Multispectral remote sensing image shallow water region extraction method based on object adaptive NDWI threshold

Abstract

The invention relates to the technical field of remote sensing image processing, in particular to a multispectral remote sensing image shallow water region extraction method based on an object adaptive NDWI threshold value, which comprises the following steps of: S1, receiving an image, segmenting the image to generate an object region, and extracting a shallow water region by taking an object as a basic unit; s2, calculating an NDWI value of each object, determining an optimal NDWI threshold value of the shallow water region, and obtaining an initial shallow water region extraction result based on the optimal NDWI threshold value; and S3, performing optimization processing on the initial shallow water region extraction result in the S2 based on an object region growing algorithm, and outputting a final remote sensing image shallow water region extraction result. The method can effectively overcome the problems that the remote sensing image shallow water region extraction process is easily affected by image noise and the optimal NDWI threshold is difficult to accurately determine; and accurate shallow water area extraction results of the offshore and island reef areas in different environments are obtained.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing image processing, and in particular to a method for extracting shallow water areas from multispectral remote sensing images based on object-adaptive NDWI thresholds. Background Art

[0002] Currently, scholars at home and abroad have conducted diverse research on water extraction from remote sensing imagery, which can be categorized into traditional water extraction methods and deep learning-based methods. Traditional water extraction methods typically require the design of water index features, which are then used to extract water from remote sensing images. Commonly used water index features include the normalized difference water index (NDWI), modified NDWIW (MNDWI), normalized multi-band water index (NDMBWI), and contrast difference water index (CDWI). While traditional water index feature extraction methods can achieve good results in water extraction from remote sensing images, they are susceptible to image noise, and the difficulty in accurately setting an appropriate water index threshold leads to limited robustness and generalizability. Convolutional neural networks (CNNs), currently the most popular deep learning model, have been widely used in water extraction from remote sensing imagery. CNN-based water extraction methods for remote sensing images can achieve good water extraction results in various scenarios and are more robust and universal than traditional water extraction algorithms based on artificially designed water features. However, existing CNN-based water extraction methods can only distinguish between water bodies and non-water bodies, without further distinguishing between shallow water areas (water depths not exceeding 20 meters) and deep water areas (water depths greater than 20 meters). Satellite-driven bathymetry (SDB) typically requires obtaining shallow water areas from remote sensing images and then performing water depth inversion on these areas. Furthermore, there are currently no publicly available datasets for shallow water extraction from remote sensing images, which has greatly limited research progress on CNN-based shallow water area extraction from remote sensing images.

[0003] Current SDB research typically uses two methods: visual interpretation and NDWI-based shallow water extraction to identify shallow water areas in remote sensing imagery. Visual interpretation can achieve high-precision shallow water extraction, but suffers from low efficiency and limited automation. NDWI-based shallow water extraction is susceptible to image noise, and its accuracy is highly correlated with the NDWI threshold setting. However, the optimal NDWI threshold is often difficult to determine.

[0004] To solve the above problems, the present invention proposes a shallow water area extraction method for multispectral remote sensing images based on object-adaptive NDWI threshold. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to overcome the deficiencies of the prior art and provide a method for extracting shallow water areas from multispectral remote sensing images based on object-adaptive NDWI thresholds.

[0006] The present invention is achieved through the following technical solution: A method for extracting shallow water areas from multispectral remote sensing images based on object-adaptive NDWI thresholds comprises the following steps: S1. Receive an image, segment the image to generate an object region, and extract the shallow water region using the object as a basic unit to eliminate the influence of image noise on the shallow water region extraction; S2. Calculate the NDWI value of each object, determine the optimal NDWI threshold of the shallow water area, and obtain the initial shallow water area extraction result based on the optimal NDWI threshold; S3, the object-based region growing algorithm optimizes the initial shallow water area extraction result in S2 and outputs the final remote sensing image shallow water area extraction result.

[0007] In the above-mentioned S1, a simple linear iterative clustering algorithm is used to segment the image to generate object areas, which can effectively overcome the influence of "salt and pepper" noise in the image on image processing.

[0008] In the simple linear iterative clustering algorithm, the compactness coefficient is 10, and the number of superpixels generated is M / 25, where M is the number of pixels in the image.

[0009] The S2 includes the following sub-steps: S2-1. Calculate the NDWI value of each subject; S2-2, sorting in ascending order based on the NDWI value of each object; S2-3, using the maximum inter-class variance in two stages to determine the optimal NDWI threshold for shallow water areas in remote sensing images; S2-4. Classify the objects and obtain the initial shallow water area extraction results of the multispectral remote sensing image.

[0010] The calculation formula of the NDWI value of each subject in S2-1 is as follows: ; in, is the NDWI value for each object, To be included in the object label NDWI value of the pixel; is the number of pixels contained in the object.

[0011] The NDWI values ​​of objects in the land, deep water area and shallow water area in S2-3 show a step distribution trend. Based on the step distribution trend, the objects are divided into two stages, and the inflection points between different stages are obtained, thereby obtaining the optimal NDWI threshold for shallow water area extraction in remote sensing images. The two stages include stage one and stage two. The scope of Phase 1 is , is the number of objects, The first inflection point threshold is obtained by calculating the maximum inter-class variance , based on the threshold It can eliminate the impact of land areas on subsequent shallow water extraction; The scope of Phase II is , In ascending order The corresponding object number is in The second inflection point threshold is obtained by calculating the maximum inter-class variance , This is the optimal NDWI threshold in shallow water areas.

[0012] The objective function of maximum inter-class variance It is expressed as follows: ; in, and The foreground and background objects are The proportion of and are the mean NDWI values ​​of foreground objects and background objects, respectively.

[0013] The objects are classified in S2-4 according to the following formula:

[0014] in, Indicates the object category, 1 represents the object is in shallow water area, 0 represents the object is not in shallow water area; All objects judged as 1 are collected together to obtain the initial shallow water area extraction results of the remote sensing image.

[0015] In S3, an object-based region growing algorithm is used to optimize the initial shallow water area extraction result. The optimization process is to perform region growth with objects as basic units. Objects of the same category are merged into one region, and then the number of pixels contained in the region is classified and judged, thereby achieving optimization of the initial shallow water area extraction result.

[0016] The calculation formula for classifying and judging the number of pixels contained in the region in S3 is as follows: ; Where: Represents the area category; 1 and 0 represent shallow water area and non-shallow water area category labels respectively; Indicates the number of pixels contained in the area; and are the lower and upper thresholds of the number of pixels in the region, respectively. and are set to and , is the number of image pixels.

[0017] Compared with the prior art, the present invention has the following beneficial effects: The present invention provides a method for extracting shallow water areas from multispectral remote sensing images based on object-adaptive NDWI thresholds. This method can effectively overcome the problems of susceptibility to image noise and difficulty in accurately determining the optimal NDWI threshold during the shallow water area extraction process of remote sensing images, and obtain accurate shallow water area extraction results in near-coastal and island reef areas under different environments.

[0018] This application uses a simple linear iterative clustering algorithm to segment the image, generate object areas and use objects as basic processing units, overcoming the defect of traditional pixel-level methods that are susceptible to noise interference, making the shallow water area extraction results more stable.

[0019] This application proposes a two-stage maximum inter-class variance calculation, eliminating interference from land areas in stage 1 and distinguishing between shallow and deep water areas in stage 2. It automatically adapts to the NDWI value distribution characteristics of different scenarios without the need for manual threshold setting, significantly improving the universality of the method in different environments.

[0020] This application optimizes the initial results based on the object's region growing algorithm. By merging similar objects to form regions and determining the number of pixels in the regions, it can effectively eliminate false detections and missed detections caused by interference factors such as clouds, coastal waves, and ships on the sea, making the final extraction results more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is the application flow chart; Figure 2It is an ascending trend chart of the object's NDWI value; Figure 3 This is the optimization result diagram of the initial shallow water area extraction result; Figure 4 Schematic diagram of region growing based on object categories. DETAILED DESCRIPTION

[0022] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0023] Reference Figures 1-4 ,The shallow water area extraction method of multispectral remote sensing image based on object ,adaptive NDWI threshold includes the following steps: S1. Receive an image, segment the image to generate an object region, and extract the shallow water region using the object as a basic unit to eliminate the influence of image noise on the shallow water region extraction; In S1, a simple linear iterative clustering algorithm is used to segment the image to generate object regions, which can effectively overcome the influence of "salt and pepper" noise in the image on image processing.

[0024] Furthermore, in the process of generating objects using the simple linear iterative clustering algorithm, the image color is first converted from the RGB color space to the CIELAB color space; then the color attribute and spatial attribute of each pixel in the image are combined into a five-dimensional feature vector , all seed points are evenly distributed in the image, based on the five-dimensional feature vector at the seed point Clustering and segmenting pixels in the neighborhood of is used to generate objects. The similarity measure between pixels and seed points is calculated as follows.

[0025] ; ; ; Where, and Represent the spectral distance and spatial distance between the pixel and the seed point respectively; Represents the similarity metric distance between the pixel and the seed point; is the five-dimensional feature vector of the pixel; is the five-dimensional feature vector of the seed point; is the distance between seed points; is the compactness coefficient in the simple linear iterative clustering algorithm, which is set to 10 in this application. The number of superpixels generated is M / 25, where M is the number of pixels in the image.

[0026] S2. Calculate the NDWI value of each object, determine the optimal NDWI threshold of the shallow water area, and obtain the initial shallow water area extraction result based on the optimal NDWI threshold; The S2 includes the following sub-steps: S2-1. Calculate the NDWI value of each subject; The calculation formula of the NDWI value of each subject in S2-1 is as follows: ; in, is the NDWI value for each object, To be included in the object label NDWI value of the pixel; is the number of pixels contained in the object.

[0027] in, , and are the green band and near-infrared band reflectance of the pixel, respectively.

[0028] S2-2, sort in ascending order based on the NDWI value of each object; the sorting results refer to Figure 2 .

[0029] S2-3, using the maximum inter-class variance in two stages to determine the optimal NDWI threshold for shallow water areas in remote sensing images; Reference Figure 2 ,The NDWI values ​​of objects in land, deep water area and shallow water area in S2-3 have an obvious stage step distribution trend, based on which they are divided into two stages. By obtaining the inflection points between different stages, the optimal NDWI threshold for shallow water area extraction in remote sensing images can be obtained. ,Furthermore, the two stages include stage one and stage two; The scope of Phase 1 is , is the number of objects, The first inflection point threshold is obtained by calculating the maximum inter-class variance , based on the threshold It can eliminate the impact of land areas on subsequent shallow water extraction; The scope of Phase II is , In ascending order The corresponding object number is in The second inflection point threshold is obtained by calculating the maximum inter-class variance , This is the optimal NDWI threshold in shallow water areas.

[0030] The objective function of maximum inter-class variance It is expressed as follows: ; in, and The foreground and background objects are The proportion of and are the mean NDWI values ​​of foreground objects and background objects, respectively.

[0031] S2-4, based on the optimal NDWI threshold The objects are classified and the initial shallow water area extraction results of multispectral remote sensing images are obtained.

[0032] The objects are classified in S2-4 according to the following formula:

[0033] in, Indicates the object category, 1 represents the object is in shallow water area, 0 represents the object is not in shallow water area; All objects judged as 1 are collected together to obtain the initial shallow water area extraction results of the remote sensing image.

[0034] S3, the object-based region growing algorithm optimizes the initial shallow water area extraction result in S2 and outputs the final remote sensing image shallow water area extraction result.

[0035] Due to the influence of objects such as clouds, coastal waves and ships on the sea surface in the image, there are a small number of missed detections and false detections in the initial shallow water area extracted, such as Figure 3 The initial shallow water area extraction result is shown in FIG. Therefore, the initial shallow water area extraction result needs to be further optimized to eliminate the influence of the above objects on the shallow water area extraction. The object-based region growing algorithm is used in S3 to optimize the initial shallow water area extraction result. The optimization process is as follows: Figure 4 shown.

[0036] exist Figure 4 In the process, the object is used as the basic unit for region growth. Objects of the same category are merged into one region. Then the number of pixels contained in the region is classified and judged, thereby optimizing the initial shallow water area extraction results. The results after optimization are as follows: Figure 3 shown.

[0037] The calculation formula for classifying and judging the number of pixels contained in the region in S3 is as follows: ; Where: Represents the area category; 1 and 0 represent shallow water area and non-shallow water area category labels respectively; Indicates the number of pixels contained in the area; and are the lower and upper thresholds of the number of pixels in the region, respectively. and are set to and , is the number of image pixels.

[0038] The above descriptions are merely optional embodiments of the present invention and do not limit the patent scope of the present invention. All equivalent structural transformations made using the contents of the present invention specification under the concept of the present invention, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present invention.

Claims

1. A shallow water area extraction method based on multispectral remote sensing imagery based on object-adaptive NDWI threshold, characterized by: The steps include: S1, receiving an image, segmenting the image to generate object regions, and extracting shallow water areas using the objects as basic units; S2. Calculate the NDWI value of each object, determine the optimal NDWI threshold of the shallow water area, and obtain the initial shallow water area extraction result based on the optimal NDWI threshold; S3, the object-based region growing algorithm optimizes the initial shallow water area extraction result in S2 and outputs the final remote sensing image shallow water area extraction result.

2. The shallow water area extraction method of multispectral remote sensing image based on object adaptive NDWI threshold according to claim 1 is characterized in that: In S1, a simple linear iterative clustering algorithm is used to segment the image to generate object regions.

3. The shallow water area extraction method of multispectral remote sensing image based on object adaptive NDWI threshold according to claim 2 is characterized in that: In the simple linear iterative clustering algorithm, the compactness coefficient is 10, and the number of superpixels generated is M / 25, where M is the number of pixels in the image.

4. The shallow water area extraction method of multispectral remote sensing image based on object adaptive NDWI threshold according to claim 1 or 2 is characterized in that: The S2 includes the following sub-steps: S2-1. Calculate the NDWI value of each subject; S2-2, sorting in ascending order based on the NDWI value of each object; S2-3, using the maximum inter-class variance in two stages to determine the optimal NDWI threshold for shallow water areas in remote sensing images; S2-4. Classify the objects and obtain the initial shallow water area extraction results of the remote sensing image.

5. The shallow water area extraction method of multispectral remote sensing image based on object adaptive NDWI threshold according to claim 4 is characterized in that: The calculation formula of the NDWI value of each subject in S2-1 is as follows: ； in, is the NDWI value for each object, To be included in the object label NDWI value of the pixel; is the number of pixels contained in the object.

6. The method for extracting shallow water areas from multispectral remote sensing images based on object-adaptive NDWI threshold according to claim 4, characterized in that: The NDWI values ​​of objects in the land, deep water area and shallow water area in S2-3 show a step distribution trend, which is divided into two stages based on the step distribution trend, and the inflection points between different stages are obtained, thereby obtaining the optimal NDWI threshold for shallow water area extraction in remote sensing images. The two stages include stage 1 and stage 2. The scope of Phase 1 is , is the number of objects, The first inflection point threshold is obtained by calculating the maximum inter-class variance ; The scope of Phase II is , In ascending order The corresponding object number is in The second inflection point threshold is obtained by calculating the maximum inter-class variance , This is the optimal NDWI threshold in shallow water areas.

7. The method for extracting shallow water areas from multispectral remote sensing images based on object-adaptive NDWI threshold according to claim 6, characterized in that: The objective function of maximum inter-class variance It is expressed as follows: ； in, and The foreground and background objects are The proportion of and are the mean NDWI values ​​of foreground objects and background objects, respectively.

8. The method for extracting shallow water areas from multispectral remote sensing images based on object-adaptive NDWI threshold according to claim 6, characterized in that: The objects are classified in S2-4 according to the following formula: in, Indicates the object category, 1 represents the object is in shallow water area, 0 represents the object is not in shallow water area; All objects judged as 1 are collected together to obtain the initial shallow water area extraction results of the remote sensing image.

9. The method for extracting shallow water areas from multispectral remote sensing images based on object-adaptive NDWI threshold according to claim 1, characterized in that: In S3, an object-based region growing algorithm is used to optimize the initial shallow water area extraction result. The optimization process is to perform region growth with objects as basic units. Objects of the same category are merged into one region, and then the number of pixels contained in the region is classified and judged, thereby achieving optimization of the initial shallow water area extraction result.

10. The method for extracting shallow water areas from multispectral remote sensing images based on object-adaptive NDWI threshold according to claim 9, characterized in that: The calculation formula for classifying and judging the number of pixels contained in the region in S3 is as follows: ； Where: Represents the area category; 1 and 0 represent shallow water area and non-shallow water area category labels respectively; Indicates the number of pixels contained in the area; and are the lower and upper thresholds of the number of pixels in the region, respectively. and are set to and , is the number of image pixels.

Images