A SAR image quality assessment method based on correlation and discriminativeness

Abstract

This invention discloses a SAR image quality assessment method based on correlation and discriminative analysis, comprising the following steps: S1, acquiring two single-look complex images of the same ground feature at two different times: one when the SAR is not disturbed by noise from a civilian radiation source or active interference source of the same frequency, and the other when it is disturbed by noise. The correlation coefficient of the two images after registration is calculated, and the corresponding interference-to-signal ratio is obtained by establishing a quantitative relationship between the correlation coefficient and the interference-to-signal ratio (CNR). S2, establishing the forms of the interference-to-signal ratio and the CNR, and converting the interference-to-signal ratio into the CNR of the noise-disturbed SAR image. S3, calculating the US National Image Interpretability Rating Scale (NIIRS) based on the CNR, converting it into the corresponding NIRS for infrared detectors, and then calculating the target identification probability for the target to be identified in the image. This invention can quantitatively predict and analyze the impact of intentional and unintentional noise disturbances on SAR systems, helping image users to better utilize images.

Description

Technical Field

[0001] This invention belongs to the field of signal and information processing technology, and in particular relates to a SAR image quality assessment method based on correlation and discriminativeness. Background Technology

[0002] Synthetic Aperture Radar (SAR) is an active microwave remote sensing imaging radar. Compared to optical and infrared images, SAR is unaffected by lighting and weather conditions, enabling all-weather, high-resolution, and large-area imaging. It has become an irreplaceable tool in military intelligence reconnaissance, land resource monitoring, and geographic information mapping. SAR image quality assessment is a crucial foundation for interpreting information such as target detection, identification, and confirmation (detection refers to determining the presence of a target; identification refers to determining the type of target; confirmation refers to determining the specific model of the target, such as distinguishing between a T62 and a T72 tank) based on SAR images, and for evaluating image usability. Especially in increasingly complex electromagnetic environments, where SAR is disturbed by civilian radiation sources of the same frequency or active interference sources, image quality assessment is extremely critical for analyzing image quality degradation and the resulting decrease in image usability.

[0003] Currently, SAR image quality assessment methods are mainly divided into subjective and objective assessments. Subjective assessment methods rely on professional image interpreters to score image quality based on specific application purposes, offering the advantage of comprehensive consideration of various factors; however, the assessment process is time-consuming, labor-intensive, and inefficient. Objective assessment methods include those based on basic image parameters such as image resolution, peak-to-sidelobe ratio, and integral-to-sidelobe ratio, as well as image feature assessment methods based on human visual characteristics. These methods are relatively simple to calculate and have clear physical meanings; however, these indicators are usually at a lower level and cannot directly reflect image usability information most important to image users, such as target identification probability (identification includes detection, recognition, and confirmation). Furthermore, determining the minimum level of objective assessment indicators required to meet usability requirements still requires expert interpretation to establish thresholds, introducing significant human factors into the assessment and hindering efficiency. Establishing concise objective evaluation indicators and defining their quantitative relationship with indicators reflecting image usability, such as target identification probability, to better assist frontline SAR image users in image analysis and usability prediction, has always been a core issue in SAR image quality assessment. Summary of the Invention

[0004] To address the shortcomings of existing objective image quality assessment methods, such as insufficient guidance for image usage, excessive subjective evaluation, and low efficiency, this invention aims to provide a SAR image quality assessment method based on correlation and discriminativeness. Specifically, for complex electromagnetic environments where SAR is disturbed by noise from civilian radiation sources or active interference sources of the same frequency, this method uses SAR images acquired at two different times. By calculating the quantitative relationship between the correlation coefficient and signal-to-interference ratio of the two images, it estimates the target discrimination probability, which reflects the image's usability. This probability is specifically divided into detection probability, recognition probability, and confirmation probability.

[0005] To achieve the above-mentioned objectives, the present invention adopts the following technical solution:

[0006] A SAR image quality assessment method based on correlation and discriminativeness includes the following steps:

[0007] S1. At two different times—one when the SAR is not disturbed by noise from civilian radiation sources or active interference sources of the same frequency, and the other when it is disturbed by noise—two single-look complex images of the same ground object are acquired by the SAR. The correlation coefficient of the two images after registration is calculated. By establishing a quantitative relationship between the correlation coefficient and the interference-to-signal ratio (CNR), the CNR χ corresponding to the correlation coefficient is obtained. JSR ;

[0008] S2. Establish the interference-to-signal ratio χ JSR Hexin dry matter ratio χ SJCR The form of the interference-to-signal ratio χ obtained in step S1 is expressed as follows: JSR Transformed into signal-to-interference-to-clutter ratio (χ) of a SAR image subject to noise perturbation SJCR ;

[0009] S3. Based on the signal-to-interference-to-noise ratio χ obtained in step S2 SJCR The National Image Interpretability Rating Scale (NIIRS) based on SAR images is calculated and converted into the corresponding NIIRS for infrared detectors. Then, the probability of identifying a target in the image is calculated.

[0010] Furthermore, step S1 above includes the following sub-steps:

[0011] Step S1.1: Using a combination of map-based operations and field surveys, determine the area containing the target and background that needs SAR imaging. At two different times, when there is no noise disturbance and when there is noise disturbance, the SAR obtains single-view multiple images of the area with the same operating parameters.

[0012] Step S1.2: Perform image registration on the two single-look complex SAR images obtained in step S1.1 to obtain an image IM0 that is not disturbed by noise and an image IM0 that is disturbed by noise, with matching spatial locations of the imaging regions. JAnd calculate the correlation coefficient ρ according to formula (1). c

[0013]

[0014] In the formula, I0, I J These represent the image IM0, which is unaffected by noise, and the image IM, which is affected by noise, respectively. J An image vector is a vector formed by arranging the pixels of a rectangular image in a top-to-bottom, left-to-right order; ||·|| represents the magnitude of the vector.

[0015] Step S1.3: ρ calculated based on step S1.2 c The interference-to-signal ratio χ is calculated according to formula (2). JSR

[0016]

[0017] Furthermore, step S2 above includes the following sub-steps:

[0018] Step S2.1: For target identification, based on the noise-free SAR intensity image, i.e., |IM0| 2 Calculate the corresponding interference-to-signal ratio χ according to formula (4). JSR Average interference power

[0019]

[0020] In the formula, m is the total number of pixels in the SAR image;

[0021] p represents the target being detected, R. T Pixel count;

[0022] q represents clutter background R C Number of pixels, m = p + q, p / m is the target sparsity;

[0023] Step S2.2: Based on the average interference power calculated in step S2.1 Calculate the signal-to-interference-to-noise ratio (χ) according to formula (5). SJCR

[0024]

[0025] In the formula, R JC For SAR intensity images affected by noise, i.e., |IM J | 2 Target R in the middle T External interference and clutter superimposed on pixel areas;

[0026] This represents the average pixel value in this region.

[0027] Furthermore, step S3 above includes the following sub-steps:

[0028] Step S3.1: Based on the signal-to-interference-to-mixture ratio χ obtained in step S2 SJCR The NIIRS of the SAR image is calculated according to formula (6), i.e.

[0029]

[0030] In the formula, BW 2D =1 / (ρ a ρ g ) represents the two-dimensional spatial spectral bandwidth, ρ a ρ g These are the azimuth resolution and the ground distance resolution, respectively. g =ρ r / cosβ,ρ r Where β is the slant range resolution and β is the ground rubbing angle;

[0031] C J The channel capacity corresponding to the SAR image;

[0032] Step S3.2: Calculate the NIIRS obtained in step S3.1. SAR Converted to the corresponding NIIRS for the infrared detector according to formula (7), i.e.

[0033]

[0034] Then, calculate the given NIIRS IR Minimum sampling interval required for infrared detectors

[0035]

[0036] Step S3.3: For the target to be identified, calculate the minimum sampling interval d obtained in step S3.2. min The discrimination probability is calculated according to formula (9).

[0037]

[0038] In the formula, M represents the number of pixel periods corresponding to the target size;

[0039] l and w represent the length and width of the target, respectively;

[0040] M 50 This refers to the number of pixel cycles required to complete the corresponding identification task with a 50% probability; for detection, identification, and verification tasks, M... 50 Take values ​​of 1.5, 6, and 12 respectively.

[0041] Due to the adoption of the technical solution described above, the present invention has the following advantages:

[0042] This SAR image quality assessment method based on correlation and discriminative analysis establishes a quantitative relationship between the SAR image coherence coefficient and the target discrimination probability to evaluate SAR image quality. It can quantitatively predict and analyze the impact of intentional and unintentional noise disturbances on the SAR system, helping image users to better utilize images. It comprehensively considers factors such as target, environment, and interference to achieve objective image quality evaluation without requiring professional image interpretation. This overcomes the problems of weak guidance in traditional testing methods, excessive human factors, and low efficiency in image quality assessment standards. The calculation process does not require complex computations and has low computational intensity, making it well-suited for engineering practice and applicable to image quality assessment of systems such as airborne synthetic aperture radar, spaceborne synthetic aperture radar, and synthetic aperture radar seekers. Attached Figure Description

[0043] Figure 1 This is a flowchart of the SAR image quality assessment method based on correlation and discriminativeness of the present invention;

[0044] Figure 2 It is a SAR image that has not been disturbed by Gaussian noise;

[0045] Figure 3 It is a SAR image disturbed by Gaussian noise;

[0046] Figure 4 Is the dry signal ratio χ JSR Calculation diagram;

[0047] Figure 5 Is the information-to-do ratio χ SJCR Calculation diagram;

[0048] Figure 6 It is the correlation coefficient ρ c Compared with the dry signal ratio χ JSR Quantitative relationship diagram;

[0049] Figure 7 It is the correlation coefficient ρ c With the discrimination probability P d Quantitative relationship diagram. Detailed Implementation

[0050] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0051] like Figure 1 As shown, a SAR image quality assessment method based on correlation and discriminativeness includes the following steps:

[0052] S1. At two different times—one when the SAR is not disturbed by noise from civilian radiation sources or active interference sources of the same frequency, and the other when it is disturbed by noise—two single-look complex images of the same ground object are acquired by the SAR. The correlation coefficient of the two images after registration is calculated. By establishing a quantitative relationship between the correlation coefficient and the interference-to-signal ratio (CNR), the CNR χ corresponding to the correlation coefficient is obtained. JSR The specific steps are as follows:

[0053] Step S1.1: Using a combination of map-based operations and field reconnaissance, determine the area containing the target and background that needs SAR imaging. At two different times, when there is no noise disturbance and when there is noise disturbance, the SAR uses the same operating parameters to obtain single-look complex images of the area. The same operating parameters include, but are not limited to, resolution, viewing angle, and polarization mode.

[0054] Step S1.2: Perform image registration on the two single-look complex SAR images obtained in step S1.1 to obtain an image IM0 that is not disturbed by noise and an image IM0 that is disturbed by noise, with matching spatial locations of the imaging regions. J And calculate the correlation coefficient ρ according to formula (1). c

[0055]

[0056] In the formula, I0, I J These represent the image IM0, which is unaffected by noise, and the image IM, which is affected by noise, respectively. J An image vector is a vector formed by arranging the pixels of a rectangular image in a top-to-bottom, left-to-right order; ||·|| represents the magnitude of the vector.

[0057] Step S1.3: ρ calculated based on step S1.2 c The interference-to-signal ratio χ is calculated according to formula (2). JSR

[0058]

[0059] The derivation of the above formula (2) is as follows: Let the image IM be disturbed by noise. J Affected by Gaussian noise disturbance, i.e., IM J =IM0+IM s IM s The signal is perturbed by Gaussian noise; based on the statistical distribution law of SAR single-look complex images and Gaussian noise interference, the real part I of pixel IM0 is... 0i_rel Imaginary part I 0i_img It is a joint circular Gaussian distribution, i.e. N(·) represents a Gaussian distribution. Let I be the variance, and I 0i_rel with I0i_img Unrelated; IM J The real part I of a pixel Ji_rel Imaginary part I Ji_img It is a joint circular Gaussian distribution, and IM0 and IM J Irrelevant, that is For noise variance; at the same time, image pixel {I 0i}、{I Ji All are mutually independent, and the theoretical correlation coefficient ρ is calculated according to formula (3). c for

[0060]

[0061] In the formula, E(·) represents the expectation;

[0062] S2. For target detection, recognition, and confirmation applications in SAR images, establish the interference-to-signal ratio χ. JSR Hexin dry matter ratio χ SJCR The form of the interference-to-signal ratio χ obtained in step S1 is expressed as follows: JSR Transformed into signal-to-interference-to-clutter ratio χ of a noisy SAR image SJCR The specific steps are as follows:

[0063] Step S2.1: For target identification, based on the noise-free SAR intensity image, i.e., |IM0| 2 Calculate the corresponding interference-to-signal ratio χ according to formula (4). JSR Average interference power

[0064]

[0065] In the formula, m is the total number of pixels in the SAR image;

[0066] p represents the target being detected, R. T Pixel count;

[0067] q represents clutter background R C Number of pixels, m = p + q, p / m is the target sparsity;

[0068] In the calculation of the correlation coefficient, the interference-to-signal ratio χ² JSR The "signal" in this context refers to the noise-free SAR intensity image |IM0| 2 The mean of all pixels, including clutter background R C The pixels also include the detected target R. T The pixels; interference refers to the intensity of the image caused by noise disturbance signals | IM s | 2 The mean of all pixels;

[0069] Step S2.2: Based on the average interference power calculated in step S2.1 Calculate the signal-to-interference-to-noise ratio (χ) according to formula (5). SJCR

[0070]

[0071] In the formula, R JC For SAR intensity images affected by noise, i.e., |IM J | 2 Target R in the middle T External interference and clutter superimposed on pixel areas;

[0072] This represents the average pixel value in this region.

[0073] For target identification, the signal-to-interference-to-noise ratio (χ²) SJCR The "signal" in this context refers to the power of the identified target, i.e., the SAR intensity image affected by noise perturbation |IM J | 2 Mean pixel value of the target portion "Clutter" refers to the pixel area in a SAR intensity image that is superimposed with interference and clutter outside the detected target area due to noise disturbance. JC mean Since interference and clutter are uncorrelated

[0074] S3. Based on the signal-to-interference-to-noise ratio χ obtained in step S2 SJCR The National Imagery Interpretability Rating Scale (NIIRS) based on SAR images is calculated and converted into the corresponding NIIRS for infrared detectors. Then, the probability of identifying the target is calculated. The specific steps are as follows:

[0075] Step S3.1: Based on the signal-to-interference-to-mixture ratio χ obtained in step S2 SJCR The NIIRS of the SAR image is calculated according to formula (6), i.e.

[0076]

[0077] In the formula, BW 2D =1 / (ρ a ρ g ) represents the two-dimensional spatial spectral bandwidth, ρ a ρ g These are the azimuth resolution and the ground distance resolution, respectively. g =ρ r / cosβ,ρ r Where β is the slant range resolution and β is the ground rubbing angle;

[0078] C J The channel capacity corresponding to the SAR image;

[0079] Step S3.2: Calculate the NIIRS obtained in step S3.1. SAR Converted to the corresponding NIIRS for the infrared detector according to formula (7), i.e.

[0080]

[0081] Then, calculate the given NIIRS IR Minimum sampling interval required for infrared detectors

[0082]

[0083] Step S3.3: For the target to be identified, calculate the minimum sampling interval d obtained in step S3.2. min The discrimination probability is calculated according to formula (9).

[0084]

[0085] In the formula, M represents the number of pixel periods corresponding to the target size;

[0086] l and w represent the length and width of the target, respectively;

[0087] M 50 This refers to the number of pixel cycles required to complete the corresponding identification task with a 50% probability; for detection, identification, and verification tasks, M... 50 Take values ​​of 1.5, 6, and 12 respectively.

[0088] The implementation of the technical solution of the present invention will be described in detail through the following embodiments.

[0089] This embodiment is a simulation example. It utilizes an airborne SAR single-look complex image with an azimuth resolution and slant range resolution of 0.3m, a pixel size of 789×798, and an imaging ground-touching angle of 20°. Gaussian noise with an interference-to-signal ratio of 10dB is superimposed on the single-look complex image to generate a noise-perturbed image. A tank with dimensions of 3.6m×7.9m is used as the target to be detected, and the tank is located in a farmland environment.

[0090] Reference Figure 1 A SAR image quality assessment method based on correlation and detection includes the following steps:

[0091] Step S1, refer to Figure 2The method involves acquiring two single-look complex images of the same ground feature at two different times, one without noise disturbance and the other with noise disturbance. The correlation coefficient between the two images, before and after noise disturbance and after image registration, is calculated. Then, the relationship between the correlation coefficient and the interference-to-signal ratio (ISR) is used to obtain the ISR corresponding to the correlation coefficient. The specific steps are as follows:

[0092] Step S1.1, with Figure 2 The image shown is IM0, which is an image undisturbed by noise. By superimposing noise perturbation onto this image, a new image is formed. Figure 3 The image IM shown is affected by noise. J ;

[0093] Step S1.2: Calculate the image IM0 unaffected by noise and the image IM0 affected by noise according to formula (1). J The correlation coefficient ρ between them c

[0094] ρ c =0.3

[0095] Step S1.3: Based on the correlation coefficient ρ c Compared with the dry signal ratio χ JSR The quantitative relationship is calculated using the following formula to determine the corresponding interference-to-information ratio χ². JSR

[0096]

[0097] Step S2: For target detection, recognition, and confirmation applications in SAR images, establish the interference-to-signal ratio χ. JSR Hexin dry matter ratio χ SJCR The form of the interference-to-signal ratio χ obtained in step S1 will be... JSR Transformed into signal-to-interference-to-clutter ratio χ of a noisy SAR image SJCR The specific steps are as follows:

[0098] Step S2.1, with Figure 2 The target is a tank, and the background is farmland. Figure 4 P is 2448, q is 627174, and m is 629622. It is 7791. The value is 715.7, calculated according to the following formula.

[0099]

[0100] Step S2.2, refer to Figure 5 Calculate according to the following formula

[0101]

[0102] Step S3: Based on the χ obtained in step S2 SJCR ,calculate Figure 3 The SAR image shown is converted to its corresponding infrared detector NIRS, and then the discrimination probability against the tank target is calculated. The specific steps are as follows:

[0103] Step S3.1, based on the χ obtained in step S2 SJCR Calculate according to the following formula Figure 3 The SAR image shown has NIIRS.

[0104] C J =BW 2D log2(1+χ SJCR ) = 10.1

[0105] NIIRS SAR =3.5455+0.3660log2(C J ) = 4.77

[0106] Step S3.2: The NIIRS calculated in step S3.1 SAR Convert to the corresponding NIIRS for the infrared detector using the following formula.

[0107]

[0108] Then, calculate the given NIIRS IR Minimum sampling interval required for infrared detectors

[0109]

[0110] Step S3.3: For the tank target, based on d calculated in step S3.2... min The discrimination probability is calculated according to the following formula.

[0111]

[0112] Detection probability P d (M,M 50 =1.5) =0.90,

[0113] Recognition probability P d (M,M 50 =6) = 0.08,

[0114] Confirmation probability P d (M,M 50 =12)=0.01.

[0115] In turn Figure 3 By superimposing Gaussian noise with different interference-to-signal ratios, corresponding noise-perturbed images are obtained, and then... Figure 1The flowchart shown calculates the image correlation coefficient and corresponding discrimination probability under different interference-to-signal ratio simulation conditions, thus obtaining... Figure 6 The correlation coefficient ρ shown c Compared with the dry signal ratio χ JSR The quantitative relationship diagram, and Figure 7 The correlation coefficient ρ shown c With the discrimination probability P d Quantitative relationship diagram.

[0116] The above description is only a preferred embodiment of the present invention and not a limitation thereof. Any equivalent changes and modifications made in accordance with the scope of the present invention without departing from the spirit and scope of the present invention shall be within the scope of patent protection of the present invention.

Claims

1. A SAR image quality assessment method based on correlation and discriminative power, characterized by: It includes the following steps: S1. At two different times—one when the SAR is not disturbed by noise from civilian radiation sources or active interference sources of the same frequency, and the other when it is disturbed by noise—two single-look complex images of the same ground object are acquired by the SAR. The correlation coefficient of the two images after registration is calculated. By establishing a quantitative relationship between the correlation coefficient and the interference-to-signal ratio (CNR), the CNR χ corresponding to the correlation coefficient is obtained. JSR ; S2. Establish the interference-to-signal ratio χ JSR Hexin dry matter ratio χ SJCR The form of the interference-to-signal ratio χ obtained in step S1 is expressed as follows: JSR Transformed into signal-to-interference-to-clutter ratio (χ) of a SAR image subject to noise perturbation SJCR ; Includes the following sub-steps: Step S2.1: For target identification, based on the noise-free SAR intensity image, i.e., |IM0| 2 Calculate the corresponding interference-to-signal ratio χ according to formula (4). JSR Average interference power In the formula, m is the total number of pixels in the SAR image; p represents the target being detected, R. T Pixel count; q represents clutter background R C Number of pixels, m = p + q, p / m is the target sparsity; Step S2.2: Based on the average interference power calculated in step S2.1 Calculate the signal-to-interference-to-noise ratio (χ) according to formula (5). SJCR In the formula, R JC For SAR intensity images affected by noise, i.e., |IM J | 2 Target R in the middle T External interference and clutter superimposed on pixel areas; This represents the average pixel value in this region. S3. Based on the signal-to-interference-to-noise ratio χ obtained in step S2 SJCR The National Image Interpretability Rating Scale (NIIRS) based on SAR images is calculated and converted into the corresponding NIIRS for infrared detectors. Then, the probability of identifying a target in the image is calculated.

2. The SAR image quality assessment method based on correlation and discriminative power according to claim 1, characterized in that: Its steps S1 Includes the following sub-steps: Step S1.1: Using a combination of map-based operations and field surveys, determine the area containing the target and background that needs SAR imaging. At two different times, when there is no noise disturbance and when there is noise disturbance, the SAR obtains single-view multiple images of the area with the same operating parameters. Step S1.2: Perform image registration on the two single-look complex SAR images obtained in step S1.1 to obtain an image IM0 that is not disturbed by noise and an image IM0 that is disturbed by noise, with matching spatial locations of the imaging regions. J And calculate the correlation coefficient ρ according to formula (1). c In the formula, I0, I J These represent the image IM0, which is unaffected by noise, and the image IM, which is affected by noise, respectively. J An image vector is a vector formed by arranging the pixels of a rectangular image in a top-to-bottom, left-to-right order; ||·|| represents the magnitude of the vector. Step S1.3: ρ calculated based on step S1.2 c The interference-to-signal ratio χ is calculated according to formula (2). JSR 3. The SAR image quality assessment method based on correlation and discriminative power according to claim 1, characterized in that: Step S3 includes the following sub-steps: Step S3.1: Based on the signal-to-interference-to-mixture ratio χ obtained in step S2 SJCR The NIIRS of the SAR image is calculated according to formula (6), i.e. In the formula, BW 2D =1 / (ρ a ρ g ) represents the two-dimensional spatial spectral bandwidth, ρ a ρ g These are the azimuth resolution and the ground distance resolution, respectively. g =ρ r / cosβ,ρ r Where β is the slant range resolution and β is the ground rubbing angle; C J The channel capacity corresponding to the SAR image; Step S3.2: Calculate the NIIRS obtained in step S3.

1. SAR Converted to the corresponding NIIRS for the infrared detector according to formula (7), i.e. Then, calculate the given NIIRS IR Minimum sampling interval required for infrared detectors Step S3.3: For the target to be identified, calculate the minimum sampling interval d obtained in step S3.

2. min The discrimination probability is calculated according to formula (9). In the formula, M represents the number of pixel periods corresponding to the target size; l and w represent the length and width of the target, respectively; M 50 This refers to the number of pixel cycles required to complete the corresponding identification task with a 50% probability; for detection, identification, and verification tasks, M... 50 Take values ​​of 1.5, 6, and 12 respectively.

Images