Multi-Feature Continuous Extraction Method for Multiple Types of Coastal Water Bodies Based on Landsat-8 Images

Abstract

The present invention provides a multi-feature continuous extraction method for multi-type waters in the coastal area based on Landsat-8 images, comprising the following steps: S1. Obtain Landsat-8 images, and preprocess the Landsat-8 images to obtain spectral reflectance images; S2. Based on the spectral reflectance images, extract waters in the coastal zone to achieve a rough positioning of the coastal zone. The extraction of waters in the coastal zone includes water body extraction and water body edge enhancement; S3. Continuously extract the extracted water body areas according to preset water body types, and the water body types include aquaculture ponds, salt pans, rivers and lakes. The present invention continuously extracts multi-type waters in the coastal area based on regional projection, spectral and geometric features, and can improve the accuracy of water body extraction.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular, to a multi-feature continuous extraction method for multi-type waters along the coast based on Landsat-8 images. Background Art

[0002] The ground objects in the coastal area are complex and the water types are diverse. There are the following four challenges in identifying multi-type waters along the coast using Landsat-8 multispectral images:

[0003] (1) The edges of the waters are blurred. In medium-resolution images, due to the relatively low image resolution, the aquaculture ponds and salt pans are separated only by dams. The dams are likely to form mixed pixels with the ponds, resulting in a change in their spectral characteristics. The edges of some aquaculture ponds may be broken, or there may be a phenomenon of "adhesion" to each other, which brings difficulties to the extraction of aquaculture ponds.

[0004] (2) The spectral characteristics of different waters are similar.

[0005] (3) The shapes and sizes of the aquaculture ponds and salt pans are different. Due to geographical location and spatial structure, etc., there are large differences in the construction specifications of aquaculture ponds themselves. Aquaculture ponds are divided into regularly aggregated ponds and irregular large ponds, and it is difficult to describe aquaculture ponds with unified features. The salt pan consists of evaporation ponds, storage ponds and crystallization ponds, and their shapes and sizes are also different.

[0006] (4) The geometric characteristics of the broken rivers are similar to those of aquaculture ponds, and the geometric characteristics of isolated reservoirs and lakes are similar to those of aquaculture ponds.

[0007] In summary, at present, the coastal environment is complex, various ground objects are unevenly distributed, and the extraction of waters is easily interfered by other nearby high-reflectivity building roofs and shadows, making the extraction more difficult. In addition, the spectral and geometric characteristics between different waters are similar. Therefore, it is very difficult to effectively extract multi-type waters in a large range only by relying on traditional classification methods or single spectral indices. Summary of the Invention

[0008] In view of the deficiencies of the prior art, the present invention provides a multi-feature continuous extraction method for multi-type waters along the coast based on Landsat-8 images. The present invention continuously extracts multi-type waters along the coast based on regional projection, spectral and geometric features, and can improve the accuracy of water extraction.

[0009] The technical means adopted by the present invention are as follows:

[0010] A multi-feature continuous extraction method for multi-type waters along the coast based on Landsat-8 images, comprising the following steps:

[0011] S1. Obtain a Landsat-8 image, preprocess the Landsat-8 image, and obtain a spectral reflectance image;

[0012] S2. Based on the spectral reflectance image, extract the water bodies within the coastal zone to achieve a rough positioning of the coastal zone. The extraction of the water bodies within the coastal zone includes water body extraction and water body edge enhancement;

[0013] S3. Continuously extract the extracted water body areas according to the preset water body types. The water body types include aquaculture ponds, salt pans, rivers, and lakes.

[0014] Further, preprocessing the Landsat-8 image includes radiometric calibration and atmospheric correction of the Landsat-8 image;

[0015] The radiometric calibration includes: using the radiometric calibration tool in ENVI5.3 software to automatically read the calibration parameters in the header file to complete the radiometric calibration work;

[0016] The atmospheric correction includes: using the Radiometric Calibration toolbox to complete the atmospheric correction work according to the preset parameters. The preset parameters include the image longitude and latitude, imaging time, and sensor type.

[0017] Further, in S2, water body extraction and water body edge enhancement include:

[0018] Adopt an improved normalized difference water index to extract water bodies;

[0019] Adopt the Laplacian operator and the Hessian operator for water body edge enhancement.

[0020] Further, in S3, continuously extracting the extracted water body areas according to the preset water body types includes extracting aquaculture ponds based on the spatial region projection rule, specifically including:

[0021] Divide the image according to the extension angles of each water body to generate water body maps at different angles

[0022] Locate the aggregated feature water bodies for each angle of the water body map;

[0023] Obtain the projection differences of the located aggregated feature water bodies, and perform aquaculture pond extraction according to the projection differences.

[0024] Further, in S3, continuously extracting the extracted water body areas according to the preset water body types also includes extracting salt pans based on spectral and spatial search, specifically including:

[0025] Obtain each water body object in the water body set;

[0026] Calculate the spectral reflectance of each speed limit in each water body object;

[0027] Judge whether the maximum band value of the spectral reflectance at each pixel position is equal to the red band, and determine whether the water body belongs to a salt pan according to the judgment result.

[0028] Furthermore, for salt pan extraction based on spectral and spatial search, it further includes:

[0029] Based on the geometric characteristics of the water body, conduct a secondary judgment on the water bodies determined to be salt pans, and successively remove the river area, lake area, and tidal flat area to obtain the final salt pan extraction result.

[0030] Furthermore, in S3, continuous extraction of the extracted water body areas according to the preset water body types further includes river and lake extraction based on the geometric characteristics of roundness, specifically including:

[0031] Obtain the roundness of each water body object, where the roundness of the water body object is obtained according to the area of the water body object, the contour length of the water body, and the maximum distance from the center of the water body object to its contour;

[0032] Judge the water body category as a river or a lake according to the roundness of the water body object, where the roundness value of the river is less than that of the lake.

[0033] Compared with the prior art, the present invention has the following advantages:

[0034] This application uses Landsat-8 images and presents a continuous binary feature discrimination-based coastal multi-type water body extraction algorithm based on regional projection, spectral, and geometric features. First, this method utilizes the similarity of the local spatial direction and distance of aquaculture ponds, and further extracts aquaculture ponds through regional projection. Secondly, this method utilizes the spectral characteristics of salt pans and combines the spatial distribution characteristics of salt pans to extract salt pans. Finally, this method utilizes the differences in the geometric characteristics of rivers and lakes to extract rivers and lakes respectively. Through experimental comparison, it is proved that the method of this application can quickly extract four types of water bodies and improve the extraction accuracy at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0036] Figure 1This is the multi-feature continuous extraction flow chart of multi-type waters along the coast based on Landsat-8 images in the embodiments of the present invention.

[0037] Figure 2 This is the schematic diagram of the pond distribution in the aquaculture area in the embodiments of the present invention.

[0038] Figure 3 This is the schematic diagram of the projection of aquaculture ponds and salt pans in the embodiments of the present invention. Among them, (a) represents the schematic diagram of aquaculture ponds, (b) represents the schematic diagram of salt pans, (c) represents the projection schematic diagram of aquaculture ponds, and (d) represents the projection schematic diagram of salt pans.

[0039] Figure 4 This is the schematic diagram of the angle distribution and water bodies in the embodiments of the present invention.

[0040] Figure 5 This is the schematic diagram of the aggregated water bodies in a certain angle range in the embodiments of the present invention.

[0041] Figure 6 This is the multi-band spectral curve diagram of multi-water bodies along the coast in the embodiments of the present invention. Among them, (a) represents the spectral curve diagram of Liaoning region, (b) represents the spectral curve diagram of Tianjin region, (c) represents the spectral curve diagram of Shandong region, and (d) represents the spectral curve diagram of Jiangsu region.

[0042] Figure 7 This is the schematic diagram of removing non-salt pan areas in the embodiments of the present invention. Among them, (a) represents the distribution schematic diagram of C large , (b) represents that the length of the buffer zone extension is 60m, (c) represents detecting whether there are other potential salt pan water bodies in the detection buffer zone, (d) represents the salt pan, and (e) represents the removed non-salt pan. Detailed implementation manners

[0043] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0044] As Figure 1 shown, this embodiment provides a multi-feature continuous extraction method for multi-type waters along the coast based on Landsat-8 images, including the following steps:

[0045] S1. Obtain the Landsat-8 image, and preprocess the Landsat-8 image to obtain the spectral reflectance image;

[0046] S2. Extract the water bodies within the coastal zone based on the spectral reflectance image to achieve a rough positioning of the coastal zone. The extraction of the water bodies within the coastal zone includes water body extraction and water body edge enhancement;

[0047] S3. Continuously extract the extracted water body areas according to the preset water body types. The water body types include aquaculture ponds, salt pans, rivers, and lakes.

[0048] Specifically, the method of the present invention mainly includes three stages. The first stage is the preprocessing of Landsat-8 images, mainly radiometric calibration and atmospheric correction, the purpose of which is to obtain spectral reflectance with actual physical significance. The second stage is the extraction of water bodies within the coastal zone, including water body extraction, water body edge enhancement, and rough positioning of the coastal zone, the purpose of which is to extract the water bodies within the coastal zone. The third stage is the continuous extraction of multi-type water bodies. The main idea is to analyze the characteristics of different water bodies and propose extraction methods. First, in order to reduce the interference of other water bodies on the extraction of aquaculture ponds, an extraction algorithm for aquaculture ponds based on aggregated area projection is given; the main idea is to obtain potential aggregated areas by using the characteristics of aggregated distribution of aquaculture ponds and the similarity of the relative angles and distances between water bodies, and projection can obtain the regularity of the aquaculture pond area and the proximity of adjacent aquaculture ponds. Then, the final result of aquaculture ponds can be obtained according to the statistical characteristics of the length and width of the aggregated area after projection. Second, in order to extract salt pans, a spectral-spatial search extraction algorithm is given; the main idea is to first give a significant reflectance area feature to describe the number of pixels with the red band as the maximum reflectance band within a water body object, which is used to extract the crystallization ponds and evaporation ponds of salt pans. In order to reduce the interference of other water bodies, spatial search is combined with the characteristics of aggregated distribution of salt pans to remove interference. In addition, some reservoirs may not be extracted. Subsequently, using the spatial structure relationship of each salt pond in the salt pan, the extracted crystallization ponds and evaporation ponds are locally extended in the direction to obtain the evaporation ponds of the salt pan, and the final result of the salt pan is obtained by combining the above extraction results. Finally, in order to extract rivers and lakes, a geometric feature based on roundness is given to describe the roundness of water body objects to distinguish rivers and lakes, and finally a complete extraction result of multi-type water bodies is obtained.

[0049] The following combines specific application examples to elaborate on the specific solutions of each step.

[0050] S1. Data preprocessing.

[0051] Preprocess the downloaded Landsat-8 original remote sensing images, which are divided into radiometric calibration and atmospheric correction. Radiometric calibration is completed using the Radiometric Calibration tool in ENVI 5.3 software, which can automatically read the calibration parameters in the header file to complete the radiometric calibration work. Atmospheric correction is also completed under the Radiometric Calibration toolbox, which requires manual setting of some parameters to complete the atmospheric correction work, including image longitude and latitude, imaging time, and sensor type. The preprocessing of the image is completed through the above two parts of processing.

[0052] S2, water body extraction and edge enhancement.

[0053] This application uses the improved Normalized Difference Water Index (Modified NDWI, MNDWI) to extract water bodies, and this index can enhance the edge structure characteristics of aquaculture ponds. However, due to the limited resolution of the image, there are fractures in the dams of some aquaculture ponds and salt pans. Therefore, in order to highlight their spatial structure characteristics, this application uses the Laplace operator and the Hessian operator. The Laplace operator can enhance the edge and detail information of the image, while the Hessian operator maintains the overall structure of the water body well.

[0054] S3, continuous discrimination of multiple types of coastal water bodies based on regional projection, spectral and geometric features.

[0055] Although the water index, Hessian matrix, and Laplace operator can effectively extract water bodies and enhance the edges of water bodies, due to the similar spectral characteristics of multiple types of water bodies, relying solely on spectral features cannot effectively discriminate water bodies such as lakes, rivers, aquaculture ponds, and salt pans. For these four different water bodies, the present invention has respectively studied their characteristics and extraction methods.

[0056] S301, aquaculture pond extraction algorithm based on spatial region projection rules

[0057] Different from the spatially random distribution of isolated water bodies such as lakes and reservoirs, salt pans and aquaculture ponds are artificially constructed. To minimize their construction costs, they show an aggregated distribution. The characteristics of the two are analyzed below. The structure of the salt pan is fixed and mainly consists of relatively large storage ponds, evaporation ponds with regular shapes, and crystallization ponds, with a relatively high degree of aggregation. Aquaculture ponds are generally rectangular and are separated into enclosed small water bodies by narrow roads or dikes, but often gather together to form a relatively large aquaculture area. It is often difficult to extract salt pans and aquaculture ponds only considering the spatial proximity of water bodies. The difficulty lies in that the spatial positions of aquaculture ponds and each pond in the salt pan are usually relatively close, which makes it difficult to distinguish between salt pans and aquaculture ponds. Nevertheless, there are still differences in the spatial characteristics of salt pans and aquaculture ponds. The differences are reflected in the different shapes and sizes of each salt pond in the salt pan. The storage pond is generally large, while the evaporation pond and crystallization pond are small. Aquaculture ponds are composed of densely adjacent aquaculture ponds with similar shapes and sizes. This provides inspiration for distinguishing the two types of water bodies. Below, this application will study the method for extracting aquaculture ponds.

[0058] Aquaculture ponds usually consist of multiple aggregated areas. Suppose they are composed of three aggregated areas A, B, and C, and suppose each aggregated area contains K ponds. As Figure 2 shown, the aquaculture ponds in each aggregated area have similar shape characteristics, including similar lengths and widths between adjacent ponds, and similar distances and directions between adjacent ponds. These characteristics characterize the aggregation characteristics of an aggregated area. The ponds in the salt pan also aggregate. How to use the aggregation characteristics to describe the spatial differences between aquaculture ponds and salt pans is a key issue. Can the projection be used to characterize the aggregation differences between the salt pan and the aquaculture area? Below, the aquaculture ponds and salt pan water bodies in the vertical direction are taken as examples for analysis respectively. Figure 3 (a) shows four aquaculture ponds. Figure 3 (b) shows the salt pan. The salt ponds in the salt pan usually have different shapes and sizes. The frames from left to right are the storage pond, evaporation pond, and crystallization pond respectively. After the projection of these two types of water bodies, they are shown as Figure 3 (c) and (b) respectively. The aquaculture ponds show four rectangles. The heights and widths of these rectangles are similar, and the distances between adjacent rectangles are similar. The projection of the salt pan shows three rectangles. Although the distances between adjacent rectangles are similar, the heights and widths of these rectangles are not as similar as those of the aquaculture ponds. Therefore, this application describes the differences between aquaculture ponds and salt pans based on the aggregated projection of water body space to extract aquaculture ponds.

[0059] The general idea is: (1) Divide each water body in the entire image according to angles to generate water body maps at different angles; (2) Locate the water bodies with aggregation characteristics for each angle of the water body map; (3) Extract aquaculture ponds using the projection differences of the aggregated water bodies. First, the generation of a water body map at a certain angle is elaborated. For example, divide the interval into nine subsets, asFigure 4 For the equal - angle division shown in (a), the central - line angle θ of each subset is j as follows:

[0060]

[0061] Suppose an aggregated water - body region S is composed of several water - body objects. For any water - body object S q , the direction angle is defined as the angle between the major axis of the ellipse with the same standard second - order central moment as its water - body region and the horizontal direction. Then, for this water - body object S q existing in the interval where θ j is located, it must satisfy:

[0062] |θ(S q ) - θ j | ≤ T(3.2)

[0063] where T is 10 degrees. In this way, any closed water - body can determine the direction interval where it exists through Equation (3.2). All water - bodies in the entire image can be decomposed into water - body maps generated by multiple angle intervals, as shown in Figure 4 (b), and examples of water - bodies in each interval are given.

[0064] Next, locate the water - bodies with aggregation features in the water - body map at a certain angle. The water - bodies with aggregation features refer to those where the distance between adjacent water - bodies in the aggregation region is close, and the relative angles between adjacent water - bodies are approximately parallel. To illustrate this feature, first, we use distance to measure the former. For this purpose, the Euclidean - distance difference between the centroids of the three nearest single water - bodies is defined to measure the adjacency between adjacent water - bodies. Second, the angular difference between the centroids of the three nearest single water - bodies is defined to measure the similarity of the directions of the three water - bodies. Suppose the water - bodies in a certain angle interval, as shown in Figure 5 , assume the current water - body is S i , and its two nearest water - bodies are S i-1 and S i+1 . If the Euclidean distance between S i-1 and S i is defined as d i-1 , and the distance between S i and S i+1 is defined as d i . Then d i-1 and d i are respectively expressed as:

[0065]

[0066] and

[0067]

[0068] Among them, (x i+1 , y i+1 ), (x i , y i ) and (x i-1 , y i-1 ) are the centroids of water bodies S i+1 , S i and S i-1 respectively. If the distance between the centroids of two adjacent water bodies is large, the two water bodies may not be in the same aggregation area; if the distance between the centroids of two adjacent water bodies is small, they may be in the same aggregation area.

[0069] In addition to the close distance between adjacent water bodies, the approximate parallelism between adjacent water bodies is another feature. The measurement of parallelism can be carried out by measuring the angle difference between the centroids of three adjacent water bodies. If is defined as the angle of the centroid of water body S i-1 with respect to the horizontal direction to the centroid of water body S i , is the angle of the centroid of S i with respect to the horizontal direction to the centroid of S i+1 , then and are respectively expressed as:

[0070]

[0071] and

[0072]

[0073] If the angular differences of equations (3.5) and (3.6) are small, the three water body objects are approximately parallel, otherwise there must be at least one non-approximately parallel water body. Therefore, the approximately parallel water bodies at angle θ j form the following set:

[0074]

[0075] Among them, M is the total number of water body objects at the current angle, and d T are the thresholds of angle and distance respectively.

[0076] When approximately parallel water bodies are detected in each angular interval, the salt pan and the aquaculture pond often satisfy this feature at the same time, and it is necessary to distinguish them by projection. Before projection, the current approximately parallel water bodies need to be rotated, and the purpose is to make the water bodies convenient for horizontal and vertical projection after rotation. Taking a single water body S i as an example, assuming its angle is θ i , then rotating it to the vertical direction requires coordinate transformation. If (x, y) is S iThe coordinate values, then the rotated coordinates are expressed as (x T , y T ), where Project it at the current angle, then it can be expressed as:

[0077]

[0078] and

[0079]

[0080] Among them, V x is the vertical projection of the water body, V y is the horizontal projection of the water body. Both projections are one-dimensional vectors. S i (x T , y T ) is the binary image value (0 or 1) of the coordinate point (x T , y T ). The ideal single water body presents a single strip or an approximate rectangle in the vertical projection. In fact, due to resolution reasons, a single aquaculture pond cannot have a regular rectangle and often presents an irregular rectangle. In this way, the vertical projection will show a single strip with multiple peaks. To determine the length of the water body, currently, the length of the water body corresponding to the maximum value of the vertical projection of the current approximately parallel water body is selected. Here, it can also be called the height of the projection. To determine the width of the water body, the maximum value of the horizontal projection of the current approximately parallel water body is selected as the current water body width. Then the height and width of the single water body after projection are respectively expressed as:

[0081] H(S i ) = max{V x}(3.10)

[0082] and

[0083] K(S i ) = max{V y}(3.11)

[0084] Among them, H(S i ) is the length or height of the water body object S i , and K(S i ) is the width of the water body object.

[0085] According to the analysis of the single water body projection in formulas (3.8)-(3.11), then in an angle interval, how to utilize the horizontal and vertical projection characteristics of multiple approximately parallel water bodies to distinguish the aquaculture pond and the salt field is the key. According to Figure 3 characteristics, this application uses the means of the heights and widths of multiple such water body objects as the features for measuring the difference between the two, then the two are respectively expressed as:

[0086]

[0087] and

[0088]

[0089] In most cases, since salt pans are composed of evaporation ponds, reservoirs and crystallization ponds of different shapes and sizes, their strip distribution is irregular after projection. At the same time, the strips of projection of a few salt pans are more regular, which makes it difficult to distinguish salt pans from aquaculture ponds. However, salt pans show block characteristics, so the overall characteristics of salt pans can be seen from their average height H. R and the average width K R Compared with salt pans, the aggregated aquaculture ponds are composed of aquaculture ponds of similar shapes and sizes. After projection, they will show a more regular strip shape. The average height H after projection is R and the average width K R It is approximately equal to the length and width of the aquaculture pond. At the same time, the aquaculture pond presents the characteristics of a long and thin strip. Therefore, its average height H R Larger, average width K R Small, so the paper gives a judgment based on the mean of length and width. R >H T At the same time K R <K T When , the water body is a breeding pond at the current angle, otherwise it is a salt pan. R and K R are the height and width means of the clustered area, H T and K T are the thresholds for height and width respectively. In this way, the above discrimination method can effectively extract the breeding pond at each angle interval.

[0090] S302. Salt pan extraction based on spectral and spatial search

[0091] This step studies the method of extracting salt pans, and masks the aquaculture ponds to be extracted in order to reduce the interference with the extraction of salt pans. The salt pan consists of an evaporation pond, a reservoir pond, and a crystallization pond. The salt-making process in the salt pan is to introduce seawater into the evaporation pond. The water in the seawater is evaporated, and crystal salts will gradually precipitate, usually showing a bright white color, while the colors of other water bodies are usually dark blue or blue. In order to analyze the spectral differences between salt pans and other water bodies, this application selects 4 main salt pan production areas from north to south in China, namely Liaoning, Tianjin, Shandong, and Jiangsu. 100 sample points are respectively selected for the areas with bright white colors in the crystallization ponds and evaporation ponds of the salt pans, aquaculture ponds, lakes, and rivers. Samples for 4 quarters of 2021 are selected for each area. Due to the influence of clouds in Tianjin and Jiangsu areas, only samples for the 2nd and 3rd quarters are selected. The mean values of the spectral reflectance of each band at the sample points are statistically calculated, and the spectral response curves of each ground object are drawn according to the calculated mean values, as Figure 6 shown in the multi-band spectral curve diagram of multiple types of coastal water bodies.

[0092] As Figure 6 (a) shows the spectral curve diagram of each ground object in the Liaoning region, Figure 6 (b) is the spectral curve diagram of each ground object in the Tianjin region, Figure 6 (c) is the spectral curve diagram of each ground object in the Shandong region, Figure 6 (d) is the spectral curve diagram of each ground object in the Jiangsu region. The solid line is the mean value of the reflectance of the crystallization pond and the evaporation pond in multiple bands, the dotted line with dots represents the mean value of the reflectance of the aquaculture pond in multiple bands, the dotted line with triangles is the mean value of the reflectance of the lake in multiple bands, and the dotted line with squares is the mean value of the reflectance of the river in multiple bands. From Figure 6 it can be seen that the reflectance of the crystallization pond and the evaporation pond of the salt pan is the highest in the RED band, while the reflectance of other water bodies in the RED band is lower. Based on this, the difference in the RED band is used to distinguish the salt pan from other water bodies. We use the characteristic that the reflectance of the salt pan is the largest in the red band to propose a spectral-spatial metric to discriminate the spectral differences between the salt pan and other water bodies. Assume that the set R of water bodies includes several water body objects, expressed as:

[0093]

[0094] where l is the number of water body objects. Here, the water body object Ω t consists of several water body pixels with a pixel value of 1 and being connected. According to the above spectral analysis, we need to find out the maximum band of the spectral reflectance of each pixel of the water body object Ω t , expressed as:

[0095]

[0096] where, For the water body object Ω t The band with the maximum spectral value at the coordinate point (i, j), ρ k (i, j) is the water body object Ω t (i, j) is the reflectance value at the k-th band, K is the number of bands, K = 7. The maximum band value at each pixel position is obtained according to the above formula. If it is a salt pan, its maximum band value is the red band. Therefore, we need to determine whether the obtained maximum band value is equal to the red band (in the Landsat-8 image, the red band is the 4th band), denoted as It can be expressed as:

[0097]

[0098] When the maximum band is the red band, equals 1, otherwise equals 0. And if the water body object is a crystallization pond or evaporation pond of a salt pan, the number of bands with the maximum reflectance value in the red band within its area is relatively large. We use RPI to represent it:

[0099]

[0100] If the value of RPI is large, the possibility of it being a crystallization pond or evaporation pond of a salt pan is relatively large. Therefore, in this application, the salt pan and other water bodies are discriminated according to RPI. If RPI(Ω t ) > T RPI , then the current water body object Ω t is a salt pan, T RPI is the threshold, otherwise the current water body object Ω t is other water bodies (such as rivers and lakes). However, rivers, lakes and tidal flats on the coast are easily confused with salt pans. The reason is that the spectral reflectance of the sediment deposition area of coastal rivers and tidal flats is the largest in the red band, resulting in their RPI being similar to the characteristics of crystallization ponds and evaporation ponds. Therefore, these rivers, lakes and tidal flats need to be removed from the extracted potential salt pans. Rivers are usually slender. Combining geometric features, such as aspect ratio, density, area ratio, etc., can further remove rivers. And lakes and tidal flats generally show isolated distribution. They are removed using neighborhood features, such as Figure 7 shown, where C large represents the water body object that conforms to the above discrimination rules. Salt pans are separated by dams or roads and generally show an aggregated distribution, as shown by the solid box area in Figure 7 (a), while most tidal flats and lakes show isolated distribution, as shown by Figure 7As shown in the dashed box area in (a). Since the salt ponds of the salt pan are usually close to each other, their neighborhood features are used to remove non-salt pans from potential salt pans. The salt pans are usually separated by dams, and the width of the dams in the image is only one or two pixels. Since the resolution of the Landsat-8 image used in this paper is 30m, each area is extended outward by 60m to form a buffer zone, as shown in Figure 7 As shown in (b), a buffer zone is established in each area, such as Figure 7 As shown in (c), based on the densely distributed characteristics of salt pans, it is necessary to detect whether there are other potential salt pan water bodies in the detection buffer zone. If so, they are salt pan crystallization pools and evaporation pools. Figure 7 (d) If it does not exist, it is not a salt pan. Figure 7 (e) is shown, it will be removed. The main salt pan crystallization pool and evaporation pool area can be obtained through the above. Using the spatial distribution characteristics of the salt pan, the water reservoir of the salt pan is usually close to one side of the evaporation pool and crystallization pool, and is similar to the direction of the water reservoir, that is, approximately parallel. Therefore, by statistically analyzing the local directionality of the area obtained above, and using straight lines to represent the salt pools in different rows, the water reservoirs on both sides are found through the parallel characteristics to obtain the final salt pan result.

[0101] S303. Extract rivers and lakes based on the geometric features of roundness.

[0102] Next, we extract rivers and lakes. Rivers are generally winding and usually have elongated shape characteristics, while lakes are more round. Therefore, we propose a roundness E to distinguish rivers from lakes, which is defined as follows:

[0103]

[0104] Where F is the area of the water object, l max Indicates the maximum distance from the center of the water object to its contour, and L represents the contour length of the water body. The range of E is (0,1], which helps to describe the degree to which a closed area object is biased towards a circle. That is, the closer the shape of the water object is to a regular circle, the closer its corresponding E value is to 1. The slender structure corresponds to a small exponent value, so rivers usually correspond to small roundness values, and lakes correspond to larger roundness values.

[0105] In order to verify the effectiveness of the proposed method, the overlap rate (Intersection Over Union, IOU) and Kappa coefficient are used as accuracy indicators. IOU is the repetition rate between the extracted result and the true value image, expressed as:

[0106]

[0107] Among them, TP is the number of misclassified positive pixels, TN is the number of misclassified negative pixels, FP is the number of misclassified positive pixels, and FN is the number of misclassified negative pixels. It refers to the proportion of the number of correctly predicted pixels to the total number of pixels, that is, the accuracy rate. The larger the IOU value, the better the extraction performance.

[0108] The Kappa coefficient is defined as:

[0109]

[0110] Among them, p0 is the overall classification accuracy, that is, the sum of the number of correctly classified samples in each category divided by the total number of samples. p0 is defined as

[0111]

[0112] p e represents the sum of the products of the actual and predicted quantities corresponding to all categories respectively, and then divided by the square of the total number of samples, and is defined as

[0113]

[0114] To test the performance of the multi-type water body extraction algorithm of this patent, the water body classification algorithm based on pixels and object levels (Combining Pixel-and Object-Based Machine Learning for Identification of Water-Body Types, POB) is used as a comparison method.

[0115] To visually compare the performance of the two methods, this application uses two evaluation metrics, namely the Intersection over Union (IOU) and the Kappa coefficient, to quantitatively analyze the extraction results, as shown in Table 1. In the Dalian and Dongying regions, the IOU is approximately 75%, while the Kappa is around 80%. Compared with the comparative algorithm, both the IOU and the Kappa coefficient have increased by more than 5%. From the data in the table, it can be seen that the algorithms of this application and the comparative algorithm have better extraction effects on aquaculture ponds and rivers in the Dongying area. The reason is that the aquaculture ponds in the Dalian area are scattered, while those in the Dongying area are more regularly constructed and show a large-area aggregated distribution. However, the algorithms of this application and the comparative algorithm have better extraction effects on rivers in the Dongying area. The reason is that the rivers in the Dalian area are relatively narrow, resulting in breaks and missed extractions of rivers due to pixel reasons, and the inaccurate extraction of water body results also leads to a decrease in accuracy. In contrast, the rivers in the Dongying area are wide and long, and their geometric features are more obvious. For the extraction results of lakes, the algorithms of this application and the comparative algorithm have better extraction effects on lakes in the Dalian area. The reason is that the water system in the Dongying area is relatively developed, with a large number of various water bodies distributed, and it is easy to misclassify between water bodies. Among them, the extraction effect of the comparative algorithm on lakes is relatively poor. The main reason is that a large number of salt pans are distributed in the Dongying area, resulting in misclassification between a large number of lakes and salt pans. This is also because the comparative algorithm does not fully consider water bodies such as salt pans, resulting in a decrease in the accuracy of lake extraction.

[0116] Table 1 Comparison of Experimental Performance between the Method of the Invention and the Comparative Method

[0117]

[0118]

[0119] The present invention identifies multiple types of coastal waters based on Landsat-8 images. However, the coastal zone environment is complex, with uneven distribution of various ground objects. The extraction of water bodies is easily interfered by nearby building roofs with high reflectivity and shadows, making the extraction difficult. In addition, the spectral and geometric features between different water bodies are similar. Therefore, it is very difficult to effectively extract multiple types of water bodies over a large area relying solely on traditional classification methods or single spectral indices. To solve the above problems, this application uses Landsat-8 images and presents an extraction algorithm for multiple types of coastal waters based on continuous binary feature discrimination of regional projection, spectral, and geometric features. First, the algorithm utilizes the similarity of the local spatial direction and distance of aquaculture ponds and further extracts aquaculture ponds through regional projection. Second, the algorithm utilizes the spectral features of salt pans and combines them with the spatial distribution features of salt pans to extract salt pans. Finally, the algorithm extracts rivers and lakes separately using the differences in the geometric features of rivers and lakes. To verify the effectiveness of the patented algorithm of the present invention, experiments are carried out in two areas, Dongying, Shandong and Dalian, Liaoning, and verified through accuracy evaluation and comparison. The experimental results show that the Kappa and IOU of this patent are basically around 80%.

[0120] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some or all of the technical features. These modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-feature continuous extraction method for multi-type water bodies on the coast based on Landsat-8 images, characterized in that It includes the following steps: S1. Obtain Landsat-8 images, preprocess the Landsat-8 images, and obtain spectral reflectance images; S2. Based on the spectral reflectance images, extract the water bodies within the coastal zone to achieve a rough positioning of the coastal zone. The extraction of the water bodies within the coastal zone includes water body extraction and water body edge enhancement; S3. Continuously extract the extracted water body areas according to the preset water body types. The water body types include aquaculture ponds, salt pans, rivers, and lakes. Among them, continuously extracting the extracted water body areas according to the preset water body types includes extracting aquaculture ponds based on the spatial area projection rule, specifically including: Divide the image according to the extension angles of each water body to generate water body maps at different angles; Locate the water bodies with aggregated features in each water body map at different angles; Obtain the projection differences of the aggregated feature water bodies at each location, and extract the aquaculture ponds according to the projection differences, including obtaining the average height H after projection R and the average width K R , when H R >H T and at the same time K R <K T , the water body at the current angle is an aquaculture pond, where H T and K T are the discrimination thresholds for height and width respectively.

2. The multi-feature continuous extraction method for multi-type water bodies on the coast based on Landsat-8 images according to claim 1, wherein Preprocessing the Landsat-8 images includes radiometric calibration and atmospheric correction of the Landsat-8 images; The radiometric calibration includes: using the radiometric calibration tool in ENVI5.3 software to automatically read the calibration parameters in the header file to complete the radiometric calibration work; The atmospheric correction includes: using the Radiometric Calibration toolbox to complete the atmospheric correction work according to the preset parameters. The preset parameters include the longitude and latitude of the image, the imaging time, and the sensor type.

3. A multi-feature continuous extraction method for multi-type water bodies on the coast based on Landsat-8 images according to claim 1, characterized in that, In S2, the water body extraction and water body edge enhancement are carried out, including: Adopt an improved normalized difference water index to extract water bodies; Adopt the Laplace operator and the Hessian operator to enhance the water body edge.

4. A multi-feature continuous extraction method for multi-type water bodies on the coast based on Landsat-8 images according to claim 1, characterized in that, In S3, continuously extracting the extracted water body areas according to the preset water body types also includes extracting salt pans based on spectral and spatial search, specifically including: Obtain each water body object in the water body set; Calculate the spectral reflectance of each pixel in each water body object; Judge whether the maximum band value of the spectral reflectance at each pixel position is equal to the red band, and determine whether the water body to which it belongs is a salt pan according to the judgment result.

5. A multi-feature continuous extraction method for multi-type water bodies on the coast based on Landsat-8 images according to claim 4, characterized in that, Extracting salt pans based on spectral and spatial search also includes: Based on the geometric features of the water body, conduct a secondary judgment on the water bodies determined to be salt pans, and sequentially remove the river areas, lake areas, and tidal flat areas to obtain the final salt pan extraction result.

6. A multi-feature continuous extraction method for multi-type water bodies on the coast based on Landsat-8 images according to claim 4, characterized in that In S3, continuously extracting the extracted water body areas according to the preset water body types also includes extracting rivers and lakes based on the geometric feature of roundness, specifically including: Obtain the roundness of each water body object. The roundness of the water body object is obtained according to the area of the water body object, the contour length of the water body, and the maximum distance from the center of the water body object to its contour; Judge the water body category as a river or a lake according to the roundness of the water body object, where the roundness value of the river is less than that of the lake.

Images