SAR image ship target Wilcoxon-GO nonparametric detection method based on large selection logic

Abstract

The invention discloses an SAR (Synthetic Aperture Radar) image ship target Wilcoxon-GO nonparametric detection method based on large-scale selection logic, and relates to the technical field of image detection. The method comprises the following steps: firstly, setting a detection sliding window which comprises a detection unit, a reference sliding window consisting of a front edge sliding window and a rear edge sliding window, and a protection area; traversing the SAR ship remote sensing image pixel by pixel by using the detection sliding window, wherein the moving distance each time is equal to the width of the detection unit; then calculating the mean ratio MR of the front edge sliding window and the rear edge sliding window, and judging whether clutters are uniform or non-uniform by means of a threshold; selecting a reference sliding window according to a judgment result; the uniform clutter selects the whole reference sliding window, and the non-uniform clutter selects a leading edge / trailing edge sliding window with a large mean value and widens the window by one time; and finally, calculating a detection statistic R, setting a detection threshold T, and judging whether the detection unit has a target or not according to the relationship between R and T. According to the method, the false alarm rate is approximately constant in the non-uniform background of the clutter edge and is obviously superior to the situation that the false alarm of the traditional Wilcoxon is excessively increased, and the false alarm control capability is greatly improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image detection, and particularly relates to a SAR image ship target Wilcoxon-GO non-parametric detection method based on a large-logic selection. BACKGROUND

[0002] In the field of modern marine monitoring, synthetic aperture radar (SAR) has become a core technical means for ship target identification and tracking due to its unique advantages of all-weather, all-day, wide coverage, high resolution, and the ability to penetrate clouds and fog. With the increasing frequency of global marine activities, whether it is maritime traffic regulation, fishery resource protection in the civil field, or ship detection and situation awareness in other fields, the demand for accurate, fast and stable detection of ship targets in SAR images is increasingly urgent.

[0003] Among various technologies for SAR image ship target detection, radar target constant false alarm rate (CFAR) detection technology has become the most widely used and effective method due to its ability to stably control false alarm probability in complex backgrounds. Since the two-parameter CFAR detection method based on Gaussian distribution was proposed, the industry has long focused on parametric CFAR detection algorithms for SAR image ship target detection. The core idea is to assume that the sea clutter background in the SAR image follows a specific statistical distribution (such as Gaussian distribution, negative exponential distribution, Rayleigh distribution, K distribution, alpha stable distribution, G0 distribution, lognormal distribution, Gamma distribution, or mixed Rayleigh distribution), then estimate the distribution parameters using clutter sample selection or rejection techniques, and finally set the detection threshold based on the parameters to realize target discrimination.

[0004] However, the sea clutter in SAR images is strongly influenced by oceanic environments such as wind, waves, ocean currents, and internal waves, as well as imaging conditions, showing strong complexity and variability. The statistical characteristics of actual sea clutter often deviate from the distribution models assumed by parametric CFAR. Once a mismatch occurs, the detection performance of parametric CFAR will significantly deteriorate, and even a large number of missed detections or false alarms will occur. At the same time, parametric CFAR requires complex calculations to estimate distribution parameters, which has the inherent defect of excessive computational load and difficulty in meeting real-time detection requirements, limiting its application in high-speed SAR data processing scenarios.

[0005] To break through the limitation of the parameter CFAR, the applicant of the present application previously tried to use the Wilcoxon non-parametric CFAR detection technology for SAR image ship target detection. However, practice shows that the Wilcoxon non-parametric CFAR still has a significant technical shortcoming: in the SAR image ship target detection scene, affected by factors such as wind speed change, ocean current movement, internal wave or sea surface oil pollution, the sea surface often appears a special phenomenon of division between light and dark areas, that is, the clutter edge, resulting in non-uniform distribution of background clutter. Under this non-uniform background, the false alarm probability of the Wilcoxon non-parametric CFAR will be excessively high - a large number of non-target area clutter signals are misjudged as ship targets, which greatly restricts its practical value in complex marine environments.

[0006] Therefore, how to improve the false alarm control ability of the Wilcoxon non-parametric CFAR detection technology in the non-uniform background of the clutter edge has become a key technical problem to be solved in the field of SAR image ship target detection at present, and is also the core improvement direction of the present application.

[0007] The information disclosed in this BACKGROUND section is only for the purpose of increasing the understanding of the general background of the present application and should not be taken as an acknowledgment or any form of suggestion that this information forms prior art that is already known in this field. SUMMARY

[0008] In view of the above technical problems, the embodiments of the present application provide a SAR image ship target Wilcoxon-GO non-parametric detection method based on a max selection logic to solve the problems raised in the above background technology.

[0009] The present application provides the following technical solutions: a SAR image ship target Wilcoxon-GO non-parametric detection method based on a max selection logic, comprising the following steps: setting a detection sliding window, which includes a detection unit located in the middle, a reference sliding window surrounding the detection unit, and a protection area between the detection unit and the reference sliding window; the reference sliding window includes a front edge sliding window and a rear edge sliding window; using the detection sliding window to traverse the SAR collected ship target remote sensing image pixel by pixel, the distance of each movement of the detection sliding window being equal to the width of the detection unit, that is, t one pixel; obtaining a mean ratio MR of the front edge sliding window and the rear edge sliding window; setting a threshold , comparing the MR with and respectively; if , it is determined that the reference samples of the front edge sliding window and the rear edge sliding window are uniform clutter; if If the reference samples of the leading edge sliding window and the trailing edge sliding window are determined to be non-uniform clutter, it is determined that the edge of the clutter appears in the sliding window (such as the edge of the sea surface caused by the wind speed, ocean current, internal wave or sea surface oil stain, which forms the boundary between the light and dark areas of the sea surface); If it is determined to be uniform clutter, the entire reference sliding window is selected as the reference sample for threshold calculation; If it is determined to be non-uniform clutter, the leading edge sliding window or the trailing edge sliding window with a larger sample mean is selected as the reference sample for threshold calculation, and the width of the leading edge sliding window or the trailing edge sliding window is expanded by 1 time, i.e., the width of the leading edge sliding window or the trailing edge sliding window is changed to 2 q ; expanded by 1 time to 2 q After that, when the Wilcoxon-GO non-parametric detector sets the detection threshold by using the entire reference sliding window, the leading edge sliding window or the trailing edge sliding window, the number of reference units in the entire reference sliding window, the leading edge sliding window or the trailing edge sliding window is the same, which is n .

[0010] According to formula (1), a statistical quantity for detecting a ship target in a SAR image is calculated R ; Formula (1) is: ; wherein, is a detection sample of a detection unit, is a reference sample of a reference sliding window, u is a unit step function, m is the number of detection samples, n is the number of reference samples; The detection threshold is set as T , if R > T , it is determined that the ship target exists in the detection unit; if R ≤ T , it is determined that the ship target does not exist in the detection unit.

[0011] It is explained that the Wilcoxon-GO non-parametric detector is an improved SAR image ship target Wilcoxon non-parametric detector based on the “Greatest Of” logic, and therefore, the method of the present application is simply referred to as the Wilcoxon-GO non-parametric detector.

[0012] Preferably, the width of the detection unit is t pixels, the width of the leading edge sliding window and the trailing edge sliding window is q pixels, and the width of the protection area is g pixels.

[0013] Preferably, the mean ratio MR is calculated according to formula (2); formula (2) is: ; wherein, and are the mean values of the reference samples in the leading edge sliding window and the trailing edge sliding window, respectively.

[0014] Preferably, the detection threshold T According to the preset false alarm probability P fa is calculated according to formula (3); formula (3) is: ; represents from 1 to m + n is selected from m the number of values and the sum is k the number of possible ways; n is the number of reference samples, m is the number of detection samples, N = m + n .

[0015] Preferably, the calculation formula of in formula (2) is calculated according to formula (4); formula (4) is: .

[0016] Preferably, the reference sliding window is used to set the detection threshold; and the protection unit is used to prevent the pixel energy of adjacent ship targets from leaking into the detection unit.

[0017] Preferably, the threshold is determined according to the probability that the uniform background is incorrectly judged as non-uniform, which is 0.1.

[0018] The Wilcoxon-GO non-parametric detection method for ship targets in SAR images based on the large selection logic provided by the embodiment of the application has the following beneficial effects: (1) The Wilcoxon-GO non-parametric detection method of the application maintains the original detection ability of Wilcoxon in the uniform background, and the false alarm rate in the clutter edge is approximately constant, which is significantly better than the case that the false alarm of the traditional Wilcoxon non-parametric detector is excessively high, and the false alarm control ability is greatly improved.

[0019] (2) In the non-uniform background (clutter edge), the leading edge / trailing edge sliding window with a larger sample mean is selected and widened to 2 q ; in the clutter edge scene, the mean value of the strong clutter region is large, and the "large selection" can set a higher detection threshold based on the strong clutter sample, so as to avoid the low threshold of the weak clutter region leading to the rise of the false alarm rate. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1A schematic diagram of the detection sliding window of the Wilcoxon-GO nonparametric detector for ship targets in SAR images based on the selection logic; In the diagram, the yellow area on the right is the leading edge sliding window, and the green area on the left is the trailing edge sliding window. Figure 2 A schematic diagram showing clutter edges entering the Wilcoxon-GO and the Wilcoxon nonparametric detector reference sliding window; in, Figure 2 The upper part is the Wilcoxon nonparametric detector reference sliding window; Figure 2 The lower half of the image is the Wilcoxon-GO nonparametric detector reference sliding window; Figure 3 False alarm probability of Wilcoxon-GO nonparametric detector P fa clutter edge position L The curve of change; Figure 4 This is a schematic diagram illustrating the specific implementation process of the Wilcoxon-GO nonparametric detector. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] The following is in conjunction with the appendix Figures 1-4 The present invention will be further described in detail below, along with specific embodiments.

[0023] To address the problems mentioned in the background section, this invention provides a Wilcoxon-GO nonparametric detection method for ship targets in SAR images based on selection logic, in order to solve the aforementioned technical problems. The specific technical solution is as follows.

[0024] (a) Detection using the Wilcoxon-GO nonparametric detection method The steps of the Wilcoxon-GO nonparametric detection method for ship targets in SAR images based on the selection of the largest value logic include: Step 1: Initialize the detection sliding window and traversal method For remote sensing images acquired by spaceborne synthetic aperture radar (SAR), a Wilcoxon-GO nonparametric detector is used for traversal detection: The step size of the detection sliding window moving along the image each time is the width of the detection unit. t 100 pixels (of which)t =2, which is the recommended parameter); The detection sliding window consists of two parts: a leading edge sliding window (right side area) and a trailing edge sliding window (left side area), both with an initial width of [missing information]. q (in q =3, which is the recommended parameter); A protective area with a width of [missing information] is provided between the detection unit and the reference unit (front / back sliding window). g (in g =60, which needs to be greater than the maximum length of the ship in the SAR image to prevent energy leakage from adjacent targets.

[0025] Step 2: Refer to the dynamic selection and width adjustment of the sliding window. The Wilcoxon-GO nonparametric detector selects the reference cell used to set the detection threshold through the following logic: When "Entire Reference Sliding Window" is selected, all reference cells of the leading edge sliding window and the trailing edge sliding window are used directly; When selecting "Front Edge Sliding Window" or "Rear Edge Sliding Window", its width needs to be increased by 1 to 2 times. q This ensures that regardless of which reference window is selected, the number of reference cells it contains is always the same. n (To ensure statistical consistency).

[0026] Step 3: Calculate the mean ratio (MR) and determine clutter uniformity. Define the mean ratio (MR): Calculate the mean of the reference sample in the leading-edge sliding window. M A and the mean of the trailing edge sliding window reference sample M B, the ratio of the two is the mean ratio: ; Determine clutter uniformity: Compare MR with threshold and its reciprocal K Comparison (where) =1.143, determined through simulation, the probability of a uniform background being misclassified as non-uniform is 0.1). like The reference samples for the leading and trailing edges of the sliding window are determined to be a uniform background. like It is determined to be a non-uniform background (i.e., there are "clutter edges").

[0027] Step 4: Select a reference sliding window based on clutter type Uniform background: The detection threshold is set using the reference cell of the entire reference sliding window; Non-uniform background (clutter edges): Select the leading edge sliding window or the trailing edge sliding window with a larger mean (i.e., the "select larger (GO)" logic) to set the detection threshold.

[0028] Step 5: Calculate the detection statistic R And determine whether the target exists. Define detection statistics R Let the sample of the detection unit be... x 1, x 2,…, The sample of the reference unit is y 1, y 2,…, Statistic R The total number of times the detected sample was greater than the reference sample: ; Determine the detection threshold T And determine: threshold T Based on the set false alarm probability The solution is obtained using the following formula: ; in, Indicates from 1 to m + n Selected from m There are 5 values ​​and their sum is 1 k The number of possible choices; n This is the number of reference samples. m It is the number of samples tested. N = m + n .

[0029] like R >T indicates that a ship target is detected in the detection unit; like R ≤ T The system determined that there were no ship targets in the detection unit.

[0030] Additional notes: Clutter edge modeling parameters In this method, clutter edge scenes are modeled based on the following parameters (used to verify method performance): Strong clutter region: conforms to shape parameters c =1.2 Weibull distribution; Weak clutter region: conforms to shape parameters c =2.0 Weibull distribution (i.e. Rayleigh distribution); The power ratio of strong clutter to weak clutter is CNR = 5dB.

[0031] (ii) Detection using the Wilcoxon nonparametric detection method The conventional Wilcoxon nonparametric detection method was used for detection; it shares the same experimental scenario as the Wilcoxon-GO nonparametric detection method: that is, the clutter edge is modeled as a Weibull distribution, and the strong clutter region is a region with shape parameters of . c The shape parameter of the weak clutter region is given by the Weibull distribution with a value of 1.2. c =2.0 (i.e. Rayleigh distribution), the power ratio of strong clutter to weak clutter is called CNR=5dB.

[0032] Furthermore, the parameters of both the Wilcoxon nonparametric detector and the Wilcoxon-GO nonparametric detector are set to... t =2, q =3 and g =60. Among them, Figure 2 The diagram shows the clutter edge entering the Wilcoxon-GO nonparametric detector and the Wilcoxon nonparametric detector reference sliding window from the right side (leading sliding window). Strong clutter regions are shown in gray, and the locations of the strong clutter are indicated by "". L ".

[0033] (III) Test Results Figure 3 The false alarm probabilities of the Wilcoxon-GO nonparametric detector and the Wilcoxon nonparametric detector are given. clutter edge position L The curves show the changes in the false alarm probability of the detectors. It can be seen that when the detectors are not in the strong clutter region, their false alarm probabilities decrease. However, when the detectors enter the strong clutter region, the false alarm probability of the Wilcoxon nonparametric detector increases excessively, which severely violates the constraint requirements for the detector's false alarm probability.

[0034] However, after the detection unit enters the strong clutter region, the false alarm probability of the Wilcoxon-GO nonparametric detector remains approximately constant near the set value, with a lower "secondary false alarm spike" only appearing after the strong clutter region enters the tail of the reference sliding window.

[0035] Therefore, the Wilcoxon-GO nonparametric detector has significantly improved false alarm control capability at clutter edges compared to the Wilcoxon nonparametric detector. This is mainly because the Wilcoxon-GO nonparametric detector uses a leading-edge sliding window with a large sample mean to set the detection threshold with a higher probability after strong clutter enters the reference sliding window.

[0036] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A Wilcoxon-GO nonparametric detection method for ship targets in SAR images based on the selection of the largest value logic, characterized in that, Includes the following steps: The detection sliding window includes a detection unit in the middle, a reference sliding window surrounding the detection unit, and a protection area between the detection unit and the reference sliding window; the reference sliding window includes a front edge sliding window and a rear edge sliding window; The detection sliding window is used to traverse the SAR-acquired ship target remote sensing images pixel by pixel, and the distance the detection sliding window moves each time is equal to the width of the detection unit; Obtain the mean ratio MR of the leading edge sliding window and the trailing edge sliding window; Set threshold MR and and Compare; like If so, the reference samples of the leading edge sliding window and the trailing edge sliding window are determined to be uniform clutter; like If so, the reference samples of the leading edge sliding window and the trailing edge sliding window are determined to be non-uniform clutter, that is, the clutter edge situation is detected in the sliding window; If it is determined to be uniform clutter, the entire reference sliding window is selected as the reference sample for threshold calculation; If it is determined to be non-uniform clutter, select the leading edge window or trailing edge window with a larger sample mean as the reference sample for threshold calculation, and double the width of the leading edge window or trailing edge window when selecting it. The statistical measures for detecting ship targets in SAR images are calculated according to equation (1). R ; Equation (1) is: ; in, It is the test sample of the detection unit. It is a reference sample for the sliding window. u It is a unit step function. m It refers to the number of samples tested. n This is the reference sample size; Set detection threshold T ,like R > T If so, it is determined that a ship target exists in the detection unit; if R ≤ T If no ship target is detected, it is determined that there is no ship target in the detection unit.

2. The Wilcoxon-GO nonparametric detection method for ship targets in SAR images based on the maximum selection logic as described in claim 1, characterized in that, The width of the detection unit is t Each pixel has a width of [number] for both the leading and trailing edges of the sliding window. q [Number] pixels, the width of the protected area is [Number]. g 1 pixel.

3. The Wilcoxon-GO nonparametric detection method for ship targets in SAR images based on the maximum selection logic as described in claim 1, characterized in that, The mean ratio MR is calculated according to equation (2); equation (2) is: ； in, and These are the mean values ​​of the reference samples in the leading edge sliding window and the trailing edge sliding window, respectively.

4. The Wilcoxon-GO nonparametric detection method for ship targets in SAR images based on the maximum selection logic as described in claim 1, characterized in that, Detection threshold T Based on the preset false alarm probability Calculated according to equation (3); equation (3) is: ； Indicates from 1 to m + n Selected from m There are 5 values ​​and their sum is 1 k The number of possible choices; n This is the number of reference samples. m It is the number of samples tested. N = m + n .

5. The Wilcoxon-GO nonparametric detection method for ship targets in SAR images based on the maximum selection logic as described in claim 1, characterized in that, In formula (2) The calculation formula is obtained from formula (4); formula (4) is: 。 6. The Wilcoxon-GO nonparametric detection method for ship targets in SAR images based on the maximum selection logic as described in claim 1, characterized in that, The reference sliding window is used to set the detection threshold; the protection unit is used to prevent the pixel energy of adjacent ship targets from leaking into the detection unit.

7. The Wilcoxon-GO nonparametric detection method for ship targets in SAR images based on the maximum selection logic as described in claim 1, characterized in that, Threshold The probability of a uniform background being incorrectly identified as non-uniform is determined based on 0.1.

Images