address this, this invention first performs preprocessing operations on the input SAR image, including normalization, contrast enhancement, and optional smoothing filtering, to enhance the saliency of true edges and suppress speckle noise interference. Then, a multi-stage edge detection and quality scoring mechanism is introduced. Candidate linear edges are extracted by combining gradient edge recognition and probabilistic Hough line fitting. A joint scoring model is constructed using length, brightness gradient, structural straightness, and imaging direction consistency to achieve reliable discrimination and automatic acquisition of target edges. Combined with a geometric correction strategy, the target edges are aligned in the local coordinate system, effectively avoiding the accumulation of projection errors caused by edge tilt and improving the spatial consistency of edge grayscale transitions.

[0131] To accurately characterize the point spread effect generated by the imaging chain in the edge region, this embodiment of the invention extracts the gray-level distribution along the edge normal direction, constructs the edge spread function (ESF) using a noise suppression statistical method, and performs normalization and smoothing processing to reduce the impact of SAR speckle noise on the gray-level transition expression. By differentiating the first order of the ESF, the line spread function (LSF) is obtained, highlighting the actual response of the imaging system to high-frequency information in the edge transition region, thus accurately quantifying the diffusion range of detailed features.

[0132] Furthermore, a Fast Fourier Transform (FFT) is performed on the LSF to obtain the Modulation Transfer Function (MTF) curve, which comprehensively reflects the system's detail preservation capability at different spatial frequencies. Since finite-length LSF sequences are prone to spectral leakage, sidelobe energy diffusion, and insufficient sampling resolution when directly performing FFT, leading to unstable frequency domain estimation, this embodiment of the invention introduces windowing and zero-padding strategies before FFT calculation to alleviate spectral leakage and discrete sampling errors, ensuring that the MTF curve is smooth, continuous, and comparable. Typical performance indicators such as MTF50, MTF30, and MTF10 are extracted to achieve quantitative evaluation of imaging resolution and texture fidelity. Through collaborative optimization of multiple stages including feature selection, geometric correction, noise suppression, response modeling, and frequency domain solution, this embodiment of the invention achieves autonomous assessment of imaging quality under real natural scene conditions without artificial targets. This method can accurately quantify the spatial frequency transfer capability of the SAR system and thus evaluate SAR image quality without the need for specially designed targets or additional prior information.

[0133] Example 3

[0134] Based on the same inventive concept, embodiments of the present invention also provide a high-resolution SAR image MTF quality evaluation system based on natural scene edges.

[0135] Specifically, such as Figure 2 As shown, the high-resolution SAR image MTF quality assessment system based on natural scene edges provided in this embodiment may include:

[0136] Image input module, used to import and preprocess SAR image data;

[0137] The edge filtering module is used to filter high-contrast linear edges in natural scenes that meet the requirements of linearity, contrast, and continuity according to a preset edge selection standard system.

[0138] The edge detection and ROI extraction module is used to automatically detect high-contrast linear edges and extract the region of interest (ROI) for analysis based on the edge location.

[0139] The ESF building module is used to project and uniformly resample pixels within the ROI along the target edge normal direction to obtain the normalized edge spread function ESF.

[0140] The LSF acquisition module is used to perform spatial domain discretization of the edge spread function (ESF) and extract the effective response segment to obtain the line spread function (LSF) that describes the edge detail response capability of the imaging system.

[0141] The MTF analysis module is used to perform a Fast Fourier Transform (FFT) on the Line Spread Function (LSF) and generate a Modulation Transfer Function (MTF) curve that varies with spatial frequency based on its amplitude response.

[0142] The performance evaluation module is used to calculate multiple spatial frequency characteristic indicators, including MTF50, MTF30 and MTF10, based on MTF curves, in order to quantify the spatial resolution and detail preservation capability of the SAR imaging system.

[0143] The following provides a more detailed description of the specific contents of each functional module constituting the SAR image MTF quality evaluation system of the above embodiments of the present invention.

[0144] Edge filtering module: Used to filter high-quality linear edges from natural scene SAR images according to an automatic edge filtering standard system, which meet the requirements of linear structure, contrast prominence, and spatial continuity. Its filtering rules include:

[0145] (1) The edge body has a good linear structure, and its outline is continuous and straight with basically no jagged edges or irregular deformation;

[0146] (2) The grayscale transition at the edge is clear and steep, the contrast between the bright and dark areas is significant, and the ideal edge width is less than one pixel;

[0147] (3) The backscattering characteristics of the two sides of the edge are uniform and consistent, and there is a sufficient brightness difference;

[0148] (4) The background area is smooth and clean, without interfering targets such as trees, shadows, and dense urban textures;

[0149] (5) The edge direction is preferably consistent with the SAR main imaging direction or the deviation is within the allowable range to improve the stability of the measurement results;

[0150] (6) The edges have sufficient continuity and their spatial extension length exceeds ten pixels to ensure sufficient grayscale sampling.

[0151] Edge detection and ROI extraction module: used to perform edge detection and geometric correction on the input SAR image, and extract the region of interest (ROI) based on the target edge location for subsequent MTF estimation.

[0152] Further, this includes preprocessing the input SAR image, including performing normalization, contrast enhancement, and optional smoothing filtering operations to enhance the saliency of edge structures and suppress speckle noise interference, thereby improving the stability of subsequent edge detection stages. Edge detection is then performed on the preprocessed image using a gradient-based Canny operator, preferably employing high and low thresholds dynamically set based on gray-level histogram statistics to enhance the adaptive capability of edge detection, expressed as:

[0153]

[0154] In the formula, Indicates the input image; For dynamic thresholds; This represents the initial edge detection result.

[0155] Line segment detection is performed on the initial edge map based on probabilistic Hough transform to obtain a set of candidate edges with continuity and directional consistency:

[0156]

[0157] In the formula, The minimum line segment length, This represents the maximum interruption gap, used to constrain edge integrity.

[0158] To improve the quality of candidate edges, four indicators—length, brightness contrast, goodness of fit, and orientation consistency—are calculated for each candidate linear edge, and an edge quality scoring function is constructed:

[0159]

[0160] All indicators are normalized to interval, The weight can be set according to the platform scenario.

[0161] Length indicator:

[0162]

[0163] Brightness contrast index:

[0164]

[0165] in, and These represent the grayscale statistical values ​​of equidistant samples on the left and right sides of the edge, respectively.

[0166] Linear indicator:

[0167]

[0168] in, This represents the least squares fitting error.

[0169] Directional consistency index:

[0170]

[0171] in, The reference value for the angle of the main imaging direction.

[0172] All candidate edges are sorted according to the scoring results, and the linear edge with the highest score that meets the threshold condition is selected as the target edge to ensure that it has high contrast, continuity and geometric stability.

[0173] Based on the position and orientation of the target linear edge, a region of interest (ROI) of fixed width is automatically extracted along the normal direction of the edge, and the ROI is rotated or affine corrected to ensure that the edge is strictly aligned with the image coordinate axis direction within the ROI, thereby avoiding grayscale projection deviation caused by edge tilt.

[0174] Specifically, assuming the edge center point has been determined and its direction vector, then with A local coordinate system is constructed based on this reference. The edges are rotated to a horizontal alignment using an affine transformation, and pixels within the following range are selected as the Region of Interest (ROI):

[0175]

[0176] in, These are the geometrically corrected image pixel values. This indicates that the ROI is at the edge. This represents the sampling length of the ROI in the edge direction. This represents the sampling width of the ROI in the normal direction.

[0177] By using orientation correction and region clipping, the grayscale transitions in the ROI have a strictly consistent geometric reference frame, providing high-quality input for subsequent ESF construction.

[0178] ESF building block: Used to extract grayscale distribution along the normal direction of the target edge and construct the edge spread function (ESF) to describe the grayscale transition response of the imaging system in the edge region.

[0179] Further includes:

[0180] Using the linear edge of the target in the ROI after direction correction as a reference, the normal direction sampling axis of the edge is determined so that all projected samples are strictly perpendicular to the edge structure distribution, avoiding grayscale mixing errors caused by geometric skew.

[0181] The pixel positions within the ROI are mapped to a one-dimensional coordinate system along the normal direction. Based on this coordinate system, the pixel gray levels are statistically accumulated to obtain the gray level distribution along the normal direction. Median statistics are preferably used to suppress extreme noise interference.

[0182] To improve spatial sampling accuracy, the extracted one-dimensional grayscale sequence is resampled and sorted to ensure consistent sampling intervals, thereby constructing a continuous and comparable edge spread function (ESF). The ESF can be expressed as:

[0183]

[0184] In the formula: For the first The grayscale value at each sampling location, Let be the set of pixels projected onto the same normal position. This represents the pixel grayscale value in the ROI.

[0185] The constructed ESF curve is normalized to ensure that the grayscale range is uniformly mapped to... This ensures that the ESF has a consistent dynamic range representation capability under different scenarios and brightness conditions.

[0186] The ESF curve is subjected to noise-suppressing smoothing, preferably using moving average or Gaussian filtering, to reduce the impact of speckle noise on the grayscale transition trend and enhance the expression of actual edge response characteristics.

[0187] The LSF acquisition module is used to normalize and smooth the ESF and perform first-order spatial domain differential calculations to obtain the line spread function (LSF). It further includes:

[0188] The ESF curve is normalized to uniformly map the curve's grayscale changes to... The interval ensures that the ESF representation under different image conditions has a consistent brightness dynamic range, reducing the impact of imaging amplitude differences on subsequent modeling. The normalized form can be expressed as:

[0189]

[0190] The normalized ESF curve is smoothed by filtering to suppress noise fluctuations. Moving average or Gaussian smoothing is preferred to make the grayscale transition smoother and enhance the curve's ability to represent the true edge response. The smoothed curve is denoted as... .

[0191] For the upper and lower platform segments outside the grayscale transition area, interval truncation is performed, retaining only the effective response segments where grayscale rapidly transitions from dark to bright, in order to reduce the propagation of edge diffusion curve errors caused by edge background fluctuations. The selection of this effective segment can preferably be set to the ESF index sampling position within the grayscale range of 5% to 95%.

[0192] The smoothed ESF curve is subjected to a first-order spatial domain differential operation to obtain the line spread function (LSF), which is used to quantitatively characterize the gray-level transition velocity at the edges and the response characteristics of the imaging system to high-frequency details. The discrete gradient estimation is preferably implemented using a central difference form, as shown in the following expression:

[0193]

[0194] in, For the first Line spread function values ​​at each sampling location, These are the smoothed ESF sample values. The sampling interval along the edge normal direction

[0195] Through the above differential operation, the location with the largest gradient change in ESF can be effectively highlighted, so that the imaging system's ability to respond to edge details can be accurately expressed.

[0196] The MTF analysis module performs a Fast Fourier Transform (FFT) on the LSF to obtain the modulation transfer function (MTF) curve and extracts resolution evaluation metrics. Further features include:

[0197] To reduce spectral leakage caused by the finite length of LSF sequences, it is preferable to use weighted window functions such as the Hanning window to window the LSF, and to enhance the frequency domain sampling accuracy through zero padding, so as to obtain a smooth and continuous frequency response curve.

[0198] Perform a Fast Fourier Transform (FFT) on the processed LSF to obtain the amplitude response of the imaging system in the spatial frequency domain. Its mathematical expression preferably takes the following form:

[0199]

[0200] in express points, Represents a discrete linear diffusion function sequence. Represents pixel-domain spatial frequency. This indicates assigning a complex number.

[0201] The obtained amplitude-frequency response is normalized so that the zero-frequency response equals 1, in order to enable comparison between different SAR systems:

[0202]

[0203] Based on the normalized MTF curve, three key characteristic frequencies—MTF50, MTF30, and MTF10—are extracted to quantitatively reflect the spatial resolution performance and detail preservation capability of the SAR imaging system. The specific calculation method is as follows:

[0204]

[0205] MTF50 represents the effective resolution corresponding to perceived sharpness, MTF30 corresponds to the lower limit of recognizable detail structure, and MTF10 corresponds to the system's ability to transmit the limit of high-frequency information. The higher the characteristic frequency, the stronger the imaging system's ability to preserve details at the corresponding spatial frequency, thus quantitatively reflecting the image's edge sharpness and texture fidelity.

[0206] Example 4

[0207] The technical effects of the technical solutions provided in the embodiments of the present invention will be evaluated below with reference to specific verification examples.

[0208] In this specific verification example:

[0209] I. Data composition, including:

[0210] The data used in this embodiment of the invention comes from high-resolution SAR images acquired by the TerraSAR-X satellite, employing a HH (horizontal transmit-horizontal receive polarization) single-polarization observation method, covering typical natural scenes such as roads, buildings, and bare land. All data have undergone radiometric calibration and geometric correction, exhibiting high geometric accuracy and radiometric consistency, and providing clear and representative linear edge structures even without artificial targets.

[0211] II. Evaluation criteria, including:

[0212] To evaluate the image quality of a SAR imaging system, this embodiment of the invention uses the modulation transfer function (MTF) curve as the evaluation criterion. By comparing the attenuation trend and high-frequency response retention of the MTF curves corresponding to different images across the entire frequency range, the system's ability to transmit spatial details is determined. The smoother the curve and the gentler the descent, the higher the image sharpness and the stronger the detail fidelity, thus more objectively reflecting the differences in overall imaging quality.

[0213] III. Experimental Results:

[0214] As shown in Table 1, this table presents the imaging parameter configurations corresponding to the three sets of experimental data in Figure 1. Ground distance resolution measures the system's ability to distinguish adjacent scatterers in the distance direction; azimuth resolution reflects the ability to resolve details along the platform's flight direction; distance sight number is used to suppress speckle noise, and a higher value results in a more significant averaging effect, but may introduce detail smoothing. These parameters, in different combinations, directly affect image sharpness and texture fidelity, thus determining the overall image quality.

[0215] Table 1. Parameter configurations for three SAR imaging systems

[0216]

[0217] The MTF curve comparison results show that the dashed line group has the smallest ground distance and azimuth resolution, and uses a lower number of views. Therefore, it maintains a stronger modulation response in the high-frequency region, has the highest edge detail retention, and its image sharpness is better than other groups. The solid line group achieves a balance between resolution and number of views, with moderate MTF attenuation and moderate image sharpness. In contrast, the dotted line group uses a larger number of views and a lower resolution. Although it has a stronger noise suppression effect, its high-frequency response is significantly attenuated, its image detail retention is insufficient, and its sharpness is weaker.

[0218] The above results demonstrate that the variation law of the MTF curve obtained in this embodiment of the invention is completely consistent with the expected system resolution and number of looks parameters. More importantly, this method successfully expands the application boundary of modulation transfer function (MTF) in SAR image evaluation. Traditionally, due to the influence of differences in scattering mechanisms and multiplicative modulation of speckle noise on natural scene SAR images, MTF evaluation usually relies on specially deployed test targets. However, this embodiment of the invention can stably obtain the MTF curve using only the linear edges in the natural scene without relying on any artificial field setup, and its high-frequency attenuation trend has a clear correspondence with the system parameters. This proves that this method successfully applies the MTF evaluation system, which was originally limited to controlled conditions, directly to natural SAR scenes, realizing a more convenient and widely applicable real-world measurement of imaging quality.

[0219] In summary, the embodiments of the present invention can effectively present the differences in imaging quality between different systems. The MTF curve performance is highly consistent with the system resolution and number of views settings, verifying the accuracy and discrimination ability of the MTF estimation method based on natural scene edges under real data conditions.

[0220] This invention also provides a computer terminal, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it can be used to execute any of the methods in the above embodiments of this invention, or to run any of the systems in the above embodiments of this invention.

[0221] Optionally, the memory is used to store programs; the memory may include volatile memory, such as random-access memory (RAM), such as static random-access memory (SRAM), double data rate synchronous dynamic random-access memory (DDR SDRAM), etc.; the memory may also include non-volatile memory, such as flash memory. The memory is used to store computer programs (such as application programs and functional modules that implement the above methods), computer instructions, etc., and the aforementioned computer programs and computer instructions can be partitioned and stored in one or more memories. Furthermore, the aforementioned computer programs, computer instructions, data, etc., can be accessed by the processor.

[0222] A processor is used to execute computer programs stored in memory to implement the various steps of the methods or various modules of the systems involved in the above embodiments. For details, please refer to the relevant descriptions in the preceding method and system embodiments.

[0223] The processor and memory can be independent structures or integrated structures. When the processor and memory are independent structures, they can be coupled together via a bus to transmit data signals. This will not be elaborated further in this embodiment of the invention.

[0224] The execution entity of the aforementioned processor and memory can be a computer terminal device with computing functions, such as a computer, a microcontroller, or a single-chip microcomputer. In specific implementation, the embodiments of the present invention do not limit the execution entity and can select it according to the needs of actual application.

[0225] A computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in or transmitted through a computer-readable storage medium. A computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium can be magnetic or semiconductor, etc.

[0226] An embodiment of the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, can be used to perform the method of any of the above embodiments of the present invention, or to run the system of any of the above embodiments of the present invention.

[0227] Computer-readable media include computer storage media and communication media, wherein communication media include any medium that facilitates the transfer of computer programs from one place to another. Storage media can be any available medium accessible to a general-purpose or special-purpose computer. An exemplary storage medium is coupled to a processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and storage medium can reside in an ASIC. Alternatively, the ASIC can reside in a user device. Of course, the processor and storage medium can also exist as discrete components in a communication device.

[0228] The computer-readable storage medium includes, but is not limited to, flash memory, hard disk, solid-state drive, etc.

[0229] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of the present invention is generated.

[0230] Unless otherwise specified, the model numbers of the various devices in this embodiment of the invention are not limited, and any device that can perform the above functions is acceptable.

[0231] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of a preferred embodiment, and the sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0232] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for evaluating the MTF quality of SAR images based on natural scene edges, characterized in that, The method includes: A selection standard system for automatic edge screening of SAR images of natural scenes is established, and high-contrast linear edges are determined based on the standard system. Automatically detect high-contrast linear edges in natural scenes and extract regions of interest for subsequent calculations based on the position of the linear edges; The gray-level distribution is extracted along the normal direction of the linear edge and an edge diffusion function is constructed to describe the gray-level transition response of the imaging system in the edge region. The edge diffusion function is normalized and smoothed, and the line diffusion function is obtained by calculating the first-order derivative in the spatial domain. Perform a Fast Fourier Transform on the line spread function to obtain the modulation transfer function curve, which characterizes the system's detail transfer capability at different spatial frequencies.

2. The SAR image MTF quality assessment method based on natural scene edges according to claim 1, characterized in that, The selection criteria system for the automatic edge filtering is as follows: The outline of the main body at the edge is continuous and straight without jagged edges or irregular deformation; The edge grayscale transition is clear and steep, and the contrast between bright and dark areas is significant, meaning the contrast between bright and dark areas meets the requirements. ≥0.20, where and These are the median brightness values ​​of the local pixel sets on both sides of the edge normal direction after linear normalization. The backscattering characteristics of the two edge regions are uniform and consistent, but there are differences in brightness; The background area is smooth and clean, without trees, shadows, or dense urban textures; The edge direction is consistent with or deviates from the SAR main imaging direction within a preset range, with a deviation angle not exceeding 15°; the edge has continuity, maintaining structural continuity and integrity within at least 10 consecutive pixels, and meets the requirements of a linear index. ≥ ,in The mean square error is the result of fitting a straight line to the edge line segment using the least squares method. To preset the straightness threshold, ideally... It should be greater than 0.

85.

3. The SAR image MTF quality assessment method based on natural scene edges according to claim 1, characterized in that, The brightness difference is as follows: Based on local regions of SAR images, the gray values ​​of the pixel sets on both sides of the edge normal are uniformly and linearly normalized to... Interval; calculate the median brightness of the normalized pixel sets on both sides respectively. and Brightness difference is defined as ,satisfy This ensures clear transitions in grayscale at the edges.

4. The SAR image MTF quality assessment method based on natural scene edges according to claim 1, characterized in that, The automatic detection of high-contrast linear edges in natural scenes, and the extraction of regions of interest for subsequent calculations based on the positions of these linear edges, includes: For each candidate linear edge, four indicators are calculated: length, brightness contrast, goodness of fit of the line, and orientation consistency. An edge quality scoring function is then constructed. Sort all candidate edges according to the scoring results, and select the linear edge with the highest score that meets the threshold condition as the target edge; Based on the position and orientation of the target linear edge, a region of interest of fixed width is automatically extracted along the normal direction of the edge, and the region of interest is rotated or affine corrected so that the edge is aligned with the image coordinate axis direction within the region of interest.

5. The SAR image MTF quality assessment method based on natural scene edges according to claim 1, characterized in that, The edge diffusion function curve is smoothed for noise suppression using a moving average or Gaussian filter; interval truncation is performed on the upper and lower platform segments outside the grayscale transition region to retain the response segments where the grayscale transitions from dark to bright, in order to reduce the error propagation of the edge diffusion curve.

6. The SAR image MTF quality assessment method based on natural scene edges according to claim 5, characterized in that, The sample indexes in the edge diffusion function curve that fall within the grayscale range of 5% to 95% are considered as valid response segments.

7. A SAR image MTF quality assessment system based on natural scene edges, characterized in that, The system includes: Image input module, used to import and preprocess SAR image data; The edge filtering module is used to filter high-contrast linear edges in natural scenes that meet the requirements of linearity, contrast, and continuity according to a preset edge selection standard system. The edge detection and region of interest extraction module is used to automatically detect high-contrast linear edges and extract regions of interest for analysis based on the edge positions. The ESF building block is used to project and uniformly resample pixels within the region of interest along the normal direction of the target edge to obtain a normalized edge spread function. The edge spread function acquisition module is used to perform spatial domain discretization first-order differentiation on the edge spread function and extract the response segment to obtain the line spread function that describes the edge detail response capability of the imaging system. The MTF analysis module is used to perform a fast Fourier transform on the line spread function and generate a modulation transfer function (MTF) curve that varies with spatial frequency based on its amplitude response. The performance evaluation module is used to calculate multiple spatial frequency characteristic indicators, including MTF50, MTF30 and MTF10, based on MTF curves, in order to quantify the spatial resolution and detail preservation capability of the SAR imaging system.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, the computer program including program instructions that, when executed by a processor, cause the processor to perform the method described in any one of claims 1-6.

Images