A method and system for extracting a single aquaculture pond

Abstract

The application discloses a single aquaculture pond extraction method and system, relates to the technical field of target extraction, and comprises the following steps: acquiring a Sentinel-2 time series image set of a target aquaculture area; determining a maximum normalized difference water index image and a binary water body image corresponding to the target aquaculture area according to the Sentinel-2 time series image set; based on the maximum normalized difference water index image, gray-scale morphological erosion and Canny edge detection, performing multiple iterative segmentations on the binary water body image to obtain multiple initial water body objects; constructing a decision tree based on prior knowledge of the target aquaculture area; and screening the multiple initial water body objects based on the decision tree to obtain single aquaculture ponds corresponding to the target aquaculture area. The application realizes fine extraction of single aquaculture ponds.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of target extraction, in particular to a single aquaculture pond extraction method and system. BACKGROUND

[0002] The remote sensing extraction of aquaculture ponds mainly has the following challenges: (1) The extraction of aquaculture ponds is often based on high-resolution images as data sources. However, high-resolution images are difficult to obtain, and the spatial coverage and spectral resolution of the images are limited. Medium-resolution images are often used to extract aquaculture ponds connected together, i.e. aquaculture areas, and the extraction of aquaculture ponds is less. Current research tends to extract all ponds in the aquaculture area as a whole. However, compared to extracting aquaculture areas, extracting all aquaculture ponds is more important. Ponds are the smallest unit of aquaculture production, and different ponds may breed different aquatic products with large economic value differences. Therefore, the accurate extraction and mapping of individual aquaculture ponds are crucial for the fine management of aquaculture. Since aquaculture ponds are a whole, when disasters occur, it is more meaningful to assess the losses of the aquaculture industry based on the number of ponds than based on the area, which is a major feature of aquaculture industry different from other types of agriculture. (2) Medium-resolution images represented by Sentinel-2, Landsat-5 / -8 / -9 have larger coverage, finer spectral resolution and shorter revisit period. Some researchers have made corresponding work on medium-resolution aquaculture ponds, but the extraction results have a large number of interconnected ponds, and the extraction effect for individual ponds is not good. SUMMARY

[0003] The purpose of the present application is to provide a single aquaculture pond extraction method and system, which realizes the fine extraction of single aquaculture ponds.

[0004] To achieve the above purpose, the present application provides the following solutions:

[0005] A single aquaculture pond extraction method, comprising:

[0006] acquiring a Sentinel-2 time series image set of a target aquaculture area;

[0007] determining a maximum normalized difference water index image and a binary water image corresponding to the target aquaculture area according to the Sentinel-2 time series image set;

[0008] based on the maximum normalized difference water index image, gray-scale morphological erosion and Canny edge detection, performing multiple iterations of segmentation on the binary water image to obtain a plurality of initial water objects;

[0009] constructing a decision tree based on prior knowledge of the target aquaculture area;

[0010] filtering the plurality of initial water body objects based on the decision tree to obtain single aquaculture ponds corresponding to the target aquaculture area.

[0011] Optionally, according to the Sentinel-2 time series image set, the maximum normalized difference water body index image and the binary water body image corresponding to the target aquaculture area are determined, specifically including:

[0012] calculating the normalized difference water body index of each Sentinel-2 image in the Sentinel-2 time series image set; the normalized difference water body indexes of the plurality of Sentinel-2 images constitute NDWI time series data;

[0013] According to the Lyapunov criterion, the NDWI time series data is denoised;

[0014] Based on the maximum synthesis method, the NDWI time series data after denoising is combined to obtain the maximum normalized difference water body index image;

[0015] The maximum normalized difference water body index image is segmented by threshold segmentation method to obtain a binary water body image.

[0016] Optionally, based on the maximum normalized difference water body index image, gray-scale morphological erosion and Canny edge detection, each iteration of the multiple iterations of the binary water body image includes:

[0017] The maximum normalized difference water body index image is sequentially subjected to gray-scale morphological erosion processing and Canny edge detection processing to obtain a Canny edge image;

[0018] determining whether the current iteration value is a preset initial value;

[0019] If the current iteration value is the preset initial value, the binary water body image is segmented using the Canny edge image to obtain a plurality of first water body objects;

[0020] If the current iteration value is not the preset initial value, the Canny edge image is superimposed with the Canny edge image in the previous iteration segmentation to obtain an updated Canny edge image; then the binary water body image is segmented using the updated Canny edge image to obtain a plurality of first water body objects;

[0021] performing connectivity detection on any of the first water body objects to obtain a detection result;

[0022] if the detection result indicates that the first water body object passes the connectivity detection, marking the first water body object as an initial water body object, and stopping iteration;

[0023] if the detection result indicates that the first water body object fails the connectivity detection, entering the next iteration segmentation.

[0024] Optionally, performing connectivity detection on any of the first water body objects to obtain a detection result, specifically comprising:

[0025] calculating a landscape shape index and a convex hull perimeter ratio of the first water body object; the convex hull perimeter ratio is a ratio of a perimeter of the first water body object to a convex hull perimeter;

[0026] determining whether the landscape shape index satisfies a first preset threshold range and whether the convex hull perimeter ratio satisfies a second preset threshold range;

[0027] if the landscape shape index satisfies the first preset threshold range and the convex hull perimeter ratio satisfies the second preset threshold range, the detection result of the connectivity detection is that the first water body object passes the connectivity detection;

[0028] if the landscape shape index does not satisfy the first preset threshold range or the convex hull perimeter ratio does not satisfy the second preset threshold range, the detection result of the connectivity detection is that the first water body object fails the connectivity detection.

[0029] Optionally, screening the plurality of initial water body objects based on the decision tree to obtain a single water aquaculture pond corresponding to the target water aquaculture area, specifically comprising:

[0030] for any initial water body object, obtaining area data of the initial water body object, a median of corresponding NDWI time series data, a number of adjacent water body objects, and a land type in which the initial water body object is located; the land type includes farmland, building area, and water aquaculture area;

[0031] determining whether the area data of the initial water body object is greater than a first set value to obtain a first result;

[0032] if the first result indicates no, discarding the initial water body object;

[0033] if the first result indicates yes, determining whether the median of the NDWI time series data corresponding to the initial water body object is greater than or equal to a second set value to obtain a second result;

[0034] if the second result indicates no, discarding the initial water body object;

[0035] If the second result indicates yes, it is determined whether the number of adjacent water body objects of the initial water body object is greater than or equal to a third set value, to obtain a third result;

[0036] If the third result indicates no, the initial water body object is discarded;

[0037] If the third result indicates yes, it is determined whether the type of the land block where the initial water body object is located is a water body breeding area, to obtain a fourth result;

[0038] If the fourth result indicates no, the initial water body object is discarded;

[0039] If the fourth result indicates yes, the initial water body object is determined as a single aquaculture pond corresponding to the target aquaculture area.

[0040] To achieve the above purpose, the present application also provides the following technical solutions:

[0041] A single aquaculture pond extraction system, comprising:

[0042] An image acquisition module, configured to acquire a Sentinel-2 time series image set of a target aquaculture area;

[0043] An image processing module, configured to determine a maximum normalized difference water index image and a binary water image corresponding to the target aquaculture area according to the Sentinel-2 time series image set;

[0044] An image segmentation module, configured to perform multiple iterations of segmentation on the binary water image based on the maximum normalized difference water index image, gray-scale morphological erosion and Canny edge detection, to obtain a plurality of initial water body objects;

[0045] A decision tree construction module, configured to construct a decision tree based on prior knowledge of the target aquaculture area;

[0046] A single screening module, configured to screen the plurality of initial water body objects based on the decision tree, to obtain a single aquaculture pond corresponding to the target aquaculture area.

[0047] According to the specific embodiments provided by the present application, the following technical effects are disclosed:

[0048] The application discloses a single aquaculture pond extraction method and system, taking a 10-meter resolution Sentinel-2 image as a data source to explore the potential of medium-resolution images in aquaculture pond extraction. First, a water body is roughly extracted from a Sentinel-2 time series image set to obtain a maximum normalized difference water body index image and a binary water body image corresponding to a target aquaculture area. Then, a fusion grayscale morphology and edge detection iterative segmentation algorithm is proposed to extract potential aquaculture ponds from the water body index image. Subsequently, based on prior knowledge of the characteristics of aquaculture ponds, other water bodies and ground objects are removed from the potential aquaculture ponds to obtain single aquaculture ponds corresponding to the target aquaculture area, thereby realizing fine extraction of single aquaculture ponds and meeting the needs of actual production and life. BRIEF DESCRIPTION OF DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description only constitute some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0050] Figure 1 It is a flowchart of the single aquaculture pond extraction method of the present application.

[0051] Figure 2 It is a structure diagram of the decision tree of the present application.

[0052] Figure 3 It is a comparison diagram of the aquaculture pond extraction result obtained by the method of the present application and the pond area of the artificial sample.

[0053] Figure 4 It is a comparison diagram of the aquaculture pond extraction result obtained by the K-Means and the pond area of the artificial sample.

[0054] Figure 5 It is a comparison diagram of the aquaculture pond extraction result obtained by the G-Means and the pond area of the artificial sample.

[0055] Figure 6 It is a comparison diagram of the aquaculture pond extraction result obtained by the SNIC and the pond area of the artificial sample.

[0056] Figure 7 It is a structure diagram of the single aquaculture pond extraction system of the present application. DETAILED DESCRIPTION

[0057] With reference to the accompanying drawings, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all the other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0058] The present application aims to provide a single aquaculture pond extraction method and system, the input data is Sentinel-2 time series image and land use (Land Use-Land Cover, LULC) image, the output data is a single aquaculture pond vector, which can be realized based on the Google EarthEngine (GEE) platform, and the API of GEE can be called in the open-source python interactive interface of Google Colab to realize the fine extraction of the single aquaculture pond.

[0059] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below in combination with the drawings and specific embodiments.

[0060] Embodiment one

[0061] As shown in the figure, the present embodiment provides a single aquaculture pond extraction method, comprising: Figure 1

[0062] Step 100, acquiring a Sentinel-2 time series image set of a target aquaculture area.

[0063] Step 200, determining the maximum normalized difference water body index image and the binary water body image corresponding to the target aquaculture area according to the Sentinel-2 time series image set.

[0064] Step 200 specifically includes:

[0065] 1) calculating the normalized difference water body index (Normalized Difference Water Index, NDWI) of each Sentinel-2 image in the Sentinel-2 time series image set; the normalized difference water body indexes of multiple Sentinel-2 images constitute NDWI time series data.

[0066] The calculation formula of the normalized difference water body index of the Sentinel-2 image is:

[0067]

[0068] ​wherein, NDWI represents the normalized difference water index of the Sentinel-2 image, B3 represents the green band data corresponding to the Sentinel-2 image, and B8 represents the near-infrared band data corresponding to the Sentinel-2 image.

[0069] 2) According to the Laiyida criterion (3σ criterion), the NDWI time series data is denoised.

[0070] 3) Based on the maximum synthesis method, the NDWI time series data after denoising is combined to obtain a maximum normalized difference water index image (MNI).

[0071] 4) The threshold segmentation method is used to segment the maximum normalized difference water index image to obtain a binary water image. Specifically, the threshold segmentation method is used to identify the pixels with a value greater than or equal to 0 in the MNI as water pixels, thereby generating a binary water image (BWI).

[0072] In one specific practical application, step 200 can be performed on the Google EarthEngine (GEE) platform, and the Sentinel-2 time series image is input to the GEE platform. The output of the GEE platform is: a maximum NDWI image (MNI); and a binary water image (BWI).

[0073] Step 300, based on the maximum normalized difference water index image, gray-scale morphological erosion and Canny edge detection, the binary water image is iteratively segmented multiple times to obtain multiple initial water objects. That is, an iterative segmentation algorithm combining gray-scale morphological and edge detection is proposed, which realizes fine segmentation of water bodies and extracts potential aquaculture ponds through segmentation degree detection.

[0074] Step 300 specifically includes:

[0075] 1) Set the total number of iterations n and the preset initial value i = 0.

[0076] 2) Start the i-th iteration segmentation. Specifically as follows:

[0077] 21) The maximum normalized difference water index image is sequentially subjected to gray-scale morphological erosion processing and Canny edge detection processing to obtain a Canny edge image (CEI).

[0078] 22) Determine whether the current iteration number value is the preset initial value; that is, determine whether i is equal to 0.

[0079] 23) If the current iteration value is the preset initial value, i.e. i equals 0, the binary water body image is segmented using the Canny edge image to obtain a plurality of first water body objects. Specifically, the Canny edge image is superimposed with the binary water body image to achieve image segmentation of the BWI.

[0080] 24) If the current iteration value is not the preset initial value, i.e. i is not equal to 0, the Canny edge image is superimposed with the Canny edge image in the previous iteration segmentation to obtain an updated Canny edge image; then the binary water body image is segmented using the updated Canny edge image to obtain a plurality of first water body objects.

[0081] 25) Any of the first water body objects is subjected to connectivity detection to obtain a detection result; specifically as follows:

[0082] A) The landscape shape index and the ratio of the convex hull perimeter of the first water body object are calculated; the ratio of the convex hull perimeter is the ratio of the perimeter of the first water body object to the convex hull perimeter.

[0083] The calculation formula of the landscape shape index (LSI) is as follows:

[0084]

[0085] wherein LSI represents the landscape shape index, P object represents the perimeter of the first water body object, A object represents the area of the first water body object.

[0086] The calculation formula of the ratio of the convex hull perimeter (RPOC) is as follows:

[0087]

[0088] wherein RPOC represents the ratio of the convex hull perimeter, P convex hull represents the perimeter of the convex hull corresponding to the first water body object.

[0089] B) It is judged whether the landscape shape index satisfies a first preset threshold range and whether the ratio of the convex hull perimeter satisfies a second preset threshold range; specifically, the first preset threshold range is that the landscape shape index is less than or equal to 2.5; the second preset threshold range is that the ratio of the convex hull perimeter is less than or equal to 1.5. That is, LSI of the water body object ≤ 2.5 and RPOC ≤ 1.5.

[0090] C) if the landscape shape index meets the first preset threshold range and the convex hull perimeter ratio meets the second preset threshold range, the detection result of the connectivity detection is that the first water body object passes the connectivity detection; if the landscape shape index does not meet the first preset threshold range or the convex hull perimeter ratio does not meet the second preset threshold range, the detection result of the connectivity detection is that the first water body object does not pass the connectivity detection.

[0091] 26) if the detection result indicates that the first water body object passes the connectivity detection, the first water body object is marked as an initial water body object, and the iteration is stopped; if the detection result indicates that the first water body object does not pass the connectivity detection, the next iteration segmentation is entered. In other words, if the first water body object passes the segmentation degree detection, the water body object is extracted from the BWI as a possible single-object aquaculture pond (SOAP), which is a water body object vector, and no longer participates in the iteration, otherwise, the next round of iteration is continued.

[0092] 3) determining whether the iteration number i is less than the total iteration number n; if yes, a new round of iteration is continued; if no, the iteration is ended.

[0093] Step 400, constructing a decision tree based on the prior knowledge of the target aquaculture area. Specifically, the prior knowledge of the aquaculture pond includes spectral features, shape features and spatial distribution features of the aquaculture pond, and the decision tree is constructed based on the above three aspects, as shown in FIG. 4. Figure 2

[0094] Step 500, screening the plurality of initial water body objects based on the decision tree to obtain the single aquaculture pond corresponding to the target aquaculture area.

[0095] Step 500, specifically comprising:

[0096] 1) for any initial water body object, acquiring the area data of the initial water body object, the median of the corresponding NDWI time series data, the number of adjacent water body objects and the type of the land block where the initial water body object is located; the type of the land block includes farmland, building area and water body breeding area.

[0097] 2) determining whether the area data of the initial water body object is greater than a first set value to obtain a first result; preferably, the first set value is 52000 square meters.

[0098] 3) if the first result indicates no, the initial water body object is discarded; if the first result indicates yes, determining whether the median of the NDWI time series data corresponding to the initial water body object is greater than or equal to a second set value to obtain a second result; preferably, the second set value is 0.2.​

[0099] 4) If the second result represents no, the initial water body object is discarded; if the second result represents yes, it is determined whether the number of adjacent water body objects of the initial water body object is greater than or equal to a third set value, to obtain a third result; preferably, the third set value is 5, wherein "adjacent" refers to a range with the initial water body object as the center and a radius of 100 meters.

[0100] 5) If the third result represents no, the initial water body object is discarded; if the third result represents yes, it is determined whether the type of the land parcel where the initial water body object is located is a water aquaculture area, to obtain a fourth result; preferably, the type of the land parcel where the initial water body object is located is determined according to a LULC image.

[0101] 6) If the fourth result represents no, the initial water body object is discarded; if the fourth result represents yes, the initial water body object is determined as a single water aquaculture pond corresponding to the target water aquaculture area.

[0102] In order to test the extraction accuracy of the water aquaculture pond of the present application, the following test method is designed, including sampling design, response design and analysis. Taking the 0.5-meter high-resolution image provided by Maxar as the base map, 742 water aquaculture pond samples are randomly extracted by manual digitization, and the area of each sample is within 0-20000 square meters. All the samples are extracted by professional personnel with image interpretation experience, and the artificial sample data set most consistent with the ground truth is obtained by majority voting. The data set extracted manually from the 0.5-meter high-resolution image is closest to the ground truth, while the present embodiment uses 10-meter resolution Sentinel-2 data, and the extraction accuracy compared with the artificial sample data set has a certain gap.

[0103] The GEE platform provides three image segmentation algorithms, K-Means, G-Means and Simple Non-Iterative Clustering (SNIC), and the NDWI image is taken as the input data. The parameters of the three algorithms are optimized to ensure that the output results are of the highest accuracy. Then the extraction results of the above three algorithms are compared with the extraction results of the water aquaculture pond of the present application, and it can be concluded that:

[0104] The extraction result of the present application is closest to the true shape of the water aquaculture pond, and most of the ponds are segmented into independent objects; in the K-Means method, there are a large number of noise points in the pond; in the G-Means method, some pond objects are incorrectly classified as a class with other ground objects; the SNIC method correctly distinguishes the water aquaculture water body from the non-water aquaculture water body, but in the water aquaculture water body, many ponds are divided into the same object.

[0105] Further, the artificial sample data set is spatially connected with the extraction result of the aquaculture pond to obtain an extraction sample corresponding to each artificial sample. The areas of the artificial sample and the corresponding extraction sample are calculated and compared to verify whether the extraction result is close to the true shape of the aquaculture pond. Five indexes, a coefficient of determination R 2 , a root mean square error RMSE, a mean absolute error MAE, a mean absolute percentage error MAPE and an accuracy Accuracy, are introduced to test the extraction effect of each method (the present application, K-Means, G-Means and SNIC) on the aquaculture pond, and the results are shown in Figure 3 、 Figure 4 、 Figure 5 and Figure 6 .

[0106] (1) coefficient of determination (R 2 )

[0107] The coefficient of determination is introduced to test the fitting degree of the extraction result and the artificial sample in the area, and the formula is as follows:

[0108]

[0109] wherein R 2 is the coefficient of determination between the segmentation result and the artificial sample, n is the number of mutual matching of the extraction result and the artificial sample; for the i th spatially matched aquaculture pond extraction result and the artificial extraction result, x i is the area of the aquaculture pond extraction result, y i is the area of the artificial extraction result, is the average of the areas of all extraction results, is the average of the areas of all artificial samples.

[0110] The results show that the R 2 of the present application is the highest, which is 0.5007, much higher than that of other methods. For samples of different areas, the segmentation result of the present application is as close to the true value as possible, and the error distribution is also relatively uniform, which is the reason for the high R 2 . The R 2 of the SNIC method is the lowest, which is 0.04. SNIC obtains finer polygons by setting a closer seed point distance, so that more aquaculture ponds are segmented correctly; but for larger ponds, SNIC still segments them into small polygons, which is the reason for the low R 2 of SNIC.

[0111] (2) Root Mean Squared Error (RMSE)

[0112] The RMSE is introduced to compare the extraction accuracy of different methods as a whole, and the formula is:

[0113]

[0114] The results show that the RMSE of the application is the lowest, which is 2427.19, and the RMSE of the segmentation method of GEE is more than 3000.

[0115] (3) Mean Absolute Error (MAE)

[0116] In order to improve the robustness of the detection result to abnormal values, MAE is introduced for testing, and the formula is:

[0117]

[0118] The results show that the MAE of the application is the lowest, which is 1222.02, far lower than 1777.10 of SNIC which is the second. The MAE of K-Means is the highest, which is 2321.93, close to twice of the application. This shows that even if the influence of abnormal values of extraction error is controlled, the extraction effect of GEE method is still not as good as the method.

[0119] (4) Mean Absolute Percentage Error (MAPE)

[0120] Since the area of aquaculture ponds is different, in order to eliminate the influence of different sample data levels on the results, the relative error of pond area is calculated for the spatial matching artificial sample and the extraction sample, and MAPE is introduced, and the formula is:

[0121]

[0122] The results show that the MAPE of the application is 30.36%, that is, on average, the extraction result of the application has an area error of 30.36% with the true value. The MAPE of SNIC is slightly higher than that of the method, because SNIC divides the image into small patches, which is exactly the area characteristics of the aquaculture ponds in the study area, but the cost is that SNIC cannot well segment larger aquaculture ponds. The MAPE value of G-Means is the lowest, which is 60.67%, which is likely to be because the segmentation result of G-Means has a large number of connected aquaculture ponds.

[0123] (5) Accuracy

[0124] For each pair of spatially matched artificial samples and extraction samples, the extraction sample with a relative error less than or equal to 0.2 is defined as the correct extraction sample to measure the accuracy of the method, and the formula is as follows

[0125]

[0126] Wherein, Accuracy is the accuracy of the extraction result, TP is the sample with a relative error less than or equal to 0.2, and FP is the sample with a relative error greater than 0.2. This is a very strict standard, especially considering that the artificial samples are extracted from 0.5-meter resolution images, and the data source of the extraction result is 10-meter images. The results show that the accuracy of the present application is 65.63%, which is significantly improved compared with 56.8% of SNIC. The accuracy of K-Means is 24.2%, and the accuracy of G-Means is 27.42%, which are relatively close.

[0127] After a series of tests and comparisons with other methods, it is found that the present application can obtain relatively accurate aquaculture pond extraction results under limited resolution through a refined segmentation process. Compared with the general method provided on GEE, the extraction result is significantly improved.

[0128] In summary, the present application proposes a single aquaculture pond extraction method, which extracts the pond as an independent object from the image to realize fine mapping of aquaculture ponds. Relying on the Google Earth Engine (GEE) remote sensing data processing cloud platform, the present application realizes large-scale extraction of aquaculture ponds. The present application can be further implemented on the open-source python interactive interface of Google Colab, and the extraction result is compared with the extraction result of the existing segmentation method of GEE to test the extraction accuracy of the present application. The results show that the average absolute error of the extraction result of the present application is 1222.02m 2 , which is 555.08m 2 higher than other methods 2 ; the extraction accuracy is 65.63%, which is 8.83%-41.43% higher than other methods.

[0129] Example 2

[0130] As Figure 7 shown, in order to realize the technical solutions in Example 1, the present embodiment provides a single aquaculture pond extraction system, which comprises:

[0131] An image acquisition module 101 is configured to acquire a Sentinel-2 time series image set of a target aquaculture area.

[0132] The image processing module 201 is configured to determine a maximum normalized difference water body index image and a binary water body image corresponding to the target aquaculture area according to the Sentinel-2 time series image set.

[0133] The image segmentation module 301 is configured to perform multiple iterations of segmentation on the binary water body image based on the maximum normalized difference water body index image, gray-scale morphological erosion, and Canny edge detection, to obtain a plurality of initial water body objects.

[0134] The decision tree construction module 401 is configured to construct a decision tree based on prior knowledge of the target aquaculture area.

[0135] The single body screening module 501 is configured to screen the plurality of initial water body objects based on the decision tree, to obtain a single water body aquaculture pond corresponding to the target aquaculture area.

[0136] The embodiments in the specification are described in a progressive manner, and each embodiment focuses on the difference from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method part.

[0137] The principles and implementation manners of the present application are described by using specific examples in the present application. The above description of the embodiments is only used to help understand the method of the present application and its core idea; meanwhile, for those skilled in the art, according to the idea of the present application, the specific implementation manner and application range can be changed. In conclusion, the content of the specification should not be understood as a limitation of the present application.

Claims

1. A method for extracting from a single aquaculture pond, characterized in that the method... include: Obtain the Sentinel-2 time-series image set of the target aquaculture area; Based on the Sentinel-2 time series image set, determine the maximum normalized difference water body index image and binarized water body image corresponding to the target aquaculture area; Based on the maximum normalized difference water index image, grayscale morphological erosion, and Canny edge detection, the binarized water image is iteratively segmented multiple times to obtain multiple initial water objects. Based on the maximum normalized difference water index image, grayscale morphological erosion, and Canny edge detection, when performing multiple iterative segmentations on the binarized water image, each iterative segmentation specifically includes: The image of the maximum normalized difference water index is subjected to grayscale morphological erosion processing and Canny edge detection processing in sequence to obtain a Canny edge image; it is then determined whether the current iteration number is a preset initial value. If the current iteration number is a preset initial value, the binarized water image is segmented using the Canny edge image to obtain multiple first water objects; if the current iteration number is not a preset initial value, the Canny edge image is overlaid with the Canny edge image from the previous iteration to obtain an updated Canny edge image; then the updated Canny edge image is used to segment the binarized water image to obtain multiple first water objects. Connectivity detection is performed on any of the first water bodies to obtain the detection results; If the detection result indicates that the first water body object passes the connectivity detection, then the first water body object is marked as the initial water body object, and the iteration stops; if the detection result indicates that the first water body object fails the connectivity detection, then proceed to the next iteration segmentation. Connectivity detection is performed on any of the first water bodies to obtain a detection result. Specifically, this includes: calculating the landscape shape index and convex hull perimeter ratio of the first water body; the convex hull perimeter ratio is the ratio of the perimeter of the first water body to the perimeter of the convex hull; determining whether the landscape shape index meets a first preset threshold range and whether the convex hull perimeter ratio meets a second preset threshold range; if the landscape shape index meets the first preset threshold range and the convex hull perimeter ratio meets the second preset threshold range, then the connectivity detection result indicates that the first water body passes the connectivity detection; if the landscape shape index does not meet the first preset threshold range or the convex hull perimeter ratio does not meet the second preset threshold range, then the connectivity detection result indicates that the first water body fails the connectivity detection. Based on prior knowledge of the target aquaculture area, a decision tree is constructed. The decision tree is used to filter the multiple initial water bodies to obtain the individual aquaculture ponds corresponding to the target aquaculture area.

2. The method for extracting from a single aquaculture pond according to claim 1, characterized in that, Based on the Sentinel-2 time-series image set, determine the maximum normalized difference water body index image and binarized water body image corresponding to the target aquaculture area, specifically including: Calculate the normalized differential water index for each Sentinel-2 image in the Sentinel-2 time series image set; the normalized differential water indices of multiple Sentinel-2 images constitute NDWI time series data. The NDWI time series data are denoised according to the Raida criterion. Based on the maximum value synthesis method, the NDWI time series data after noise reduction are merged to obtain the maximum normalized difference water index image. The image of the maximum normalized difference water index was segmented using a threshold segmentation method to obtain a binarized water image.

3. The extraction method for a single aquaculture pond according to claim 2, characterized in that, The formula for calculating the normalized differential water index of the Sentinel-2 image is as follows: in, This represents the normalized differential water index of Sentinel-2 imagery. This represents the green band data corresponding to the Sentinel-2 image. This represents the near-infrared band data corresponding to the Sentinel-2 image.

4. The extraction method for a single aquaculture pond according to claim 1, characterized in that, The formula for calculating the landscape shape index of the first water body object is as follows: LSI stands for Landscape Shape Index. This represents the perimeter of the first body of water. Indicates the area of ​​the first body of water; The formula for calculating the ratio of the convex hull perimeter of the first water body object is as follows: Wherein, RPOC represents the ratio of the convex hull perimeter. This represents the perimeter of the convex hull corresponding to the first water body object.

5. The extraction method for a single aquaculture pond according to claim 1, characterized in that, The first preset threshold range is a landscape shape index of less than or equal to 2.5; the second preset threshold range is a convex hull perimeter ratio of less than or equal to 1.

5.

6. The method for extracting from a single aquaculture pond according to claim 2, characterized in that, The decision tree is used to filter the multiple initial water bodies to obtain the individual aquaculture ponds corresponding to the target aquaculture area, specifically including: For any initial water body object, obtain the area data of the initial water body object, the median of the corresponding NDWI time series data, the number of neighboring water bodies, and the type of land parcel it is located in; the land parcel type includes farmland, building area, and aquaculture area; Determine whether the area data of the initial water body object is greater than a first preset value to obtain a first result; If the first result indicates no, then the initial water body object is discarded; If the first result indicates yes, then determine whether the median of the NDWI time series data corresponding to the initial water body object is greater than or equal to the second set value, and obtain the second result; If the second result indicates no, then the initial water body object is discarded; If the second result indicates yes, then determine whether the number of neighboring water objects of the initial water object is greater than or equal to a third preset value, and obtain the third result; If the third result indicates no, then the initial water body object is discarded; If the third result indicates yes, then determine whether the land parcel type where the initial water body object is located is an aquaculture area, and obtain the fourth result; If the fourth result indicates no, then the initial water body object is discarded; If the fourth result indicates yes, then the initial water body object is determined as the single aquaculture pond corresponding to the target aquaculture area.

7. A single-unit aquaculture pond extraction system, characterized in that, The system includes: The image acquisition module is used to acquire a set of Sentinel-2 time-series images of the target aquaculture area; The image processing module is used to determine the maximum normalized difference water body index image and the binarized water body image corresponding to the target aquaculture area based on the Sentinel-2 time series image set. The image segmentation module is used to perform multiple iterative segmentations on the binarized water body image based on the maximum normalized difference water body index image, grayscale morphological erosion, and Canny edge detection to obtain multiple initial water body objects. Based on the maximum normalized difference water index image, grayscale morphological erosion, and Canny edge detection, when performing multiple iterative segmentations on the binarized water image, each iterative segmentation specifically includes: The image of the maximum normalized difference water index is subjected to grayscale morphological erosion processing and Canny edge detection processing in sequence to obtain a Canny edge image; it is then determined whether the current iteration number is a preset initial value. If the current iteration number is a preset initial value, the binarized water image is segmented using the Canny edge image to obtain multiple first water objects; if the current iteration number is not a preset initial value, the Canny edge image is overlaid with the Canny edge image from the previous iteration to obtain an updated Canny edge image; then the updated Canny edge image is used to segment the binarized water image to obtain multiple first water objects. Connectivity detection is performed on any of the first water bodies to obtain the detection results; If the detection result indicates that the first water body object passes the connectivity detection, then the first water body object is marked as the initial water body object, and the iteration stops; if the detection result indicates that the first water body object fails the connectivity detection, then proceed to the next iteration segmentation. Connectivity detection is performed on any of the first water bodies to obtain a detection result. Specifically, this includes: calculating the landscape shape index and convex hull perimeter ratio of the first water body; the convex hull perimeter ratio is the ratio of the perimeter of the first water body to the perimeter of the convex hull; determining whether the landscape shape index meets a first preset threshold range and whether the convex hull perimeter ratio meets a second preset threshold range; if the landscape shape index meets the first preset threshold range and the convex hull perimeter ratio meets the second preset threshold range, then the connectivity detection result indicates that the first water body passes the connectivity detection; if the landscape shape index does not meet the first preset threshold range or the convex hull perimeter ratio does not meet the second preset threshold range, then the connectivity detection result indicates that the first water body fails the connectivity detection. The decision tree construction module is used to construct a decision tree based on prior knowledge of the target aquaculture area. The individual screening module is used to screen the multiple initial water bodies based on the decision tree to obtain the individual aquaculture ponds corresponding to the target aquaculture area.

Images