A simplified polarization SAR method and system for detecting marine oil spills.

Abstract

This invention discloses a simplified polarimetric SAR (SAR) method and system for detecting marine oil spills. The method includes: S1, preprocessing a fully polarimetric SAR image and extracting simplified polarimetric SAR data; S2, extracting total scattering power and entropy based on the simplified polarimetric SAR data to construct a simplified polarimetric SAR oil spill detector; S3, performing random forest classification on the polarization feature map extracted by the simplified polarimetric SAR detector to obtain an oil-water separation map, and quantitatively evaluating the oil spill detection capability of the oil-water separation map. This invention features low scattering power... esa By combining polarization characteristic parameters from both backscattering energy and scattering mechanism, the extraction of oil spill information can be improved, thus making oil spill detection more accurate.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of polarimetric scattering characteristics and compact polarimetric synthetic aperture radar image detection, more particularly, to a compact polarimetric SAR offshore oil spill detection method and system. BACKGROUND

[0002] Due to the increase of maritime trade and the development of offshore oil platforms, the risk of environmental pollution caused by oil spills has increased dramatically in the past few decades. Oil spill detection has also become a hot topic of discussion among researchers. Compared with traditional optical sensors, radar has advantages in large-area, night and bad-weather detection, but is easily affected by similar objects.

[0003] Using full polarimetric (FP) SAR data can detect oil spills, but its main limitations are reduced swath width and constrained incidence span, which reduce the ability of both in oil spill detection. The compact polarimetric (CP) architecture can provide related performance close to FP SAR while avoiding the shortcomings of FP SAR. CP mode can obtain partial polarization information while maintaining the swath width of single polarization SAR. This ability makes it very suitable for large-area marine surveillance. There are currently three main CP modes: π / 4 mode, dual circular polarization (DCP) and hybrid polarization (HP) mode. SAR in the hybrid polarization architecture (HP) transmits circular polarization and receives two orthogonal linear polarizations that are mutually coherent. CP SAR provides better oil spill detection performance than traditional dual polarization mode, and CP SAR data has been successfully applied to oil spill detection. Currently, CP SAR data is usually simulated from FP SAR data.

[0004] Many scholars have proposed polarimetric feature parameters for oil spill detection, which are basically based on three directions: backscatter energy, correlation of different channels and scattering mechanism. Combining different features to make oil spill detection more accurate is a direction that needs further research in the future. Gao proposed l esa The innovation combines backscatter energy and scattering mechanism, and is successfully applied to ship detection. Both of these two polarimetric features have been successfully applied to oil spill detection, and other researchers have also proposed different new features based on the two polarimetric features respectively. However, there is no literature on combining the two features for oil spill detection. Therefore, it is necessary to improve the prior art. SUMMARY

[0005] The present application relates to the field of polarimetric scattering characteristics and compact polarimetric synthetic aperture radar image detection, more particularly, to a compact polarimetric SAR offshore oil spill detection method and system.

[0006] In order to achieve the above purpose, the technical scheme adopted by the present application is as follows:

[0007] A compact polarimetric SAR offshore oil spill detection method, comprising the following steps:

[0008] S1, preprocessing the full polarimetric SAR image, and extracting compact polarimetric SAR data;

[0009] S2, extracting total scattering power and entropy based on the compact polarimetric SAR data, and constructing a compact polarimetric SAR oil spill detector;

[0010] S3, performing random forest classification on the polarization feature map extracted by the compact polarimetric SAR detector, obtaining an oil-water separation map, and quantitatively evaluating the oil spill detection capability of the oil-water separation map.

[0011] Further, the step S1 specifically comprises:

[0012] S10, downloading the full polarimetric SAR image, and extracting the covariance matrix in the HP mode by using the polarization scattering matrix S;

[0013] S11, constructing the Stokes vector in the HP mode from the covariance matrix, wherein the Stokes vector in the HP mode is the compact polarimetric SAR data.

[0014] Further, the polarization scattering matrix S in the step S10 is specifically:

[0015]

[0016] In the formula, the scattering matrix element S xy x and y represent transmission and reception, x, y∈{h, v}, and h and v represent horizontal and vertical polarization respectively.

[0017] Further, the Stokes vector in the HP mode in the step S11 is:

[0018]

[0019] Further, the step S2 specifically comprises:

[0020] S20, constructing five polarization features for verification by using the Stokes vector: entropy H w , total scattering power g0, polarization degree DoP, correlation polarization coefficient ρ co , and Bragg scattering related parameter iB CTLR ;

[0021] S21, decomposing the total scattering power g0 according to the contribution of the entropy H w of each unit in the response by using power entropy decomposition, and decomposing the total scattering power g0 based on the entropy H WThe total scattering power g0 is decomposed into two components to obtain a low scattering power l esa The low scattering power l esa The detector is used for detecting oil spills in compact polarimetric SAR.

[0022] Further, the five polarization features in step S20 are calculated as follows:

[0023]

[0024] g0 = <|S RH | 2 + |S RV | 2 >

[0025]

[0026]

[0027] iB CTLR = (g0 + g3) / (g0 - g3)

[0028] In the formula, λ i is the i-th eigenvalue.

[0029] Further, the power entropy decomposition in step S21 is calculated as follows:

[0030]

[0031]

[0032] In the formula, h esa and l esa are the high-entropy scattering amplitude component and the low-entropy scattering amplitude component, respectively, and their relationship is:

[0033]

[0034] Further, step S3 specifically includes:

[0035] S30, performing random forest classification on the extracted polarization feature map l esa and evaluating the contribution of each parameter using variable importance;

[0036] S31, visualizing the classification map after random forest classification to obtain an oil-water separation map;

[0037] S32, quantitatively evaluating the feature extraction map of multiple polarization features based on three scenarios: different thickness of oil slick, different types of oil slick, and different area of oil slick.

[0038] Further, the two parameters quantitatively evaluated in the step S32 are the Michelson contrast MC and the JM distance, and the formulas are as follows:

[0039]

[0040] In the formula, I MAX and I MIN respectively represent the maximum and minimum average polarization eigenvalues between the two target samples to be measured, and the value range of MC is 0-1.

[0041]

[0042]

[0043] In the formula, m i and m j represent the average value, and ∑ i and ∑ j represent the covariance matrix of the classes w i and w j , the superscript T represents the transpose of the matrix, and the value range of the JM distance is 0-2.

[0044] The application also provides a system for implementing the above-mentioned compact polarization SAR offshore oil spill detection method, and the system comprises:

[0045] a preprocessing module, configured to preprocess a full polarization SAR image and extract compact polarization SAR data;

[0046] an extraction and construction module, configured to extract total scattering power and entropy based on the compact polarization SAR data and construct a compact polarization SAR oil spill detector;

[0047] a classification and quantitative evaluation module, configured to perform random forest classification on the polarization feature map extracted by the compact polarization SAR detector, obtain an oil-water separation map, and quantitatively evaluate the oil spill detection capability of the oil-water separation map.

[0048] Compared with the prior art, the application has the following advantages: the compact polarization SAR offshore oil spill detection method and system provided by the application use the CP SAR data extracted from the full polarization SAR image as input, and the low scattering power l esa detector is used for oil spill detection, and compared with other types of polarization feature parameters, the low scattering power l esa combination can improve the partial defects of the entropy H w and the total scattering power g0, and the combination of the information of the two helps to improve the extraction of the oil slick information, so that the oil spill detection becomes more accurate. BRIEF DESCRIPTION OF DRAWINGS

[0049] In order to make the technical solutions in the embodiments of the present application or the prior art clearer, the accompanying drawings needed in the embodiments or prior art description will be briefly introduced. Obviously, the accompanying drawings in the following description only aim to some embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative effort on the basis of these drawings.

[0050] Figure 1 is the overall flowchart of the method for detecting marine oil spill by using compact polarimetric SAR according to the present application.

[0051] Figure 2 is the data set for verification according to the present application.

[0052] Figure 3 is the detection result according to the present application.

[0053] Figure 4 is the classification result of the detection result according to the present application.

[0054] Figure 5 is the classification accuracy of the detection result according to the present application.

[0055] Figure 6 is the quantitative evaluation result of the detection result based on MC measurement according to the present application.

[0056] Figure 7 is the quantitative evaluation result of the detection result based on JM distance according to the present application.

[0057] Figure 8 is the principle diagram of the system for detecting marine oil spill by using compact polarimetric SAR according to the present application. DETAILED DESCRIPTION

[0058] The preferred embodiments of the present application will be described in detail below with reference to the accompanying drawings, so that the advantages and features of the present application can be more easily understood by those skilled in the art, and the protection scope of the present application can be more clearly defined.

[0059] Referring to Figure 1 and Figure 2 , the present embodiment discloses a method for detecting marine oil spill by using compact polarimetric SAR, which comprises the following steps:

[0060] Step S1, pre-process the full polarimetric SAR image and extract compact polarimetric SAR data.

[0061] Specifically, this step comprises the following steps:

[0062] Step S10: Download six fully polarimetric SAR image data from the Alaska website, including two L-band ALOSPALSAR images, one C-band RADARSAT-2 image, and three L-band UAVSAR images, and extract the covariance matrix in HP mode using the polarimetric scattering matrix S.

[0063] Specifically, the polarization scattering matrix S is:

[0064]

[0065] In the formula, the scattering matrix element S xy In the diagram, x and y represent transmission and reception, and x, y ∈ {h, v}, where h and v represent horizontal and vertical polarization, respectively.

[0066] S11. Construct the Stokes vector in HP mode from the covariance matrix. The Stokes vector in HP mode is simplified polarimetric SAR data.

[0067] In HP mode, the radar transmits a circular signal and simultaneously receives two orthogonal linearly polarized signals. Considering the right circular transmission field and scattering reciprocity, its scattering vector is:

[0068]

[0069] Its covariance matrix can be written as:

[0070]

[0071] The superscript "*" indicates complex conjugation. 〈·〉 represents the global average; |·| represents the modulus of the complex signal; the superscript T indicates the transpose of the matrix. The polarization covariance matrix is ​​a Hermitian positive semi-definite matrix. Its elements can be obtained from FP data:

[0072]

[0073]

[0074]

[0075]

[0076] Where Re represents the imaginary part. Stokes uses four real numbers to represent the monochromatic electromagnetic field, which are the well-known Stokes vectors. Considering a coherent dual-polarization SAR, the SAR's transmitted field is right-circularly polarized (R), then the four Stokes parameters are:

[0077]

[0078] Step S2: Extract the total scattering power and entropy based on the simplified polarimetric SAR data, and construct a simplified polarimetric SAR oil spill detector.

[0079] Specifically, the steps include:

[0080] Step S20: Using the Stokes vector, construct five polarization features for verification: entropy H w Total scattering power g0, polarizability DoP, co-correlated polarizability ρ co and the parameter iB related to Bragg scattering CTLR .

[0081] The formulas for the five polarization characteristics are as follows:

[0082]

[0083] g0=<|S RH | 2 +|S RV | 2 >

[0084]

[0085]

[0086] iB CTLR = (g0+g3) / (g0-g3)

[0087] In the formula, λ i It is the i-th eigenvalue.

[0088] Step S21: Utilize power entropy decomposition, based on the entropy H of each unit in the response. w The contribution of H is used to decompose the total scattering power g0, based on the entropy H W The total scattering power g0 is decomposed into two components to obtain the low scattering power l. esa The low scattering power l esa This is a detector for simplified polarization SAR oil spill detection.

[0089] The formula for power entropy decomposition is:

[0090]

[0091]

[0092] In the formula, h esa and l esa These are the high-entropy scattering amplitude components and the low-entropy scattering amplitude components, respectively, and their relationship is as follows:

[0093]

[0094] Specifically, based on the previously mentioned formula, and using entropy H... W The total scattering power g0 is decomposed into two components, which is called power entropy decomposition. This yields the low-entropy scattering amplitude component l in this embodiment. esa This was applied to six images. For example... Figure 2 As shown, (a) and (b) are used to evaluate l esa The ability to detect oil spills of varying thicknesses; (c) and (d) are used to evaluate l esa The ability to detect different types of floating oil; (e)(f) used to evaluate l esa Detection capability for oil spills of different sizes.

[0095] Figure 3 The middle shows l esa The detector's visualization results across six images show that... esa The detector can combine two different directions of oil spill detection: backscattered energy and scattering mechanism, improving some of their respective shortcomings and obtaining a more obvious oil-water contrast, thereby enabling more accurate oil spill detection.

[0096] Step S3: Perform random forest classification on the polarization feature map extracted by the simplified polarimetric SAR detector to obtain an oil-water separation map, and quantitatively evaluate the oil spill detection capability of the oil-water separation map.

[0097] Specifically, the steps include:

[0098] Step S30, for low scattering power l esa The extracted polarization feature maps are used for random forest classification, and the contribution of each parameter is evaluated using variable importance.

[0099] Step S31: Visualize the classification map after random forest classification to obtain the oil-water separation map.

[0100] Step S32: Based on three scenarios: oil spills of different thicknesses, oil spills of different types, and oil spills of different areas, quantitatively evaluate the feature extraction maps of multiple polarization characteristics.

[0101] Specifically, the two parameters quantitatively evaluated in step S32 are the Michelson contrast (MC) and the JM distance, with the following formulas:

[0102]

[0103] In the formula, I MAX and I MIN These represent the maximum and minimum average polarization eigenvalues ​​between the two target samples being tested, respectively, with MC ranging from 0 to 1.

[0104]

[0105]

[0106] In the formula, m i and m j ∑ represents the average value. i and ∑ j Represents class w i and w j The covariance matrix is ​​given by the matrix T, which denotes the transpose of the matrix. The JM distance ranges from 0 to 2. A larger JM distance indicates a higher degree of distinguishability between the two target classes, and vice versa. When the JM distance is greater than 1.9, it indicates strong distinguishability between the two target classes; when the JM distance is between 1 and 1.9, it indicates good distinguishability between the two classes; and when the JM distance is between 0 and 1, it indicates weak distinguishability.

[0107] For l esa There are several reasons why this method can be used for oil spill detection. First, the scattering mechanism of oil-free sea surfaces is mainly Bragg or tilted Bragg, which results in scattered waves with polarization characteristics very close to those of fully polarized waves: DoP is close to 1, H... w Approaching 0. However, on the oil spill surface, due to damping characteristics—that is, the damping force reduces resonance—Bragg resonance waves are significantly reduced, leading to a non-Bragg mechanism. Its polarization characteristics are close to those of unpolarized waves: DoP approaches 0, H... w The entropy of an oil spill area is close to 1. Therefore, the entropy of the oil spill area is usually greater than that of the sea surface. Secondly, the total scattered power received by SAR sensors from an oil spill surface is usually lower than that from an oil-free surface. SAR sensors are very sensitive to the roughness of the sea surface. Oil slicks make the sea surface rougher because they not only suppress ocean capillary waves and gravity waves, but also reduce the surface tension and friction between the wind and the liquid surface. Therefore, oil slicks have low backscattering characteristics, which is reflected in the brightness value.

[0108] Therefore, the low-entropy scattering amplitude component l esa From g0 and (1-H w It is composed of ( ) combinations. Through information enhancement, the oil spill surface is made l esa The value is higher than that of oil-free sea surface. esa Smaller values ​​can significantly reveal the characteristics of the oil spill area. Combining these two polarization features is more effective than simply using either g0 or H. w This will allow us to obtain more useful and effective information, which will help in the extraction of oil spilled areas.

[0109] In this embodiment, random forest is a bagging ensemble method based on CART decision trees. Previous researchers have successfully applied it to land cover classification of remote sensing images. Its principle is to generate many classifiers trained on random subsets of the dataset, each classifier being generated using a training set consisting of random vectors randomly drawn with replacement from the input vectors.

[0110] In specific classification problems, the CART classification tree algorithm introduces the Gini coefficient to represent the impurity of the model, expressed as:

[0111]

[0112] Assume there are K categories, and the probability of the k-th category is p. k For a sample M, if M is divided into two parts, M1 and M2, based on a certain value 'a' of feature A, then, given feature A, the Gini coefficient of D is:

[0113]

[0114] The Gini coefficient replaces the information gain ratio of the C4.5 algorithm; a smaller Gini coefficient indicates better features. For the discrete features extracted by polarization in this paper, the approach is to repeatedly binary-divide the feature values. Each classifier is independent, so their predictions will encompass all possibilities, resulting in a good prediction. The final classification result is the mode of all classifiers' results. The random forest bagging ensemble method essentially aggregates several weak classifiers to form a strong classifier.

[0115] Variable importance is a parameter that assesses the contribution of each factor to the classification result. We use l esa After extracting the feature maps of the study area, random forest classification was performed on the feature maps.

[0116] Visualize the classification map from the previous step in ArcGIS software. Figure 4 Six images were displayed based on l esa Random forest classification and variable importance ranking of six polarization features. Figure 5 The Kappa accuracy of random forest classification for six polarization features is shown in six cases. Where l esa In all five cases, Kappa accuracy was superior. This demonstrates the rationale behind the invention, namely... esa The detector combines the advantages of scattering mechanisms and backscattering to achieve more accurate detection of floating oil through information enhancement.

[0117] Figure 6The MC measurement results for the six polarization characteristics studied are shown in the case study. The ranking results are shown among oil float thickness (a)(b), different types of oil float (c)(d)(e), and oil floats of different areas (f)(g). Overall, l esa The detector effectively distinguishes between different oil spill scenarios and outperforms five other detectors in most cases. This is because it combines two polarization features, enhancing information from both backscattering energy and scattering mechanisms, resulting in better discriminability. 1-H w and The parameters also have good discriminative ability, but overall they do not have l esa It works well.

[0118] Figure 7 The JM distance assessment results for the study area are displayed. The JM distance reflects the distinguishability between two target categories; a JM distance greater than 1.9 indicates strong distinguishability. From the results, l esa The detector performs well in inspecting the thickness of oil slicks and oil slicks of different sizes. However, its ability to distinguish between crude oil and emulsions is not ideal in detecting different types of oil slicks. Overall, in most category comparisons, l esa The JM distances of all detectors are greater than 1.9, demonstrating their ability to highlight oil float signals and suppress other similar substances. Combined with Michelson contrast analysis, this fully demonstrates... esa The detector exhibits superior performance. This demonstrates the superior performance of the power entropy decomposition parameter l. esa It is feasible and effective for oil spill detection.

[0119] See Figure 8 The present invention also provides a system for implementing the simplified polarimetric SAR marine oil spill detection method described above, comprising: a preprocessing module 1 for preprocessing a fully polarimetric SAR image and extracting simplified polarimetric SAR data; an extraction and construction module 2 for extracting total scattering power and entropy based on the simplified polarimetric SAR data and constructing a simplified polarimetric SAR oil spill detector; and a classification and quantitative evaluation module 3 for performing random forest classification on the polarization feature map extracted by the simplified polarimetric SAR detector to obtain an oil-water separation map and quantitatively evaluating the oil spill detection capability of the oil-water separation map.

[0120] This invention uses l esa By combining polarization characteristic parameters from both backscattering energy and scattering mechanism, the extraction of oil spill information can be improved, thus making oil spill detection more accurate.

[0121] Although embodiments of the present invention have been described in conjunction with the accompanying drawings, the patent owner may make various modifications or alterations within the scope of the appended claims, as long as they do not exceed the protection scope described in the claims of the present invention, they shall be within the protection scope of the present invention.

Claims

1. A compact polarimetric SAR marine oil spill detection method, characterized in that, The method comprises the following steps: S1, preprocessing the full-polarimetric SAR image and extracting the compact polarimetric SAR data; S2, extracting the total scattering power and entropy based on the compact polarimetric SAR data, and constructing a compact polarimetric SAR oil spill detector; S3, performing random forest classification on the polarimetric feature map extracted by the compact polarimetric SAR oil spill detector to obtain an oil-water separation map, and quantitatively evaluating the oil spill detection capability of the oil-water separation map; The step S1 specifically comprises: S10, downloading the full-polarimetric SAR image and extracting the covariance matrix in the HP mode by using the polarimetric scattering matrix S; S11, constructing the Stokes vector in the HP mode from the covariance matrix, wherein the Stokes vector in the HP mode is the compact polarimetric SAR data; The step S2 specifically comprises: S20, construct five kinds of polarization characteristics for verification: entropy , total scattering power , polarization degree DoP, correlation polarization coefficient and the parameter related to Bragg scattering ; S21, decompose the total scattering power according to the contribution of the entropy of each cell in the response using power entropy decomposition a detector for compact polarimetric SAR oil spill detection;​​​​​ The formula for power entropy decomposition in the step S21 is: wherein and are high-entropy and low-entropy scattered amplitude components, respectively, with the relationship: 。 2. The compact polarimetric SAR marine oil spill detection method according to claim 1, characterized in that, The polarimetric scattering matrix S in the step S10 is specifically: where the scattering matrix elements where x and y represent transmit and receive, h and v represent horizontal and vertical polarization, respectively.

3. The compact polarimetric SAR marine oil spill detection method according to claim 1, characterized in that, The Stokes vector in the HP mode in the step S11 is: 。 4. The compact polarimetric SAR marine oil spill detection method according to claim 1, characterized in that, The formula for the five polarimetric features in the step S20 is: wherein , is the ith eigenvalue.

5. The compact polarimetric SAR sea spill detection method according to claim 4, characterized in that, The step S3 specifically comprises: S30, to low scattering power The extracted polarization feature maps are classified by random forest, and the contribution of each parameter is evaluated by variable importance. S31, visualizing the classification map after random forest classification to obtain an oil-water separation map; S32, quantitatively evaluating the feature extraction map of various polarimetric features based on three scenes: different thicknesses of floating oil, different types of floating oil, and different areas of floating oil.

6. The compact polarimetric SAR sea spill detection method according to claim 5, characterized in that, The two parameters for quantitative evaluation in the step S32 are the Michelson contrast MC and the JM distance, and the formulas are respectively: wherein, and respectively represent the maximum and minimum average polarization eigenvalue between the two target samples being measured, and the value of MC ranges from 0 to 1. wherein and denotes the average value, and denotes the covariance matrix of the classes and the value of the JM distance ranges from 0 to 2.

7. A system for implementing the method of compact polarimetric SAR marine oil spill detection according to any one of claims 1 to 6, characterized in that It comprises: A preprocessing module configured to preprocess the full-polarimetric SAR image and extract the compact polarimetric SAR data; An extraction and construction module configured to extract the total scattering power and entropy based on the compact polarimetric SAR data, and construct a compact polarimetric SAR oil spill detector; A classification and quantitative evaluation module configured to perform random forest classification on the polarimetric feature map extracted by the compact polarimetric SAR detector to obtain an oil-water separation map, and quantitatively evaluate the oil spill detection capability of the oil-water separation map.

Images