SAR oil spill automatic monitoring method based on prompt feedback and pattern spot similarity

Abstract

The invention discloses an SAR oil spill automatic monitoring method based on prompt feedback and pattern spot similarity, and the method comprises the steps: S1, obtaining a sea area remote sensing image, and carrying out the preprocessing of the sea area remote sensing image, and obtaining a processed image; s2, inputting into a U-Net model to obtain an initial oil spill area screening result; s3, a user randomly selects one pattern spot, and a judgment result that the pattern spot is an oil spill area or a non-oil spill area is obtained; s4, similarity calculation is carried out, the shape feature, the texture feature and the back scattering feature of each pattern spot are obtained, and then similarity scores corresponding to other pattern spots are obtained; s5, comparing the similarity scores corresponding to the other pattern spots with a set similarity score threshold value, and performing corresponding processing; and S6, repeatedly executing the steps S3-S5 to obtain a final oil spill area monitoring result. According to the invention, oil spill detection and analysis in a large-scale area can be automatically completed by preliminarily detecting the oil spill area and introducing a small amount of manual prompt mechanisms, so that manual intervention is effectively reduced, and the monitoring efficiency and accuracy are greatly improved.

Description

Technical Field

[0001] The invention relates to the technical field of oil spill area detection, and in particular to a SAR oil spill automatic monitoring method based on prompt feedback and image spot similarity. Background Art

[0002] Marine oil spill pollution is a serious environmental disaster that may have long-term impacts on marine ecosystems and coastal areas. Therefore, rapid and accurate detection and monitoring of oil spills is crucial to reducing the damage to the environment caused by oil spills and taking timely clean-up actions. With the advancement of SAR technology, its ability to penetrate clouds and haze around the clock and in all weather conditions has become one of the main means of oil spill monitoring. In the actual application of oil spill monitoring, the goal is to achieve an efficient and automated monitoring process, reduce dependence on human intervention, and ensure high recall and high accuracy of detection. However, existing SAR oil spill monitoring methods usually rely on a large amount of manual annotation and complex detection algorithms, which are time-consuming and labor-intensive, and often lack processing efficiency and accuracy when faced with large-scale marine monitoring. For example, current oil spill detection algorithms mostly use machine learning or deep learning models, especially models based on convolutional neural networks (CNN). These models face some challenges when applied to SAR oil spill monitoring. First, the oil spill areas in SAR images are usually irregular patches due to environmental factors such as lighting, wind speed, and sea conditions, which makes the detection of oil spill areas highly uncertain. Secondly, they rely too much on manually annotated data, which requires a huge amount of annotation work and is difficult to reduce false alarms while ensuring a high recall rate, which is not conducive to practical engineering applications. Summary of the invention

[0003] The present invention provides a SAR oil spill automatic monitoring method and device based on prompt feedback and image spot similarity, so as to overcome the technical problems that the existing SAR oil spill monitoring method usually relies on a large amount of manual annotation and complex detection algorithms, which is time-consuming and labor-intensive, and the processing efficiency and accuracy are often insufficient when facing large-scale sea area monitoring.

[0004] In order to achieve the above object, the technical solution of the present invention is:

[0005] A SAR oil spill automatic monitoring method based on prompt feedback and image similarity, the specific steps include:

[0006] S1: acquiring a sea area remote sensing image based on a SAR satellite, and preprocessing the sea area remote sensing image to obtain a processed image;

[0007] S2: inputting the processed image into the U-Net model to obtain an initial oil spill area screening result;

[0008] S3: The user selects a spot from the initial oil spill area screening results, and combines the processed image to determine whether the spot is an oil spill area or a non-oil spill area;

[0009] S4: performing similarity calculation based on the spot for which the user gives the judgment result and other spots, including obtaining the shape feature, texture feature and backscattering feature of each spot, and obtaining the similarity scores corresponding to other spots based on the shape feature, texture feature and backscattering feature of each spot;

[0010] S5: Compare the similarity scores corresponding to the other spots with the set similarity score threshold. If the spots are compared with the spots of the oil spill area given by the user, segmentation processing is performed on the spots that meet the set similarity score after comparison; if the spots are compared with the spots of the non-oil spill area given by the user, elimination processing is performed on the spots that meet the set similarity score after comparison;

[0011] S6: Update the initial oil spill area and repeat S3-S5. When all the spots are judged and processed, the final monitoring result of the oil spill area is obtained.

[0012] Furthermore, in S4, the Hausdorff distance is used to calculate the shape features corresponding to each of the other spots, and the calculation formula is:

[0013]

[0014] Among them, A and B represent A respectively. i The image spot is the image spot where the user gives the judgment result and B j The contour point set of the patch, a represents a specific point in A, and b represents a specific point in B;

[0015] The initial value d of the shape feature of the spot given the judgment result by the user is calculated based on the shape features of other spots. H (A i ), the calculation formula is:

[0016]

[0017] Among them, j∈[1,ni-1], n is the total number of oil spill spots.

[0018] Furthermore, in S4, the texture features corresponding to other spots are calculated through the gray level co-occurrence matrix, including:

[0019] Contrast calculation, the formula is:

[0020]

[0021] Correlation calculation, the formula is:

[0022]

[0023] Homogeneity calculation formula is:

[0024]

[0025] Where P(i,j) is the element at position (i,j) in the gray-level co-occurrence matrix, i and j are gray levels, N is the number of gray levels, and μ i and μ j is the mean of gray levels i and j, σ i and σ j is the standard deviation of gray levels i and j.

[0026] Further, in S4, the backscattering characteristics corresponding to other spots are calculated based on the backscattering coefficient, and the calculation formula is:

[0027]

[0028] Where P r It represents the power received by the SAR satellite, and Area represents the pixel area.

[0029] Furthermore, the process of obtaining the similarity scores corresponding to other spots based on the shape features, texture features and backscattering features of each spot is as follows:

[0030] Calculate the similarity of the shape features between the image patch for which the user gives the judgment result and other image patches. The formula is:

[0031]

[0032] Among them, d H (A i ,B j ) is A i B j Hausdorff distance of the shape contour of the patch;

[0033] Calculate the similarity between the texture features of the image patch for which the user gives the judgment result and other image patches. The formula is:

[0034]

[0035] Among them, F A,k and F B,k A i B j The value of the spot on the kth texture feature, n is the total number of texture features;

[0036] Calculate the similarity between the spot for which the user gives the judgment result and other spots in terms of backscattering characteristics. The formula is:

[0037]

[0038] Among them, σ 0,A and σ 0,B A i B j Backscatter coefficient of the pattern;

[0039] The similarity scores of each feature are weighted averaged to obtain a comprehensive similarity score, which is expressed as:

[0040] S=ω texture ·S texture +ω shape +ω SAR ·S SAR

[0041] Among them, ω texture ,ω shape ,ω SAR are the weights of each feature, and satisfy ω texture +ω shape +ω SAR =1.

[0042] Furthermore, each time the similarity calculation is performed between the image patch for which the user gives a judgment result and other image patches, the weights of the shape features, texture features, and backscattering features of the image patch are updated according to the judgment result given by the user, and the next similarity score is calculated based on the updated weights. The weight update formula is:

[0043] w texture,t+1 =w texture,t +α·Δw texture

[0044] w shape,t+1 =w shape,t +β·Δw shape

[0045] w SAR,t+1 =w SAR,t +γ·Δw SAR

[0046] Among them, α, β, γ are the adjustment step sizes, Δw texture , Δw shape , Δw SAR is the adjusted weight increment;

[0047] The formula for calculating the next similarity score based on the updated weight is:

[0048]

[0049] in, Represents the new user's judgment result after iteration; R i represents the initial set of all oil spills; w texture,t+1 、w shape,t+1 and w SAR,t+1 are the updated feature weights.

[0050] Beneficial effects: The present invention combines the U-Net model to perform preliminary oil spill area detection and introduces a small amount of manual prompt mechanism to achieve automatic monitoring of oil spill areas. Users only need to input prompts for a small number of spots to automatically complete large-scale oil spill detection and analysis, effectively reducing manual intervention and greatly improving monitoring efficiency; introducing spot similarity calculation, combined with user prompt information, similarity analysis of unlabeled spots and automatic segmentation or merging of other similar areas. This method can reduce missed reports and false alarms while ensuring a high recall rate, and is particularly suitable for complex scenes and variable image features in offshore oil spill monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0052] Figure 1 It is a flow chart of the SAR oil spill automatic monitoring method based on prompt feedback and image spot similarity in the present invention;

[0053] Figure 2 It is a remote sensing image of the sea area acquired by the SAR satellite in the embodiment of the present invention;

[0054] Figure 3 This is a diagram of the initial oil spill area screening result in an embodiment of the present invention;

[0055] Figure 4 This is the final oil spill area monitoring result diagram in the embodiment of the present invention. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0057] This embodiment provides a SAR oil spill automatic monitoring method based on prompt feedback and image similarity, such as Figure 1 As shown, the specific steps include:

[0058] S1: Based on SAR satellite acquisition Figure 2 The remote sensing image of the sea area shown in the figure is preprocessed to obtain a processed image;

[0059] Specifically, in this embodiment, the sea area remote sensing image is preprocessed by EnLee filtering and histogram equalization method to improve image quality and reduce interference information;

[0060] S2: Input the processed image into the U-Net model to obtain the initial oil spill area screening result, such as Figure 3 As shown;

[0061] Specifically, the output of the U-Net model can be expressed as a segmentation mask M pred , where each pixel p has a probability value P(p∈O), indicating whether the pixel p belongs to the oil spill area O, and the formula is expressed as:

[0062] M pred (p) = P(p∈O)

[0063] S3: The user selects a spot from the initial oil spill area screening results, and combines the processed image to determine whether the spot is an oil spill area or a non-oil spill area;

[0064] Specifically, the user can select a pixel point or area through the graphical interface and input the prompt information. The prompt can be in the form of a, correct oil spill area; b, incorrect oil spill area. point (x, y), indicating that the pixel (x, y) is a correct oil spill point or an incorrect non-oil spill point; area prompt T area (x1,y1,x2,y2), indicating that a certain rectangular area is a correct oil spill area or an incorrect non-oil spill area.

[0065] S4: performing similarity calculation based on the spot for which the user gives the judgment result and other spots, including obtaining the shape feature, texture feature and backscattering feature of each spot, and obtaining the similarity scores corresponding to other spots based on the shape feature, texture feature and backscattering feature of each spot;

[0066] Specifically, from the perspective of the shape of the oil spill area, the shape characteristics of the oil spill area may present irregular geometric forms in the actual scene, but are usually relatively continuous and non-dispersed patches. Therefore, it is necessary to use shape features to describe the contour and distribution of the oil spill area. From the perspective of the texture of the oil spill, the oil spill area is usually smooth and has a relatively uniform texture in the SAR image, which contrasts with the surrounding sea surface. Therefore, the texture features extracted using the gray level co-occurrence matrix can well describe the uniformity and texture changes of the oil spill area. The oil spill area is usually smoother than the sea surface and shows a lower contrast; the texture of the oil spill area tends to be uniform and shows a higher homogeneity; homogeneity describes the uniformity of the image grayscale in the local area and is usually used to measure the smoothness of the texture. Since the oil spill area in the SAR image has a unique backscattering characteristic. The oil spill surface will suppress the sea surface waves caused by the wind, resulting in dark spots in the SAR image. Based on this phenomenon, the backscattering coefficient can be used to distinguish the oil spill area from the normal sea surface. The oil spill area usually shows a lower backscattering value because the smooth surface of the oil spill area reduces the scattering of electromagnetic waves. In summary, the present invention combines shape features, texture features and backscattering features to calculate the similarity of oil spill spots.

[0067] In a specific embodiment, in S4, the shape features corresponding to other spots are calculated using the Hausdorff distance, and the calculation formula is:

[0068]

[0069] Among them, A and B represent A respectively. i The image spot is the image spot where the user gives the judgment result and B j The contour point set of the patch, a represents a specific point in patch A, and b represents a specific point in patch B;

[0070] The initial value d of the shape feature of the spot given the judgment result by the user is calculated based on the shape features of other spots. H (A i ), the calculation formula is:

[0071]

[0072] Among them, j∈[1,ni-1], n is the total number of oil spill spots.

[0073] In a specific embodiment, in S4, the texture features corresponding to other spots are calculated by using the gray level co-occurrence matrix, including:

[0074] Contrast calculation, the formula is:

[0075]

[0076] Correlation calculation, the formula is:

[0077]

[0078] Homogeneity calculation formula is:

[0079]

[0080] Where P(i,j) is the element at position (i,j) in the gray-level co-occurrence matrix, i and j are gray levels, N is the number of gray levels, and μ i and μ j is the mean of gray levels i and j, σ i and σ j is the standard deviation of gray levels i and j.

[0081] In a specific embodiment, in S4, the backscattering characteristics corresponding to other spots are calculated based on the backscattering coefficient, and the calculation formula is:

[0082]

[0083] Where P r It represents the power received by the SAR satellite, and Area represents the pixel area.

[0084] In a specific embodiment, the process of obtaining the similarity scores corresponding to other spots based on the shape feature, texture feature and backscattering feature of each spot is as follows:

[0085] Calculate the similarity of the shape features between the image patch for which the user gives the judgment result and other image patches. The formula is:

[0086]

[0087] Among them, d H (A i ,B j ) is A i B j Hausdorff distance of the shape contour of the patch;

[0088] Calculate the similarity between the texture features of the image patch for which the user gives the judgment result and other image patches. The formula is:

[0089]

[0090] Among them, F A,k and F B,k A i B j The value of the spot on the kth texture feature, n is the total number of texture features;

[0091] Calculate the similarity between the spot for which the user gives the judgment result and other spots in terms of backscattering characteristics. The formula is:

[0092]

[0093] Among them, σ 0,A and σ 0,B A i B j Backscatter coefficient of the pattern;

[0094] The similarity scores of each feature are weighted averaged to obtain a comprehensive similarity score, which is expressed as:

[0095] S=ω texture ·S texture +ω shape +ω SAR ·S SAR

[0096] Among them, ω texture ,ω shape ,ω SAR are the weights of each feature, and satisfy ω texture +ω shape +ω SAR =1.

[0097] In a specific embodiment, each time a similarity calculation is performed between a spot for which a user gives a judgment result and other spots, the weights of the shape features, texture features, and backscattering features of the spot are updated according to the judgment result given by the user, and the next similarity score is calculated based on the updated weights. The weight update formula is:

[0098] w texture,t+1 =w texture,t +α·Δw texture

[0099] w shape,t+1 =w shape,t +β·Δw shape

[0100] w SAR,t+1 =w SAR,t +γ·Δw SAR

[0101] Among them, α, β, γ are the adjustment step sizes, Δw texture , Δw shape , Δw SAR is the adjusted weight increment;

[0102] The formula for calculating the next similarity score based on the updated weight is:

[0103]

[0104] in, Represents the new user's judgment result after iteration; R i represents the initial set of all oil spills; w texture,t+1 、w shape,t+1 and w SAR,t+1 are the updated feature weights.

[0105] Specifically, an initial similarity calculation is performed in S4 to obtain a comprehensive similarity score. From the similarity calculation formula in S4, it can be seen that this embodiment takes into account texture, shape and backscattering features. The image patch for which the user gives the judgment result contains information from multiple aspects, such as areas with obvious texture features, areas with unique shape features, and areas with significant backscattering features, thereby updating the weight (w texture , w shape , w SAR ), the weight is updated every time the user prompts an oil spill patch. After multiple prompts from the user, the feature weight is continuously adjusted to achieve more accurate oil spill detection. Through the feedback iteration mechanism, it can cope with the challenges of oil spill monitoring in complex offshore scenes and dynamically adjust the segmentation strategy. This adaptability enables it to be flexibly applied to monitoring tasks under different oil spill forms and environmental conditions, improving the robustness of the system.

[0106] S5: Compare the similarity scores corresponding to the other spots with the set similarity score threshold. If the spots are compared with the spots of the oil spill area given by the user, segmentation processing is performed on the spots that meet the set similarity score after comparison; if the spots are compared with the spots of the non-oil spill area given by the user, elimination processing is performed on the spots that meet the set similarity score after comparison;

[0107] In a specific embodiment, in S5, the similarity scores corresponding to the other spots are compared with the set similarity score threshold. If the spots are compared with the spots of the oil spill area given by the user, the spots that meet the set similarity score are segmented after comparison; if the spots are compared with the spots of the non-oil spill area given by the user, the spots that meet the set similarity score are eliminated after comparison, which is expressed as:

[0108]

[0109] Among them, τ split is the segmentation threshold; S(p) is the similarity score of each oil spill patch; M final (p) is the updated initial oil spill area.

[0110] S6: Update the initial oil spill area and repeat S3-S5. When all the spots are judged and processed, the final monitoring results of the oil spill area are obtained, such as Figure 4 shown.

[0111] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A SAR oil spill automatic monitoring method based on prompt feedback and image similarity, characterized in that: The specific steps include: S1: acquiring a sea area remote sensing image based on a SAR satellite, and preprocessing the sea area remote sensing image to obtain a processed image; S2: inputting the processed image into the U-Net model to obtain an initial oil spill area screening result; S3: The user selects a spot from the initial oil spill area screening results, and combines the processed image to determine whether the spot is an oil spill area or a non-oil spill area; S4: performing similarity calculation based on the spot for which the user gives the judgment result and other spots, including obtaining the shape feature, texture feature and backscattering feature of each spot, and obtaining the similarity scores corresponding to other spots based on the shape feature, texture feature and backscattering feature of each spot; S5: Compare the similarity scores corresponding to the other spots with the set similarity score threshold. If the spots are compared with the spots of the oil spill area given by the user, segmentation processing is performed on the spots that meet the set similarity score after comparison; if the spots are compared with the spots of the non-oil spill area given by the user, elimination processing is performed on the spots that meet the set similarity score after comparison; S6: Update the initial oil spill area and repeat S3-S5. When all the spots are judged and processed, the final monitoring result of the oil spill area is obtained.

2. The SAR oil spill automatic monitoring method based on prompt feedback and pattern similarity according to claim 1 is characterized in that: In S4, the Hausdorff distance is used to calculate the shape features corresponding to other spots. The calculation formula is: Among them, A and B represent A respectively. i The image spot is the image spot where the user gives the judgment result and B j The contour point set of the patch, a represents a specific point in A, and b represents a specific point in B; The initial value d of the shape feature of the spot given the judgment result by the user is calculated based on the shape features of other spots. H (A i ), the calculation formula is: Among them, j∈[1,ni-1], n is the total number of oil spill spots.

3. The SAR oil spill automatic monitoring method based on prompt feedback and pattern similarity according to claim 2 is characterized in that: In S4, the texture features corresponding to other spots are calculated through the gray level co-occurrence matrix, including: Contrast calculation, the formula is: Correlation calculation, the formula is: Homogeneity calculation formula is: Where P(i,j) is the element at position (i,j) in the gray-level co-occurrence matrix, i and j are gray levels, N is the number of gray levels, and μ i and μ j is the mean of gray levels i and j, σ i and σ j is the standard deviation of gray levels i and j.

4. The SAR oil spill automatic monitoring method based on prompt feedback and pattern similarity according to claim 3 is characterized in that: In S4, the backscattering characteristics corresponding to other spots are calculated based on the backscattering coefficient. The calculation formula is: Where P r It represents the power received by the SAR satellite, and Area represents the pixel area.

5. The SAR oil spill automatic monitoring method based on prompt feedback and pattern similarity according to claim 4 is characterized in that: The process of obtaining the similarity scores corresponding to other spots based on the shape features, texture features and backscattering features of each spot is as follows: Calculate the similarity of the shape features between the image patch for which the user gives the judgment result and other image patches. The formula is: Among them, d H (A i ,B j ) is A i B j Hausdorff distance of the shape contour of the patch; Calculate the similarity between the texture features of the image patch for which the user gives the judgment result and other image patches. The formula is: Among them, F A,k and F B,k A i B j The value of the spot on the kth texture feature, n is the total number of texture features; Calculate the similarity between the spot for which the user gives the judgment result and other spots in terms of backscattering characteristics. The formula is: Among them, σ 0,A and σ 0,B A i B j Backscatter coefficient of the pattern; The similarity scores of each feature are weighted averaged to obtain a comprehensive similarity score, which is expressed as: S=ω texture ·S texture +oh shape +oh SAR ·S SAR Among them, ω texture ,ω shape ,ω SAR are the weights of each feature, and satisfy ω texture +ω shape +ω SAR =1.

6. The SAR oil spill automatic monitoring method based on prompt feedback and pattern similarity according to claim 5 is characterized in that: Each time the similarity calculation is performed between the image patch for which the user has given a judgment result and other image patches, the weights of the shape features, texture features, and backscattering features of the image patch are updated according to the judgment result given by the user, and the next similarity score is calculated based on the updated weights. The weight update formula is: In texture,t+1 =in texture,t +α Δw texture In shape,t+1 =in shape,t +β Δw shape In SAR,t+1 =in SAR,t +γ Δw SAR Among them, α, β, γ are the adjustment step sizes, Δw texture , Δw shape , Δw SAR is the adjusted weight increment; The formula for calculating the next similarity score based on the updated weight is: in, Represents the new user's judgment result after iteration; R i represents the initial set of all oil spills; w texture,t+1 、w shape,t+1 and w SAR,t+1 are the updated feature weights.

Images