Small sample SAR oil spill segmentation method based on population density analysis
Through the small sample SAR oil spill segmentation method based on population density analysis, combined with semantic segmentation network and local density characteristics, the problem that the existing technology is difficult to accurately identify oil spills under small sample conditions is solved, and efficient and accurate oil spill segmentation effect is achieved.
Patent Information
- Application Number
- CN202411821586.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-11
- Publication Date
- 2025-04-01
AI Technical Summary
The existing SAR image classification method is difficult to achieve efficient and accurate identification of oil spills under small sample conditions, and is easily disturbed by noise and imaging environment, resulting in insufficient segmentation accuracy and stability.
A small sample SAR oil spill segmentation method based on population density analysis is adopted. By comparing the global density of the intensity SAR image with the preset background threshold, the potential oil spill area is extracted, and combined with the semantic segmentation network and local density characteristics, efficient and accurate identification of oil spills is achieved.
It improves the accuracy of oil spill recognition, reduces dependence on deep learning network computing resources, enhances the application of the method under small sample conditions, and reduces the impact of noise interference on segmentation results.
Smart Images
Figure CN119741492A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of computer vision technology, and in particular to a small sample SAR oil spill segmentation method based on population density analysis. Background Art
[0002] As an active microwave sensor, synthetic aperture radar (SAR) uses radar waves to transmit and receive reflected signals, and obtains the spatial information of the target by analyzing the reflected waves to achieve characteristic analysis and detection of the target. SAR has the ability to work all day and all night, and can penetrate clouds, fog and other adverse weather conditions. It is of great significance for monitoring oil spills on the sea surface. In SAR images, the presence of oil film will affect the Bragg scattering of the sea surface and have a damping effect on short gravity waves and capillary waves, so that oil spills usually appear as dark areas on SAR images. This feature is the key principle of SAR oil spill identification.
[0003] Existing SAR image classification methods mainly include classification based on multi-source feature analysis and classification based on deep learning. Among them, classification based on multi-source feature analysis relies on backscattering information or texture background information. Such methods can extract the difference characteristics between oil spills and seawater background to a certain extent, but are easily disturbed by speckle noise and imaging environment, resulting in insufficient segmentation accuracy and stability. Classification based on deep learning can significantly improve the accuracy of oil spill detection by automatically learning features in images. However, due to its training characteristics, deep learning technology has limitations in practical application scenarios with small sample conditions.
[0004] Therefore, a method is needed to achieve efficient and accurate identification of oil spills under small sample conditions. Summary of the invention
[0005] In view of this, the present invention provides a small sample SAR oil spill segmentation method based on population density analysis, which mines the distribution differences between oil spill areas and seawater on SAR images through population density, reveals the spatial clustering of different target areas, and thus realizes efficient and accurate identification of oil spills.
[0006] To this end, the present invention provides the following technical solutions:
[0007] A small sample SAR oil spill segmentation method based on population density analysis, comprising:
[0008] Preprocess and enhance the original SAR image to obtain the intensity SAR image;
[0009] By comparing the global density of the intensity SAR image with the preset background threshold, it is determined whether there is a potential oil spill area in the intensity SAR image;
[0010] By narrowing the processing range through area restriction, the potential oil spill area image is extracted from the intensity SAR image of the potential oil spill area;
[0011] Obtain preliminary segmentation results of potential oil spill area images through semantic segmentation network;
[0012] Construct pixel density distribution map of potential oil spill area image;
[0013] The pixel density distribution map of the potential oil spill area image is segmented by presetting the local density threshold;
[0014] The pixel density distribution map of the potential oil spill area image is used to correct the boundary of the preliminary segmentation result and obtain the oil spill segmentation result map.
[0015] Furthermore, the step of comparing the global density of the intensity SAR image with a preset background threshold to determine whether there is a potential oil spill area in the intensity SAR image includes:
[0016] Calculate the global density of an intensity SAR image:
[0017]
[0018] Where M and N represent the number of rows and columns of the image, respectively, and I(x,y) represents the pixel value at position (x,y);
[0019] When the global density of the intensity SAR image is less than the preset background threshold, it is judged that there is a potential oil spill area in the intensity SAR image.
[0020] Further, the extracting of the potential oil spill area image from the intensity SAR image of the potential oil spill area by narrowing the processing range through area restriction includes:
[0021] ROI={(x,y)|I(x,y)≤θ}
[0022] Where θ represents the preset background threshold, ROI is the potential oil spill area, and I(x,y) represents the pixel value at position (x,y).
[0023] Furthermore, the loss function of the semantic segmentation network is:
[0024]
[0025] Where S is the number of pixels in the image, y i is the actual label, p i is the predicted probability.
[0026] Furthermore, the step of constructing a pixel density distribution map of a potential oil spill area image includes:
[0027] Calculate the local density of each pixel to construct a local density feature map;
[0028] Normalize the local density feature map.
[0029] Furthermore, the local density of each pixel is calculated by scanning the potential oil spill area image through a sliding window:
[0030]
[0031] Where ρ(x,y) is the local density at (x,y), I(x,y) is the pixel value at position (x,y) in the sliding window, and n is the number of pixels in the sliding window.
[0032] Furthermore, the setting of the local density threshold to segment the pixel density distribution map of the potential oil spill area image includes:
[0033] If ρ(x,y)≥T high , then the area is an oil spill area;
[0034] If ρ(x,y)≤T low , then the area is seawater area.
[0035] Furthermore, the constructing of the pixel density distribution map of the potential oil spill area image also includes: performing Gaussian smoothing processing on the pixel density distribution map of the potential oil spill area image.
[0036] Furthermore, the preprocessing of the original SAR image includes:
[0037] Orbit correction, terrain correction, radiation calibration and filtering and denoising;
[0038] The image enhancement of the original SAR image comprises:
[0039] Adjust the brightness of the original SAR image by gamma correction;
[0040] The contrast of the original SAR image is improved by using the histogram equalization algorithm.
[0041] Furthermore, the method of using the segmentation result of the pixel density distribution map of the potential oil spill area image to correct the boundary of the preliminary segmentation result to obtain the oil spill segmentation result map includes:
[0042] For the same pixel point in the potential oil spill area image, if the preliminary segmentation result is different from the segmentation result of the pixel density distribution map of the potential oil spill area image, the segmentation result of the pixel density distribution map of the potential oil spill area image shall be used as the oil spill segmentation result.
[0043] Advantages and positive effects of the present invention:
[0044] The present invention uses image group density as the feature of oil spill segmentation to describe the morphological structure, spatial distribution and other characteristics of oil spill and seawater; improves the accuracy of oil spill identification; and solves the problem of insufficient accuracy caused by image noise in the prior art when segmenting oil spill and seawater based on image texture features.
[0045] The present invention uses global density to analyze the spatial distribution characteristics of SAR images, quickly locates potential oil spill areas, reduces the volume of SAR image data to be processed, and reduces the burden of computing resources required to train deep learning networks, making the method more practical. It also combines group density with deep semantic segmentation technology, uses local density features to capture richer boundary details under small sample conditions, and assists the U-Net network to achieve refined segmentation; reduces the dependence of deep learning networks on large amounts of labeled data, and improves the performance of the algorithm under data scarcity conditions. BRIEF DESCRIPTION OF THE DRAWINGS
[0046] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. 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.
[0047] Figure 1 This is a flowchart of the original SAR image preprocessing and image enhancement processing in an embodiment of the present invention;
[0048] Figure 2 A flowchart of obtaining preliminary segmentation based on global density in an embodiment of the present invention;
[0049] Figure 3 This is a flow chart of correcting the preliminary segmentation result based on the local density feature in an embodiment of the present invention;
[0050] Figure 4 Flow chart of the oil spill segmentation method in an embodiment of the present invention. DETAILED DESCRIPTION
[0051] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme 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 only 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 should fall within the scope of protection of the present invention.
[0052] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.
[0053] The present invention provides a small sample SAR oil spill segmentation method based on group density analysis, which comprehensively considers the differences in spatial distribution and morphological structure between oil spill and seawater, uses group density as the segmentation feature of seawater and oil spill, quantitatively reflects the differences in spatial distribution and morphological structure between oil spill and seawater, and realizes efficient and accurate identification of oil spill. At the same time, it assists deep semantic segmentation technology to achieve efficient oil spill segmentation performance under the condition of insufficient data.
[0054] Combination Figure 4 The small sample SAR oil spill segmentation method based on population density analysis of the present invention comprises the following steps:
[0055] Step 1: Preprocess and enhance the original SAR image to obtain the intensity SAR image:
[0056] The original SAR image is calibrated by radiation and corrected by filtering; image enhancement is performed to improve the contrast of the oil spill area, enhance the visibility of the oil spill features, reduce the interference of background noise on the segmentation results, and obtain an intensity SAR image.
[0057] Step 2: Extract the potential oil spill area image from the intensity SAR image and perform preliminary segmentation on the potential oil spill area image:
[0058] Taking the intensity SAR image as input, the global density of the intensity SAR image is calculated, the potential oil spill area image in the intensity SAR image is locked, and the data volume is reduced;
[0059] The potential oil spill area image and real label data are used as input to train the U-Net semantic segmentation network, perform pixel-level segmentation on the potential oil spill area image, and obtain the preliminary segmentation result of the potential oil spill area image.
[0060] Step 3: construct a pixel density distribution map of the potential oil spill area image, segment the pixel density distribution map of the potential oil spill area image, and use the preliminary segmentation result to correct the boundary of the pixel density distribution map of the potential oil spill area image to obtain the oil spill segmentation result map.
[0061] The method of the present invention is further described with specific application examples:
[0062] Combination Figure 1 As shown, in this embodiment, S1 performs preprocessing and image enhancement on the original SAR image, including:
[0063] S11. Preprocess the original SAR image and perform denoising.
[0064] S111. Use SAR impact processing software SNAP to perform orbit correction, terrain correction, and radiation calibration on the original SAR image;
[0065] S112. Set different window sizes (3x3, 5x5 and 7x7) to perform Refined-Lee filtering on the SAR images processed by S111. Experiment step by step and select the optimal filter window size to eliminate the speckle noise in the SAR image to the greatest extent and optimize the visibility of the oil spill area.
[0066] S12. Perform data enhancement on the preprocessed SAR image.
[0067] S121. Use the histogram equalization algorithm to enhance the contrast of the oil spill area, highlight the oil spill area, and improve visibility.
[0068] S122. Use gamma correction to adjust the brightness range, optimize the difference between the oil spill area and the background, and obtain a strong SAR image.
[0069] Combination Figure 2 As shown, in this embodiment, S2 extracts the potential oil spill area image from the intensity SAR image and performs preliminary segmentation on the potential oil spill area image, including:
[0070] S21. Calculate the global density of the intensity SAR image:
[0071]
[0072] Where M and N represent the number of rows and columns of the image, respectively, and I(x,y) represents the pixel value at position (x,y).
[0073] The global density reflects the average density of the intensity SAR image. The higher the value, the brighter the image as a whole, and the lower the value, the darker the image as a whole.
[0074] S22. According to the global density distribution characteristics of the intensity SAR image, a background threshold θ is set; if the global density D≤θ, it is considered that there is a potential oil spill area in the intensity SAR image.
[0075] S23. performing regional restriction on the intensity SAR image of the potential oil spill area to reduce the processing range and obtain an image of the potential oil spill area;
[0076] If the pixel intensity in the intensity SAR image is less than the background threshold θ, it is a potential oil spill area:
[0077] ROI={(x,y)|I(x,y)≤θ}
[0078] Where ROI is the potential oil spill area, and I(x,y) represents the pixel value at position (x,y).
[0079] S24. Construct a U-Net semantic segmentation network, take the potential oil spill area image and label data as input, and use the cross entropy loss function as the loss function for segmentation training. The cross entropy loss function is defined as:
[0080]
[0081] Where S is the number of pixels in the image, y i is the actual label, P i is the predicted probability.
[0082] The U-Net network outputs the segmentation result of each pixel, generates preliminary dark area segmentation results, and identifies the location and shape of potential oil spill areas.
[0083] Combination Figure 3 As shown, in this embodiment, S3 constructs a pixel density distribution map of the potential oil spill area image, and segments the pixel density distribution map of the potential oil spill area image; and uses the preliminary segmentation result to correct the boundary of the pixel density distribution map of the potential oil spill area image to obtain an oil spill segmentation result map.
[0084] S31. Scan the potential oil spill area through sliding windows (3x3, 5x5 and 7x7) to calculate the local density of each pixel in the potential oil spill area image;
[0085]
[0086] Where ρ(x,y) is the local density at (x,y), I(x,y) is the pixel value at position (x,y) in the sliding window, and n is the number of pixels in the sliding window.
[0087] Based on the local density of each pixel, a local density feature map is constructed.
[0088] S32. Normalize the local density feature map so that the local density values of all pixels are in the range of [0, 1]. At the same time, use the Gaussian smoothing method to reduce the impact of local noise, ensure that the density feature map is smoother, and obtain the pixel density distribution map of the potential oil spill area image.
[0089] The normalization formula is as follows:
[0090]
[0091] The Gaussian smoothing formula is as follows:
[0092]
[0093] Among them, ω(i,j) is the weight of the Gaussian kernel function.
[0094] S33. In the pixel density distribution map of the potential oil spill area image, the oil spill usually appears as a wide, high-density area with spatial coherence, while the seawater distribution has more unevenness and local fluctuations, with lower local density. Setting the local density threshold quantitatively describes the above differences between seawater and oil spill.
[0095] S34. Set different local density thresholds to obtain the pixel density distribution map segmentation results of the potential oil spill area image. The specific judgment conditions are:
[0096] If ρ(x,y)≥T high , then the area is an oil spill area;
[0097] If ρ(x,y)≤T low , then the area is seawater area.
[0098] S35. The segmentation result of the pixel density distribution map of the potential oil spill area image is used to correct the preliminary segmentation result of the potential oil spill area.
[0099] For the same pixel point in the potential oil spill area image, if the preliminary segmentation result is different from the segmentation result of the pixel density distribution map of the potential oil spill area image, the segmentation result of the pixel density distribution map of the potential oil spill area image shall be used as the oil spill segmentation result.
[0100] S36. For areas with high local density but not accurately covered by the initial segmentation results, morphological operations are used to adjust the segmentation boundaries to achieve accurate segmentation of the oil spill area.
[0101] The method of the present invention outputs the segmentation result of each pixel point through the U-Net network to obtain the preliminary segmentation result of the potential oil spill area; and uses the local density distribution map to correct the preliminary segmentation result, thereby improving the accuracy of the oil spill segmentation result.
[0102] 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 small sample SAR oil spill segmentation method based on population density analysis, characterized in that: include: Preprocess and enhance the original SAR image to obtain the intensity SAR image; By comparing the global density of the intensity SAR image with the preset background threshold, it is determined whether there is a potential oil spill area in the intensity SAR image; By narrowing the processing range through area restriction, the potential oil spill area image is extracted from the intensity SAR image of the potential oil spill area; Obtain preliminary segmentation results of potential oil spill area images through semantic segmentation network; Construct pixel density distribution map of potential oil spill area image; The pixel density distribution map of the potential oil spill area image is segmented by presetting the local density threshold; The pixel density distribution map of the potential oil spill area image is used to correct the boundary of the preliminary segmentation result and obtain the oil spill segmentation result map.
2. According to the small sample SAR oil spill segmentation method based on population density analysis as described in claim 1, it is characterized in that: The method of comparing the global density of the intensity SAR image with a preset background threshold to determine whether there is a potential oil spill area in the intensity SAR image includes: Calculate the global density of an intensity SAR image: Where M and N represent the number of rows and columns of the image, respectively, and I(x,y) represents the pixel value at position (x,y); When the global density of the intensity SAR image is less than the preset background threshold, it is judged that there is a potential oil spill area in the intensity SAR image.
3. According to the small sample SAR oil spill segmentation method based on population density analysis as described in claim 1, it is characterized in that: The method of extracting the potential oil spill area image from the intensity SAR image of the potential oil spill area by narrowing the processing range through area restriction includes: ROI={(x,y)|I(x,y)≤θ} Where θ represents the preset background threshold, ROI is the potential oil spill area, and I(x,y) represents the pixel value at position (x,y).
4. The small sample SAR oil spill segmentation method based on population density analysis according to claim 1, characterized in that: The loss function of the semantic segmentation network is: Where S is the number of pixels in the image, y i is the actual label, p i is the predicted probability.
5. The small sample SAR oil spill segmentation method based on population density analysis according to claim 1, characterized in that: The step of constructing a pixel density distribution map of a potential oil spill area image includes: Calculate the local density of each pixel to construct a local density feature map; Normalize the local density feature map.
6. The small sample SAR oil spill segmentation method based on population density analysis according to claim 5, characterized in that: The local density of each pixel is calculated by scanning the potential oil spill area image through a sliding window: Where ρ(x,y) is the local density at (x,y), I(x,y) is the pixel value at position (x,y) in the sliding window, and n is the number of pixels in the sliding window.
7. The small sample SAR oil spill segmentation method based on population density analysis according to claim 1, characterized in that: The step of setting a local density threshold to segment the pixel density distribution map of the potential oil spill area image includes: If ρ(x,y)≥T high , then the area is an oil spill area; If ρ(x,y)≤T low , then the area is a seawater area.
8. The small sample SAR oil spill segmentation method based on population density analysis according to claim 5, characterized in that: The method of constructing the pixel density distribution map of the potential oil spill area image also includes: performing Gaussian smoothing processing on the pixel density distribution map of the potential oil spill area image.
9. The small sample SAR oil spill segmentation method based on population density analysis according to claim 1, characterized in that: The preprocessing of the original SAR image comprises: Orbit correction, terrain correction, radiation calibration and filtering and denoising; The image enhancement of the original SAR image comprises: Adjust the brightness of the original SAR image through gamma correction; The contrast of the original SAR image is improved by using the histogram equalization algorithm.
10. The small sample SAR oil spill segmentation method based on population density analysis according to claim 1, characterized in that: The method of using the pixel density distribution map segmentation result of the potential oil spill area image to correct the boundary of the preliminary segmentation result to obtain the oil spill segmentation result map includes: For the same pixel point in the potential oil spill area image, if the preliminary segmentation result is different from the segmentation result of the pixel density distribution map of the potential oil spill area image, the segmentation result of the pixel density distribution map of the potential oil spill area image shall be used as the oil spill segmentation result.