A multi-information fusion water-level fluctuation zone extraction method

Abstract

The application particularly relates to a multi-information fusion extraction method of a drawdown zone, which adopts a Harris corner point detection algorithm based on image gray, distinguishes a water body area from a non-water body area by obtaining a corner point density, performs color clustering on an image of the non-water body area, divides the image of the non-water body area into a vegetation area and a soil area, calculates a drawdown zone area according to a distance map between the vegetation area and the soil area and a distance to an edge area between the water body area and the non-water body area. The multi-information fusion extraction method of the drawdown zone can accurately identify the drawdown zone area in an original image by fusing various image processing methods such as corner point detection, color clustering and interference elimination, thereby providing a basis for repairing the drawdown zone in a later period.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of image processing of drawdown zone, and particularly relates to a drawdown zone extraction method based on multi-information fusion. BACKGROUND

[0002] The drawdown zone is also called water level fluctuation zone or drawdown zone, and mainly refers to a special soil area periodically exposed with water level fluctuation. The drawdown zone is directly connected with water body and belongs to the wetland category. From the perspective of ecology, the drawdown zone is the kidney of nature, and its purification function to pollutants is very significant. In addition, the drawdown zone also plays a vital role in maintaining the balance of the reservoir bank ecological system and protecting the water body of reservoirs, lakes and rivers. Therefore, with the fluctuation of the water body, the exposure of the land will have a great impact on the ecology and will also cause soil loss. Therefore, the accurate identification of the drawdown zone is of great significance to the ecological environment restoration.

[0003] However, the accurate identification of the drawdown zone in the prior art is not enough, and therefore a more accurate drawdown zone identification and extraction method is needed. SUMMARY

[0004] In order to solve the problem of identification and extraction of the drawdown zone, the application provides a drawdown zone extraction method based on multi-information fusion. In the method, various image processing methods such as corner point detection, color clustering and interference elimination are fused, so that the drawdown zone area in the original image can be accurately identified, and a basis is provided for the later repair of the drawdown zone.

[0005] To achieve the above-mentioned application purposes, the technical scheme adopted by the application is as follows: a drawdown zone extraction method based on multi-information fusion, comprising the following steps:

[0006] S1: performing gray scale processing on the image, and identifying the water body area and the non-water body area through the corner point information in the image;

[0007] S2: using an edge detection algorithm to extract the edge area between the water body area and the non-water body area in the image, and performing color clustering on the image of the non-water body area to divide the image of the non-water body area into a vegetation area and a soil area;

[0008] S3: calculating the distance between each point in the soil area and each point in the edge area, taking the minimum value set of the distance to form a distance map, and calculating the drawdown zone area between the distance map and the edge area.

[0009] Preferably, the specific method for identifying the water body area and the non-water body area through the corner point information in the image in step S1 is as follows:

[0010] S1.1: using a corner point detection algorithm to calculate the corner points of the image to obtain a corner point image, and distinguishing the corner points from the non-corner points in the image;

[0011] S1.2: Obtain the corner point density map by the corner point density algorithm, and classify the corner point density of each point in the image into 0-255;

[0012] S1.3: The classified value of the corner point density of each point in the image is distinguished by the threshold value of the corner point density map to obtain a segmented image, and the segmented image is processed;

[0013] S1.4: The processed segmented image is combined with the original image to obtain a non-water area image.

[0014] Preferably, the specific algorithm of the corner point density algorithm is as follows:

[0015]

[0016]

[0017] Wherein, K is the number of detected corner points in the corner point image, p(x i ,y j ) is any point in the image, ρ p is the corner point density, (x k ,y k ) is the position of the kth corner point, d pk is the distance from the point p in the image to the corner point k, and p is any point in the image.

[0018] Preferably, the specific algorithm for classifying the corner point density of each point in the image into 0-255 is as follows:

[0019]

[0020] Wherein, ρ max and ρ min are the maximum and minimum values of the corner point density in the corner point density map, respectively.

[0021] Preferably, the threshold value t0 of the corner point density map is calculated by using the Otsu threshold segmentation algorithm, and for any point p in the image, the segmented image is obtained by using the following formula:

[0022]

[0023] Remove the small connected domain in the obtained segmented image, perform a closing operation on the image, fill the holes, and smooth the edges to obtain the processed segmented image.

[0024] Preferably, before calculating the corner points of the image by using the corner point detection algorithm, the histogram of the image after gray processing is equalized; when calculating the corner points of the image by using the corner point detection algorithm, non-maximum suppression is adopted.

[0025] Preferably, the specific method of combining the processed segmented image with the original image is that: the pixel points on the original image and the segmented image are one-to-one corresponding, then for any position P i (x,y), the corresponding value in the original image is r i ,g i ,b i , and the corresponding value in the segmented image is th i , the original image is processed as follows:

[0026]

[0027] to obtain a non-water body image.

[0028] Preferably, the specific method of color clustering in step S2 is that: the color is divided into two categories, which are green area and yellow area respectively, and all points in the green area are set as black to obtain a soil area image.

[0029] All values of the distance map are normalized, and the threshold t1 is selected by using the Otsu algorithm, and compared with all values d i in the distance map, if d i >t1, then the position corresponding to d i in the soil area image is set as black.

[0030] Preferably, when there is cloud and fog interference in the original image, the image containing cloud and fog is decomposed by using wavelet, the low-frequency component is filtered by using homomorphism, and the high-frequency component is reconstructed after being stretched by using non-linear.

[0031] Or, the image without cloud and fog at the same spatial position is taken to compensate, specifically: the image without cloud and fog is registered and matched with the image with cloud and fog by using RGB three-color histogram, the area R covered by cloud and fog in the image is intercepted, and the registered image R' without cloud and fog is correspondingly intercepted, R' is replaced to the position corresponding to R, and then image fusion is performed.

[0032] Preferably, for multiple cloud and fog interference images, each cloud and fog area can be replaced respectively.

[0033] The present application has the following beneficial effects:

[0034] 1) In the present application, various image processing methods such as corner point detection, color clustering and interference elimination are adopted to distinguish the water body area and the non-water body area, and further distinguish the vegetation area and the soil area in the non-water body area, so that the drawdown zone area can be accurately identified, which is of great significance for ecological environment restoration.

[0035] 2) In this invention, when there is cloud or fog interference, the obscured ground shape is re-fused and matched by replacing the whole or parts, so as to obtain more accurate ground shape information, which is convenient for image recognition of the drawdown zone in the later stage. Detailed Implementation

[0036] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Unless otherwise specified, the technical means used in the embodiments are conventional means well known to those skilled in the art.

[0037] In the description of this invention, it should be understood that the terms "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicating orientation or positional relationships are only for the convenience of describing this invention, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this invention.

[0038] A method for extracting drawdown bands through multi-information fusion includes the following steps:

[0039] S1: The image is processed to grayscale, and water and non-water areas are identified based on corner information. The original image is acquired, and the Harris corner detection algorithm based on image grayscale is used to identify corner information. Specifically, the entire original image is first processed to grayscale, and then the corner detection algorithm is used to calculate its corners, resulting in a corner image. Because water bodies have weak backscattering capabilities, while land or rough terrain has high backscattering intensity, there is a significant difference in brightness between the two in the image. After obtaining the corner image, corner points are set to 255, and non-corner points to 0 to differentiate and identify water and non-water areas.

[0040] The specific method for identifying water bodies and non-water bodies using corner information in an image is as follows:

[0041] S1.1: Calculate the corner points of the image using a corner detection algorithm to obtain a corner point image, and distinguish between corner points and non-corner points in the image;

[0042] S1.2: Obtain the corner density map through the corner density algorithm, and classify the corner density of each point in the image into the range of 0-255;

[0043] The specific algorithm for corner density is as follows: assuming the image size is N×M and the number of detected corners is K, for any point p(x) in the image... i ,y j Its corner density ρp For:

[0044]

[0045]

[0046] Wherein, K is the number of corner points detected in the corner point image, p(x i ,y j ) is any point in the image, ρ p is the corner point density, (x k ,y k ) is the position of the kth corner point, d pk is the distance from the image point p to the corner point k, and p is any point in the image.

[0047] The specific algorithm for dividing the corner point density of each point in the image into 0-255 is as follows:

[0048]

[0049] Wherein, ρ max , ρ min are the maximum and minimum values of the corner point density in the corner point density map, respectively, and ρ p is the value after the division.

[0050] S1.3: The value of the corner point density of each point in the image after division The threshold of the corner point density map is used to distinguish, obtain the segmented image, and process the segmented image;

[0051] The threshold t0 of the corner point density map is calculated by using the Otsu threshold segmentation algorithm, and the segmented image is obtained by using the following formula for any point p in the image:

[0052]

[0053] The water area and the non-water area are clearly distinguished in the segmented image. The method for processing the segmented image is: removing the small connected domains in the obtained segmented image, performing a closing operation on the image, hole filling and edge smoothing to obtain the processed segmented image. In the present application, the connected domain area / image area<0.05 is considered as a small connected domain in the segmented image.

[0054] S1.4: Combine the processed segmented image with the original image to obtain the image of the non-water area.

[0055] ​The original image is an RGB image, and each pixel point corresponds to three values r∈[0, 255], g∈[0, 255], and b∈[0, 255]. The segmented image is a binary image, and each pixel point corresponds to only one value th, th=0 or 255. The original image and the segmented image are of the same size and correspond to each other pixel by pixel. Then, for any position P i (x,y), the corresponding value in the original image is r i ,g i ,b i , and the corresponding value in the segmented image is th i . The original image is processed as follows:

[0056]

[0057] to obtain a non-water body image.

[0058] To reduce the influence of the contrast difference of images obtained by different devices on detection, the histogram of the image after grayscale processing is equalized before corner detection. To ensure the accuracy of corner detection, when the corner detection algorithm is used to calculate the corners of the image, non-maximum suppression is used for processing.

[0059] S2: The edge region between the water body area and the non-water body area in the image is extracted by using an edge detection algorithm, and the image of the non-water body area is color clustered to divide the image of the non-water body area into a vegetation area and a soil area. The outer edge region of the segmented image is extracted by using an edge detection algorithm to obtain an edge point set B.

[0060] The non-water body image is color clustered into two categories, a green area and a yellow area. The green area belongs to the vegetation area, and the regions close to the green area are in this category. The yellow area belongs to the soil area, and the regions close to the yellow area are in this category. All points in the green area are set to black (RBG value is 0, 0, and 0), and a soil area image is obtained.

[0061] S3: The distance between each point in the soil area and each point in the edge region is calculated, the smallest value set is taken to form a distance map, and the distance map and the edge region are calculated to obtain the water-land transition zone.

[0062] For all points in the soil area, the pixel distance of each point to each point in the edge point set B is calculated, and the smallest value is recorded in the distance set D (distance map). D is the union set of the distances from the points in the soil area to the nearest edge, and the dimension of D is the same as the number of pixel points in the soil area. D corresponds to the soil area one by one.

[0063] To prevent the influence of the image size on the subsequent threshold selection, all values in the distance map are normalized.

[0064] Since the drawdown zone is at the edge of the soil area, all values of the distance map are normalized, and the threshold t1 is selected by using the Otsu algorithm i If d i >t1, the position corresponding to d i in the soil area image is set to black (d i The corresponding RGB value of the position where the soil area is located is set to 0, 0, 0, and the obtained area is the drawdown zone area (non-black area). d i is the value of the ith point in D.

[0065] Further, if the original image is disturbed by cloud and fog, the thin cloud and fog disturbance is considered as the ground form not being completely blocked, and the thick cloud and fog disturbance is considered as the part of the ground form being completely blocked.

[0066] If it is thin cloud and fog disturbance, the image containing cloud and fog is decomposed by using wavelet, the low-frequency component is filtered by using homomorphism, the high-frequency component is reconstructed after being stretched by using nonlinearity, and usually 3-5 layers of decomposition is the best.

[0067] If it is thick cloud and fog disturbance, the image without cloud and fog at the same spatial position is taken for compensation. Since the shooting time is different, the image may have certain scale and color difference, so it is necessary to first register the image without cloud and fog with the image with cloud and fog and match the RGB three-color histogram. In order to prevent more information from being confused, when the compensation is performed, the area R covered by the cloud and fog in the image is intercepted, and the registered image without cloud and fog R' is correspondingly intercepted. After R' is replaced to the position corresponding to R, the image fusion is performed.

[0068] For the images disturbed by multiple clouds and fogs (without distinguishing thin cloud and fog disturbance or thick cloud and fog disturbance), in order to obtain more accurate thick cloud and fog removal result, the replacement can be performed on each cloud and fog area respectively.

[0069] The above-described embodiments are only used to describe the preferred modes of the present application, and do not limit the scope of the present application. Without departing from the design spirit of the present application, various modifications, variations, modifications and replacements of the technical solutions of the present application made by those skilled in the art shall fall within the protection scope determined by the claims of the present application.

Claims

1. A method for extracting drawdown bands through multi-information fusion, characterized in that, The method comprises the following steps: S1: performing gray processing on the image, and identifying the water body area and the non-water body area through the corner point information in the image; S2: extracting the edge region between the water body area and the non-water body area in the image by using an edge detection algorithm, and performing color clustering on the image of the non-water body area, so that the image of the non-water body area is divided into a vegetation area and a soil area; S3: calculating the distance between each point in the soil area and each point in the edge region, taking the minimum value set of the distances to form a distance graph, and calculating the water-land ecotone area between the distance graph and the edge region.

2. The multi-information fused water-level-fluctuation belt extraction method according to claim 1, characterized in that: The specific method for identifying the water body area and the non-water body area through the corner point information in the image in step S1 is as follows: S1.1: calculating the corner points of the image by using a corner point detection algorithm to obtain a corner point image, and distinguishing the corner points and the non-corner points in the image; S1.2: obtaining a corner point density graph by using a corner point density algorithm, and classifying the corner point density of each point in the image into 0-255; S1.3: distinguishing the value of each point in the image after classification by using the threshold value of the corner point density graph to obtain a segmented image, and processing the segmented image; S1.4: combining the processed segmented image with the original image to obtain the image of the non-water body area.

3. The multi-information fused water-level-fluctuation belt extraction method according to claim 2, characterized in that: The specific algorithm of the corner point density algorithm is as follows: where K is the number of detected corners in the corner image, p(x i ,y j ) is an arbitrary point in the image, p p is the corner density, (x k ,y k ) is the position of the kth corner, d pk is the distance from the image point p to the corner k, and p is an arbitrary point in the image.

4. The multi-information fused water-level-fluctuation belt extraction method according to claim 3, characterized by: The specific algorithm for classifying the corner point density of each point in the image into 0-255 is as follows: where, p max , p min are the maximum and minimum values of the corner point density in the corner point density map, respectively.

5. The multi-information fused water-level-fluctuation belt extraction method according to claim 4, characterized by: The threshold value t0 of the corner point density graph is calculated by using an Otsu threshold segmentation algorithm, and the segmented image of any point p in the image is obtained by using the following formula: The obtained segmented image is removed from the small connected domain, and the image is subjected to a closing operation, hole filling and edge smoothing to obtain the processed segmented image.

6. The multi-information fused hydrograph extraction method of claim 5, wherein: Before the corner points of the image are calculated by using the corner point detection algorithm, the histogram of the image after gray processing is equalized; and when the corner points of the image are calculated by using the corner point detection algorithm, non-maximum suppression is adopted.

7. The multi-information fused hydrograph extraction method of claim 6, wherein: The specific method of combining the processed segmented image with the original image is: the pixel points on the original image and the segmented image are one-to-one corresponding, then for any position P i (x,y), the corresponding RGB value in the original image is r i ,g i ,b i , and the corresponding value in the segmented image is th i , and the original image is processed as follows: The non-water body image is obtained.

8. The multi-information fused hydrograph extraction method of claim 1, wherein: The specific method for color clustering in step S2 is as follows: the color is divided into two categories, namely a green area and a yellow area, and all points in the green area are set as black to obtain a soil area image; All values of the distance map are normalized, and a threshold value t1 is selected using the Otsu algorithm, and all values d i are compared, and if d i >t1, the position corresponding to the soil area image is set to black. i All values of the distance map are normalized, and a threshold value t1 is selected using the Otsu algorithm, and all values d i are compared, and if d i >t1, the position corresponding to the soil area image is set to black. i All values of the distance map are normalized, and a threshold value t1 is selected using 9. The multi-information fused hydrograph extraction method of claim 1, wherein: If the original image is disturbed by clouds and fog, the image containing the clouds and fog is subjected to wavelet decomposition, the low-frequency component is subjected to homomorphic filtering, and the high-frequency component is subjected to nonlinear stretching and reconstruction; Or, the image without cloud and fog at the same spatial position is compensated, specifically: the cloud and fog image and the image without cloud and fog are matched in RGB three color histogram, the area R covered by the cloud and fog in the image is intercepted, and the corresponding cloud and fog-free contrast image R is intercepted ′ After R' is replaced into the position corresponding to R, image fusion is performed.

10. The multi-information fused hydrological fluctuation belt extraction method according to claim 9, characterized in that: For multiple cloud and fog interference images, each cloud and fog region can be replaced respectively.

Images