Overlap shadow detection method and device based on superpixel segmentation

Abstract

The present invention discloses a method and device for detecting overlapping shadows based on superpixel segmentation, the method comprising: obtaining an original radar image and performing preprocessing to obtain an initial SAR image and its brightness image; performing superpixel segmentation on the initial SAR image using the brightness image to obtain a first superpixel image; performing morphological processing on the first superpixel image to obtain a second superpixel image; and performing overlapping shadow detection on the second superpixel image to obtain a detection result. The present invention combines pixels with adjacent and similar information into a primitive block by using a superpixel segmentation method, thereby greatly reducing the complexity of subsequent image processing tasks, overcoming the shortcomings of traditional pixel-by-pixel detection, such as poor noise robustness and poor integrity of detection results, and improving detection accuracy.

Description

Technical Field

[0001] The present invention belongs to the technical field of synthetic aperture radar interferometry, and in particular relates to a method and device for detecting overlapping shadows based on superpixel segmentation. Background Art

[0002] Interferometric Synthetic Aperture Radar (InSAR) is a new type of active surface deformation monitoring technology. It obtains two or more SAR images by multiple observations and imaging of the same area, and then inverts the digital elevation DEM based on the geometric relationship of interferometric imaging and the absolute phase difference between the same-name points on the complex image pair. InSAR technology has outstanding technical advantages such as high accuracy, high spatial resolution, almost no restrictions on cloud and rain weather, and aerial remote sensing. It is widely used in terrain mapping, marine monitoring, disaster monitoring and early warning, etc.

[0003] Since InSAR technology is very sensitive to tiny surface deformations, it is often difficult to obtain highly coherent interferometric SAR images in areas with complex terrain. Overlap and shadow phenomena are prone to occur, which seriously damages the interferometric nature of interferometric SAR images. Wrong phase information will lead to phase unwrapping errors in adjacent areas, which will affect the entire phase unwrapping process. Therefore, accurate detection of shadows and overlap phenomena in InSAR is very necessary for subsequent image processing tasks.

[0004] At present, the common detection methods for shadows and overlaps in InSAR mainly include threshold segmentation method based on amplitude and coherence coefficient, method based on edge features and fuzzy theory, method based on local frequency estimation, method based on eigenvalue decomposition and method based on CFAR.

[0005] However, the above methods all perform pixel-by-pixel discrimination during detection. Since each resolution unit on the SAR image is composed of many randomly distributed scattering point vectors, coherent speckle noise is formed on the SAR image, and the noise robustness is poor, the pixel-by-pixel discrimination results are relatively discrete, and it is difficult to reflect the texture information of the SAR data, thus affecting the detection accuracy. Summary of the invention

[0006] In order to solve the above problems existing in the prior art, the present invention provides a method and device for detecting overlapping shadows based on superpixel segmentation. The technical problem to be solved by the present invention is achieved through the following technical solutions:

[0007] A superpixel segmentation-based overlapping shadow detection method, comprising:

[0008] Acquire the original radar image and preprocess it to obtain the initial SAR image and its brightness image;

[0009] Performing super-pixel segmentation on the initial SAR image using the brightness image to obtain a first super-pixel image;

[0010] Performing morphological processing on the first super-pixel image to obtain a second super-pixel image;

[0011] Perform overlapping shadow detection on the second superpixel image to obtain a detection result.

[0012] In one embodiment of the present invention, an original radar image is acquired and preprocessed to obtain an initial SAR image and a brightness map thereof, including:

[0013] Performing speckle filtering on the original radar image to remove speckle noise of the image to obtain an initial SAR image;

[0014] The grayscale value of the initial SAR image is normalized and grayscale enhanced to obtain a corresponding brightness image.

[0015] In one embodiment of the present invention, super-pixel segmentation is performed on the initial SAR image using the brightness image to obtain a first super-pixel image, including:

[0016] Initializing the cluster centers of the initial SAR image and assigning a corresponding label to each cluster center;

[0017] Calculating gradient information of pixel points according to the brightness image to correct the cluster center;

[0018] Using the similarity between each pixel point and the cluster center, the cluster center is iteratively updated to obtain the final cluster center;

[0019] The initial SAR image is segmented into superpixels according to the final clustering center to obtain a first superpixel image.

[0020] In one embodiment of the present invention, calculating the gradient information of the pixel points according to the brightness image to correct the cluster center includes:

[0021] Set the shape of the superpixel grid and use the brightness information and position information to fuse each cluster center and each pixel point;

[0022] The gradient information is calculated according to the brightness information and position information of each pixel, and each cluster center is moved to the pixel position corresponding to the lowest gradient position in the eight-connected domain to correct the cluster center.

[0023] In one embodiment of the present invention, the cluster center is iteratively updated by using the similarity between each pixel point and the cluster center, and the final cluster center is obtained, including:

[0024] Define the maximum distance and maximum color difference in space;

[0025] Calculating a first similarity between each pixel and a plurality of cluster centers within a certain range around it according to the maximum spatial distance and the maximum color difference;

[0026] Calculating a second similarity according to the first phase similarity to update the first similarity;

[0027] Repeat the above steps of calculating the second similarity and updating the first similarity until the iteration stop condition is met, and obtain the final value of the first similarity between each pixel point and multiple cluster centers within a certain range around it;

[0028] The label of the cluster center with the highest final first similarity value to each pixel is assigned to the pixel to obtain the final cluster center.

[0029] In one embodiment of the present invention, calculating the second similarity according to the first phase similarity to update the first similarity includes:

[0030] Initialize the label image l = {s(i) = -1|i∈[1,N]} and the distance image D = {d(i,C k )=∞|i∈[1,N]}; where s(i) represents the label of the i-th pixel, N represents the total number of pixels in the initial SAR image, and d(i,C k ) represents the relationship between the i-th pixel and the k-th cluster center C k The first similarity between

[0031] The second similarity of each pixel is calculated according to the following formula:

[0032]

[0033] Among them, d s (i,C k ) represents the relationship between the i-th pixel and the k-th cluster center C k The second similarity between Represents the kth cluster center C k Features, represents d(i,C k ) is the weighted L2 norm of the feature vector f, f i Represents the feature of the i-th pixel, W = diag(1, λ, λ), λ = N c 2 / N s 2 , N c Indicates the maximum color difference, N sIndicates the maximum spatial distance;

[0034] If d s (i,C k ) <d(i,C k ), then let d(i,C k ) = d s (i,C k ), s(i)=k.

[0035] In one embodiment of the present invention, performing morphological opening processing on the first super-pixel image to obtain a second super-pixel image includes:

[0036] Traversing all superpixels in the first superpixel image, reallocating labels to regions that are not adjacent but have the same label, and obtaining a new label image;

[0037] Performing a morphological opening operation on the new label image to remove superpixels that are too small;

[0038] The labels of the superpixel image processed by the morphological opening operation are reassigned to obtain a second superpixel image.

[0039] In one embodiment of the present invention, the second superpixel image is subjected to overlapping shadow detection, and the detection result obtained includes:

[0040] The local frequency of each superpixel in the second superpixel image is estimated, and the superpixels with negative local frequencies are marked as overlapping shadow areas I. ls ;

[0041] Calculate the average intensity L of the initial SAR image avr and the average intensity of each superpixel in the second superpixel image and the average coherence coefficient coh(k), k∈[1,K′], where K′ represents the total number of superpixels in the second superpixel image;

[0042] Traverse the overlapping shadow area I ls The superpixels in L greater than a certain multiple avr The superpixels with a superpixel and an average coherence coefficient coh(k) less than the first threshold are determined as the overlapped area I. l ;

[0043] Let the partially shaded area I s =I ls -I l , calculate I s The average amplitude of traversing I sThe superpixels in the shadow area I are determined as superpixels with amplitudes less than a certain multiple of the average amplitude and superpixels with an average coherence coefficient coh(k) less than a second threshold. s .

[0044] In one embodiment of the present invention, performing local frequency estimation on each superpixel in the second superpixel image includes:

[0045] In the interferometric complex phase image, a window of a certain size is taken with the pixel as the center;

[0046] Performing FFT transformation on the window to obtain frequency spectrum information of the window;

[0047] The local frequency of the pixel is obtained according to the peak position of the frequency spectrum information.

[0048] Another embodiment of the present invention further provides a superpixel segmentation-based overlapping shadow detection device, comprising:

[0049] Initialization module, used to obtain the original radar image and preprocess it to obtain the initial SAR image and its brightness image;

[0050] A super-pixel segmentation module, used to perform super-pixel segmentation on the initial SAR image using the brightness image to obtain a first super-pixel image;

[0051] A morphological processing module, configured to perform morphological opening processing on the first super-pixel image to obtain a second super-pixel image;

[0052] The detection module is used to perform overlapping shadow detection on the second super-pixel image to obtain a detection result.

[0053] Beneficial effects of the present invention:

[0054] 1. The present invention combines pixels with adjacent and similar information into a primitive block by using a superpixel segmentation method, which greatly reduces the complexity of subsequent image processing tasks, overcomes the shortcomings of traditional pixel-by-pixel detection, such as poor noise robustness and poor integrity of detection results, and improves detection accuracy;

[0055] 2. The present invention improves the clustering characteristics of superpixel segmentation and further improves the detection accuracy by iteratively calculating the similarity of superpixel focus centers;

[0056] 3. After the initial image is divided into superpixels, the present invention also performs a morphological opening operation on it to form a new area with the approximate superpixel properties of the texture features, thereby avoiding the same area from being divided into multiple superpixels, and greatly improving the regional connectivity and reliability of superpixel segmentation.

[0057] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is a flowchart of a method for detecting overlapping shadows based on superpixel segmentation provided by an embodiment of the present invention;

[0059] Figure 2 It is a flowchart of another overlapping shadow detection method based on superpixel segmentation provided by an embodiment of the present invention;

[0060] Figure 3 The embodiment of the present invention provides the original radar image;

[0061] Figure 4 is a SAR image subjected to speckle filtering provided by an embodiment of the present invention;

[0062] Figure 5 is a schematic diagram of initializing cluster centers provided by an embodiment of the present invention;

[0063] Figure 6 Schematic diagram of superpixel segmentation with different K values ​​and m values ​​provided by an embodiment of the present invention;

[0064] Figure 7 is the overlapped area detection result provided by the embodiment of the present invention

[0065] Figure 8 is the result of shadow area detection provided by the embodiment of the present invention;

[0066] Fig. 9 is a priori DEM map provided by an embodiment of the present invention;

[0067] Fig.10 It is an overlapping area obtained by using a priori DEM provided in an embodiment of the present invention;

[0068] Fig.11 It is a shadow area obtained by using a priori DEM provided in an embodiment of the present invention;

[0069] Fig.12 It is a structural schematic diagram of an overlapping shadow detection device based on superpixel segmentation provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0070] The present invention is further described in detail below with reference to specific embodiments, but the embodiments of the present invention are not limited thereto.

[0071] Embodiment 1

[0072] See also Figure 1-2 , Figure 1is a flow chart of a method for detecting overlapping shadows based on superpixel segmentation provided by an embodiment of the present invention. Figure 2 It is a flowchart of another overlapping shadow detection method based on superpixel segmentation provided in an embodiment of the present invention.

[0073] Specifically, the method provided in this embodiment includes:

[0074] Step 1: Obtain the original radar image and preprocess it to obtain the initial SAR image and its brightness image.

[0075] Firstly, the original radar image is subjected to speckle filtering to remove the speckle noise of the image and obtain the initial SAR image.

[0076] Specifically, this embodiment uses the Lee filtering algorithm to filter the original radar image, and uses a local window to estimate the mean and standard deviation. The value of the pixel point at the center of the window is:

[0077]

[0078] Where I(x,y) is the center pixel value of the original image, is the mean of the local window, w(x,y) is the weight function, which can be expressed as:

[0079]

[0080] is the standard deviation of the local window, σ 2 is the standard deviation of the image.

[0081] See also Figure 3 and Figure 4 , Figure 3 The embodiment of the present invention provides an original radar image. Figure 4 This is a SAR image subjected to speckle filtering provided by an embodiment of the present invention.

[0082] Then, the grayscale value of the initial SAR image is normalized and grayscale enhanced to obtain the corresponding brightness image.

[0083] Specifically, after the grayscale value of the initial SAR image is normalized, the values ​​of all pixels are multiplied by 100 to enlarge them as the brightness information of the image, and a brightness map is obtained.

[0084] Step 2: Use the brightness image to perform super-pixel segmentation on the initial SAR image to obtain a first super-pixel image, which specifically includes:

[0085] 21) Initialize the cluster centers of the initial SAR image and assign corresponding labels to each cluster center.

[0086] See also Figure 5 , Figure 5 Schematic diagram of initializing cluster centers provided by an embodiment of the present invention. Specifically, the initial SAR image obtained in step 1 is evenly divided into K rectangular grids, and the initial adjacent cluster centers are spaced apart by

[0087] 22) Calculate the gradient information of the pixel points based on the brightness image to correct the cluster center.

[0088] First, the shape of the superpixel grid is set, and the brightness information and position information are fused to represent each cluster center and each pixel.

[0089] Specifically, this embodiment generates superpixels with hexagonal grid features for the original SAR image, such as Figure 5 As shown, calculate the number of cluster centers in the first column of the image and then update S, and represent each cluster center as C k for Each pixel i is represented by f i =[l i ,x i ,y i ] T , i∈[1,N], l i is the brightness of pixel i, l∈[0,100], x i ,y i is the position of pixel i in the image.

[0090] Then, the gradient information is calculated according to the brightness information and position information of each pixel, and each cluster center is moved to the pixel position corresponding to the lowest gradient position in the eight-connected domain to correct the cluster center.

[0091] In this embodiment, the gradient metric is expressed as:

[0092] G(x,y)=||L(x+1,y)-L(x-1,y)|| 2 +||L(x,y+1)-L(x,y-1)|| 2

[0093] Among them, L(x,y) represents the brightness information of the pixel with coordinates (x,y) in the image, and ||·|| represents the L2 norm.

[0094] This embodiment uses the above formula to calculate the gradient information, selects the y coordinate corresponding to the minimum amount, and moves the cluster center to the coordinate position.

[0095] 23) Using the similarity between each pixel and the cluster center, the cluster center is iteratively updated to obtain the final cluster center.

[0096] 23a) Define the maximum distance and the maximum color difference in space.

[0097] In this embodiment, the maximum spatial distance is defined as N. s =S, the maximum color difference is N c , N c ∈[1,m], where m is a value between 1 and 100 and can be set based on experience or actual conditions.

[0098] 23b) Calculating the first similarity between each pixel and a number of cluster centers within a certain range around it according to the maximum spatial distance and the maximum color difference.

[0099] Specifically, the spatial distance and color distance are differentially integrated, and the similarity between the cluster centers within the 2S×2S range around each pixel is calculated, also known as the first similarity, and the calculation formula is as follows:

[0100]

[0101] 23c) Calculating a second similarity based on the first similarity to update the first similarity.

[0102] First, the second similarity is calculated according to the first similarity.

[0103] Specifically, the first phase similarity d(i,C k ) is expressed as a vector, and the weighted L2 norm of its eigenvector f is as follows:

[0104]

[0105] Among them, the diagonal matrix W = diag(1,λ,λ), λ = N c 2 / N s 2 , f=[l,x,y] T .

[0106] Rewrite d(i,C k )for:

[0107]

[0108] Calculate the distance between a pixel i and other cluster centers C k The similarity of is a constant value and can be omitted, then the final d(i,C k ) is expressed as:

[0109]

[0110] In this embodiment, the similarity ds (i,C k ) is called the second similarity.

[0111] Then, the first similarity is updated according to the magnitude relationship between the second similarity and the first similarity, as follows:

[0112] Initialize the label image l = {s(i) = -1|i∈[1,N]} and the distance image D = {d(i,C k )=∞|i∈[1,N]}; where s(i) represents the label of the i-th pixel, N represents the total number of pixels in the initial SAR image, and d(i,C k ) represents the relationship between the i-th pixel and the k-th cluster center C k The first similarity between

[0113] According to the above formula, the second similarity between each pixel and the cluster center within the range of 2S×2S in its neighborhood is calculated and compared with the first similarity. s (i,C k ) <d(i,C k ), then let d(i,C k ) = d s (i,C k ), s(i)=k.

[0114] 23d) Repeat the above steps of calculating the second similarity and updating the first similarity until the iteration stop condition is met, and obtain the final value of the first similarity between each pixel and a plurality of cluster centers within a certain range around it.

[0115] Specifically, set the error term E = D n -D n-1 , D n It represents the first similarity value obtained by the nth iteration of a certain pixel point. The second similarity of each pixel point is repeatedly calculated and the first similarity is updated until E converges, and the final value of the first similarity between each pixel point and multiple cluster centers within a certain range around it is obtained.

[0116] 23e) assigning the label of the cluster center with the highest final first similarity value to each pixel to the pixel, so as to obtain the final cluster center.

[0117] In this embodiment, after obtaining the final value of the first similarity between each pixel point and several cluster centers around it, these final values ​​are compared and the largest value is selected. The cluster center corresponding to this value is the cluster center with the highest similarity to the pixel point. The label of the cluster center is assigned to the pixel, that is, the pixel is divided into the cluster center, thereby realizing the update of the cluster center and obtaining the final cluster center.

[0118] 24) Perform superpixel segmentation on the initial SAR image according to the final cluster center to obtain a first superpixel image.

[0119] After the above processing, the feature information of the new cluster center and its corresponding image label are obtained. It can be understood that after the above cluster center update, each super pixel area formed by it changes from the original hexagonal grid to an irregular grid shape.

[0120] This embodiment improves the clustering characteristics of superpixel segmentation by iteratively calculating the similarity of superpixel focus centers, which helps to further improve the detection effect.

[0121] Step 3: Perform morphological processing on the first super-pixel image to obtain a second super-pixel image.

[0122] After the superpixel segmentation in step 2, a first superpixel image with K cluster centers is obtained. Since each pixel in the image is relabeled according to its similarity with the surrounding cluster centers, it is not difficult to understand that there must be unconnected points or areas in the image. In order to enhance connectivity and facilitate subsequent further image processing, this embodiment performs morphological processing on the first superpixel obtained in step 2, specifically including the following steps:

[0123] 31) traversing all superpixels in the first superpixel image, reallocating labels to regions that are not adjacent but have the same label, and obtaining a new label image;

[0124] 32) Perform morphological opening operation on the new label image to remove superpixels that are too small;

[0125] 33) Reassign labels to the superpixel image processed by the morphological opening operation to obtain a second superpixel image.

[0126] It should be noted that the morphological opening operation is an existing processing method, and its specific implementation process is not described in detail in this embodiment.

[0127] Through the superpixels after morphological processing, some areas are associated, divided into multiple areas or absorbed by other areas, and these areas are renumbered, and finally a label image is output, which indicates the superpixel number corresponding to each pixel on the SAR image. The final number of superpixels is K′. For details, please refer to Figure 6 , Figure 6 Schematic diagram of superpixel segmentation with different K values ​​and m values ​​provided by an embodiment of the present invention.

[0128] It is not difficult to understand that the number of superpixels K′ here can be the same as or different from the original total number of superpixels K.

[0129] This embodiment uses morphological opening operations to process superpixels, forming new regions with approximate superpixel properties of texture features, avoiding the same region from being segmented into multiple superpixels, and greatly improving the regional connectivity and reliability of superpixel segmentation.

[0130] Step 4: Perform overlapping shadow detection on the second superpixel image to obtain a detection result.

[0131] Specifically, this embodiment uses multi-information fusion such as average intensity information, average coherence coefficient, and average amplitude information to detect overlapping and shadow areas, which specifically includes the following steps:

[0132] Before detection, the average intensity information C of the original pixels in all SAR images contained in each superpixel label is first calculated. k , average coherence coefficient And the superpixel feature structure is expressed as:

[0133]

[0134] in, Represents superpixel C k The average intensity information of the original pixels in all SAR images included; Represents superpixel C k The equivalent position information of Represents superpixel C k The average coherence coefficient.

[0135] 41) Estimating the local frequency of each superpixel in the second superpixel image, and marking the superpixels with negative local frequencies as overlapping shadow areas I ls .

[0136] In this embodiment, the local frequency is obtained by spectrum shift of the interferometric SAR image, and the spectrum shift is:

[0137]

[0138] Where c is the speed of light; λ is the wavelength; R is the slant distance from the radar to the target point; B is the baseline length; γ is the baseline inclination; in the overlapping area, the local incident angle θ=β-α<0, and in the shadow area, the local incident angle θ=β-α>π / 2.

[0139] It can be seen from the formula that the values ​​of Δω in the overlap and shadow areas are all negative. The local frequency is estimated by interferometric phase gradient, and the phase in the window is simulated by a linear two-dimensional model. The interferometric phase gradient can be given by the following formula:

[0140]

[0141] in, is the phase within the simulated window, is the window average phase, f x and f y are the local frequencies in the x-direction and the y-direction respectively.

[0142] Specifically, step 41) may include:

[0143] 41a) In the interference complex phase image, a window of a certain size is selected with the pixel point as the center. For example, in this embodiment, a window of 5*5 size can be selected.

[0144] 41b) Performing FFT transformation on the window to obtain the spectrum information of the window.

[0145] Specifically, a 32×32-point FFT transformation may be performed on the above window to obtain the frequency spectrum information of the window.

[0146] 41c) Obtain the local frequency of the pixel point according to the peak position of the spectrum information.

[0147] In this embodiment, the peak position corresponds to the local frequency (f x f y ), the spectrum shift only exists in the range direction, and only the local frequency f in the range direction needs to be estimated y That's it.

[0148] It should be noted that the phase gradient range here is [0, 2π], and π should be subtracted to make it range [-π, π].

[0149] 42) Calculate the average intensity L of the initial SAR image avr and the average intensity of each superpixel in the second superpixel image And the average coherence coefficient coh(k), k∈[1,K′], where K′ represents the total number of superpixels in the second superpixel image.

[0150] It is understandable that the relevant parameters here can be calculated directly after obtaining the second super-pixel image, or can be calculated when needed.

[0151] 43) Traverse the overlapping shadow area I ls The superpixels in L greater than a certain multiple avr The superpixels with a superpixel and an average coherence coefficient coh(k) less than the first threshold are determined as the overlapped area I. l .

[0152] For example, in this embodiment, the first threshold value may be set to 0.7. Greater than 2 times the average strength L avrThe superpixels with an average coherence coefficient coh(k) less than 0.7 are marked as overlapped regions I. l . See Figure 7 , Figure 7 This is the overlapped area detection result provided by the embodiment of the present invention.

[0153] 44) Let the partial shaded area I s =I ls -I l , calculate I s The average amplitude avr(L(I s )), traverse I s The superpixels in the shadow area I are determined as superpixels with amplitudes less than a certain multiple of the average amplitude and superpixels with an average coherence coefficient coh(k) less than a second threshold. s .

[0154] For example, in this embodiment, the second threshold value may be set to 0.6. Specifically, in this embodiment, the amplitude is less than 22×avr(L(I s )) and superpixels with a coherence coefficient coh(k) less than 0.6 are marked as shaded areas and recalibrated as I s . See Figure 8 , Figure 8 This is the result of shadow area detection provided by the embodiment of the present invention.

[0155] At this point, the detection of the overlapping area and the shadow area in the second superpixel image is completed.

[0156] The present invention combines pixels with adjacent and similar information into a primitive block by using a superpixel segmentation method, thereby greatly reducing the complexity of subsequent image processing tasks, overcoming the shortcomings of traditional pixel-by-pixel detection such as poor noise robustness and poor integrity of detection results, and greatly improving detection accuracy.

[0157] Embodiment 2

[0158] In order to verify the effectiveness of the overlapping shadow detection method based on superpixel segmentation provided in the above embodiment 1, this embodiment also carried out a simulation test through a priori DEM (Digital Elevation Model), and compared the test results with the test results in the above embodiment 1 to illustrate the effectiveness of the present invention.

[0159] First, extract the prior DEM of the corresponding scene, see Fig. 9 , Fig. 9 The prior DEM map provided by the embodiment of the present invention is reversely positioned on the SAR image according to the radar orbit information. The specific reverse positioning steps include:

[0160] (a) Take the latitude, longitude and elevation of a point in the prior DEM and convert them to WGS84 coordinates;

[0161] (b) Calculate the difference (absolute value) between the Doppler frequency and the Doppler center frequency according to the radar Doppler equation:

[0162]

[0163] λ is the wavelength, v s is the satellite velocity vector, v p is the target velocity vector, P s is the satellite position vector, P P is the velocity vector of the target.

[0164] (c) Find the difference (absolute value) between the minimum Doppler frequency and the Doppler center frequency and the corresponding distance to determine the coordinates of the target in the SAR image;

[0165] (d) Calculate the slant distance from the target to the Doppler center. Based on the range sampling interval and the shortest slant distance, the range coordinates of the target on the SAR image can be determined.

[0166] Then, the lower viewing angle corresponding to each pixel in the SAR image is calculated according to the prior DEM. The calculation method of the lower viewing angle is:

[0167]

[0168] Among them, (X S ,Y S ,Z S ) is the satellite coordinates at a certain moment, (X T ,Y T ,Z T ) are the target ground coordinates.

[0169] Next, take 5 pixels around the prior DEM elevation to estimate the slope α. When β-α<0, it is marked as an overlap area, and when β-α>900, it is marked as a shadow area. Figure 10-11 , Fig.10 is an overlapping area obtained by using a priori DEM provided in an embodiment of the present invention, Fig.11 This is the shadow area obtained by using the prior DEM provided in the embodiment of the present invention.

[0170] Finally, the overlapping shadow area marked by the prior DEM is used as a standard, and the detection result of step 4 in the above embodiment 1 is compared with it, and statistical analysis is performed. The statistical analysis is shown in Table 1 below.

[0171] Table 1 Analysis of overlapping and shadow area detection results

[0172] Detection result image ratio Test result accuracy Overlap Area 7.8％ 98.3％ Shadow Area 0.0016％ 99.8％

[0173] It can be seen from Table 1 that the detection accuracy of the detection method provided by the present invention for overlapping shadow areas is as high as over 98%, indicating that the method of the present invention can well detect overlapping shadow areas in radar SAR images.

[0174] Embodiment 3

[0175] Based on the above embodiment 1, this embodiment further provides a superpixel segmentation-based overlapping shadow detection device. Fig.12 , Fig.12 : is a schematic diagram of the structure of an overlapping shadow detection device based on superpixel segmentation provided by an embodiment of the present invention, which includes:

[0176] Initialization module 1 is used to obtain the original radar image and preprocess it to obtain the initial SAR image and its brightness image;

[0177] The super-pixel segmentation module 2 is used to perform super-pixel segmentation on the initial SAR image using the brightness image to obtain a first super-pixel image;

[0178] A morphological processing module 3 is used to perform morphological opening processing on the first super-pixel image to obtain a second super-pixel image;

[0179] The detection module 4 is used to perform overlapping shadow detection on the second super-pixel image to obtain a detection result.

[0180] The device provided in this embodiment can implement the method provided in the above-mentioned embodiment 1, and the detailed process will not be repeated here.

[0181] Therefore, the superpixel segmentation-based overlapping shadow detection device provided in this embodiment can greatly reduce the complexity of subsequent image processing tasks, overcome the shortcomings of traditional pixel-by-pixel detection such as poor noise robustness and poor overall detection results, and improve detection accuracy.

[0182] The above contents are further detailed descriptions of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as falling within the protection scope of the present invention.

Claims

1. A superpixel segmentation-based overlapping shadow detection method, characterized in that: include: Step 1: Obtain the original radar image and preprocess it to obtain the initial SAR image and its brightness image; Step 2: Using the brightness image to perform super-pixel segmentation on the initial SAR image to obtain a first super-pixel image; 21) Initializing the cluster centers of the initial SAR image and assigning corresponding labels to each cluster center; 22) calculating the gradient information of the pixel points according to the brightness image to correct the cluster center; 23) using the similarity between each pixel point and the cluster center, iteratively updating the cluster center to obtain the final cluster center; 23a) Define the maximum spatial distance N s and the maximum color difference N c ; 23b) calculating a first similarity between each pixel and a number of cluster centers within a certain range around it according to the maximum spatial distance and the maximum color difference; the calculation formula of the first similarity is as follows: In the formula, d(i,C k ) represents the pixel i and the cluster center C k The first similarity between i represents the brightness of pixel i, x i ,y i Indicates the position of pixel i in the image; Represents the cluster center C k Brightness, Represents the cluster center C k location; 23c) calculating a second similarity based on the first similarity to update the first similarity; 23d) Repeat the above steps of calculating the second similarity and updating the first similarity until the iteration stop condition is met, and obtain the final value of the first similarity between each pixel and a plurality of cluster centers within a certain range around it; 23e) assigning the label of the cluster center with the highest final value of the first similarity to each pixel to the pixel, so as to obtain the final cluster center; 24) performing super-pixel segmentation on the initial SAR image according to the final cluster center to obtain a first super-pixel image; Step 3: performing morphological processing on the first super-pixel image to obtain a second super-pixel image; 31) traversing all superpixels in the first superpixel image, reallocating labels to regions that are not adjacent but have the same label, and obtaining a new label image; 32) performing a morphological opening operation on the new label image to remove superpixels that are too small; 33) reallocating labels to the superpixel image processed by the morphological opening operation to obtain a second superpixel image; Step 4: Perform overlapping shadow detection on the second superpixel image to obtain a detection result.

2. The method for detecting overlapping shadows based on superpixel segmentation according to claim 1, characterized in that: The original radar image is obtained and preprocessed to obtain the initial SAR image and its brightness map, including: Performing speckle filtering on the original radar image to remove speckle noise of the image to obtain an initial SAR image; The grayscale value of the initial SAR image is normalized and grayscale enhanced to obtain a corresponding brightness image.

3. The method for detecting overlapping shadows based on superpixel segmentation according to claim 1, characterized in that: Calculating the gradient information of the pixel points according to the brightness image to correct the cluster center includes: Set the shape of the superpixel grid and use the brightness information and position information to fuse each cluster center and each pixel point; The gradient information is calculated according to the brightness information and position information of each pixel, and each cluster center is moved to the pixel position corresponding to the lowest gradient position in the eight-connected domain to correct the cluster center.

4. The method for detecting overlapping shadows based on superpixel segmentation according to claim 1, characterized in that: Calculating a second similarity according to the first similarity to update the first similarity includes: Initialize the label image l = {s(i) = -1|i∈[1,N]} and the distance image D = {d(i,C k )=∞|i∈[1,N]}; where s(i) represents the label of the i-th pixel, N represents the total number of pixels in the initial SAR image, and d(i,C k ) represents the relationship between the i-th pixel and the k-th cluster center C k The first similarity between The second similarity of each pixel is calculated according to the following formula: Among them, d s (i,C k ) represents the second similarity between the i-th pixel and the k-th cluster center Ck, Represents the kth cluster center C k Features, represents d(i,C k ) is the weighted L2 norm of the feature vector f, f i Represents the feature of the i-th pixel, W = diag(1, λ, λ), λ = N c 2 / N s 2 , N c Indicates the maximum color difference, N s Indicates the maximum spatial distance; If d s (i,C k ) <d(i,C k ), then let d(i,C k ) = d s (i,C k ), s(i)=k.

5. The method for detecting overlapping shadows based on superpixel segmentation according to claim 1, characterized in that: Performing overlapping shadow detection on the second superpixel image, and obtaining a detection result includes: The local frequency of each superpixel in the second superpixel image is estimated, and the superpixels with negative local frequencies are marked as overlapping shadow areas I. ls ; Calculate the average intensity L of the initial SAR image avr and the average intensity of each superpixel in the second superpixel image and the average coherence coefficient coh(k), k∈[1,K′], where K′ represents the total number of superpixels in the second superpixel image; Traverse the overlapping shadow area I ls The superpixels in L greater than a certain multiple avr The superpixels with a superpixel and an average coherence coefficient coh(k) less than the first threshold are determined as the overlapped area I. l ; Let the partially shaded area I s =I ls -I l , calculate I s The average amplitude of traversing I s The superpixels in the shadow area I are determined as superpixels with amplitudes less than a certain multiple of the average amplitude and superpixels with an average coherence coefficient coh(k) less than a second threshold. s .

6. The method for detecting overlapping shadows based on superpixel segmentation according to claim 5, characterized in that: Performing local frequency estimation on each superpixel in the second superpixel image includes: In the interferometric complex phase image, a window of a certain size is taken with the pixel as the center; Performing FFT transformation on the window to obtain frequency spectrum information of the window; The local frequency of the pixel is obtained according to the peak position of the frequency spectrum information.

7. A superpixel segmentation-based overlapping shadow detection device, characterized in that: include: Initialization module (1), used for acquiring the original radar image and preprocessing it to obtain the initial SAR image and its brightness image; A superpixel segmentation module (2) is used to perform superpixel segmentation on the initial SAR image using the brightness image to obtain a first superpixel image; specifically comprising: 21) Initializing the cluster centers of the initial SAR image and assigning corresponding labels to each cluster center; 22) calculating the gradient information of the pixel points according to the brightness image to correct the cluster center; 23) using the similarity between each pixel point and the cluster center, iteratively updating the cluster center to obtain the final cluster center; 23a) Define the maximum spatial distance N s and the maximum color difference N c ; 23b) calculating a first similarity between each pixel and a number of cluster centers within a certain range around it according to the maximum spatial distance and the maximum color difference; the calculation formula of the first similarity is as follows: In the formula, d(i,C k ) represents the pixel i and the cluster center C k The first similarity between i represents the brightness of pixel i, x i ,y i Indicates the position of pixel i in the image; Represents the cluster center C k Brightness, Represents the cluster center C k location; 23c) calculating a second similarity based on the first similarity to update the first similarity; 23d) Repeat the above steps of calculating the second similarity and updating the first similarity until the iteration stop condition is met, and obtain the final value of the first similarity between each pixel and a plurality of cluster centers within a certain range around it; 23e) assigning the label of the cluster center with the highest final value of the first similarity to each pixel to the pixel, so as to obtain the final cluster center; 24) performing super-pixel segmentation on the initial SAR image according to the final cluster center to obtain a first super-pixel image; A morphological processing module (3) is used to perform morphological opening processing on the first super-pixel image to obtain a second super-pixel image; specifically comprising: 31) traversing all superpixels in the first superpixel image, reallocating labels to regions that are not adjacent but have the same label, and obtaining a new label image; 32) performing a morphological opening operation on the new label image to remove superpixels that are too small; 33) reallocating labels to the superpixel image processed by the morphological opening operation to obtain a second superpixel image; A detection module (4) is used to perform overlapping shadow detection on the second super-pixel image to obtain a detection result.

Images