A method for automatic target recognition based on SAR time-series remote sensing images

Abstract

The present invention belongs to the field of satellite remote sensing technology, and provides a method for automatic target recognition based on SAR time-series remote sensing images. The method includes: obtaining a time series data set of a research area within a time period; calculating the median of the phase dimension of each pixel in the data set to obtain a median image; calculating the median water body index MWI based on the median image, and obtaining a water body classification image based on the pixel; constructing objects using a connected domain labeling algorithm to obtain an object-based water body classification image; calculating the shape features of each object in the object-based water body classification image; pre-extracting each breeding pond object based on a random forest according to the median image and the shape features; calculating the boundary direction change index of each breeding pond object, filtering unqualified objects, and obtaining the final breeding pond object data. Compared with traditional methods, this automatic target recognition method reduces manual intervention and improves the degree of automation and recognition accuracy.

Description

Technical Field

[0001] The invention relates to the field of satellite remote sensing technology, and in particular to an automatic target recognition method based on SAR time-series remote sensing images. Background Art

[0002] The rapid growth of population has greatly increased the demand for aquaculture products, so it is crucial to strengthen the sustainability and scientific management of aquaculture. The premise of scientific management of aquaculture is to accurately grasp the number, area and spatial distribution characteristics of aquaculture ponds. Traditional field surveys can accurately identify and measure aquaculture ponds, but this consumes a lot of manpower and material resources. The identification of traditional aquaculture ponds mostly relies on manual ground statistics, which not only requires a lot of manpower and material resources, but also has difficulty in ensuring accuracy and timeliness.

[0003] With the development of remote sensing technology, the application of high spatial resolution remote sensing data to identify aquaculture ponds has attracted the interest of many scholars. Remote sensing technology can achieve large-scale periodic observations and is an effective means to accurately and quickly monitor aquaculture ponds. At present, water body identification mainly relies on optical remote sensing images, and a variety of water body indexes have been created to quickly and accurately obtain information such as the spatial distribution of water bodies and their changes. However, optical images are easily affected by cloud cover and other factors, and many invalid observations often occur, which is particularly prominent in coastal areas suitable for aquaculture. For the identification of aquaculture ponds, due to the short aquaculture cycle (usually a few weeks to a few months), if there are continuous invalid observations, it is bound to have a greater impact on the recognition accuracy.

[0004] The synthetic aperture radar (SAR) sensor carried by the Sentinel-1 satellite can penetrate clouds and provide surface observation data with high temporal and spatial resolution. In contrast, radar data with better cloud penetration ability is more suitable for monitoring aquaculture ponds. Aquaculture ponds are shallow water surfaces surrounded by dams. Based on the high contrast between smooth water surfaces (low radar backscatter) and rough non-water surfaces (high radar backscatter), SAR images can be used to monitor them. However, on the one hand, there is a lack of automated recognition methods, which requires more manual intervention. On the other hand, salt pans, abandoned fish ponds, and tidal wetlands have greatly reduced the accuracy of aquaculture pond identification. Summary of the invention

[0005] In view of this, an embodiment of the present invention provides an automatic target recognition method based on SAR time-series remote sensing images, which is used to solve the problem that the identification of aquaculture ponds requires a lot of manual intervention and is difficult to distinguish from other natural water bodies, significantly improves the recognition accuracy of aquaculture ponds, and enhances the degree of automation and practicality.

[0006] In a first aspect, an embodiment of the present invention provides a method for automatic target recognition based on SAR time-series remote sensing images, comprising the following steps:

[0007] S1: Obtain all Sentinel-1 data of the area to be studied within a time period and sort them in chronological order to form a time series dataset;

[0008] S2: Calculate the median of the phase dimension of each pixel in the time series data set to obtain a median image;

[0009] S3: Calculate and obtain a median water index MWI based on the median image, and extract surface water data based on the median water index MWI to obtain a pixel-based water classification image;

[0010] S4: constructing an object using a connected domain labeling algorithm for the pixel-based water body classification image to obtain an object-based water body classification image;

[0011] S5: Calculating the shape feature of each object in the object-based water body classification image;

[0012] S6: Based on the object-based water body classification image, pre-extract each aquaculture pond object based on the median image and the shape feature based on random forest;

[0013] S7: Calculate the boundary direction change index of each of the aquaculture pond objects, filter out unqualified aquaculture pond objects according to the boundary direction change index, and obtain final data of the aquaculture pond objects.

[0014] According to a specific implementation method of an embodiment of the present invention, in step S1, the time series data set needs to be preprocessed, including orbit correction, elimination of GRD boundary noise, thermal noise removal, radiation calibration, filtering, terrain correction and decibel processing, to finally obtain the backscattering coefficient.

[0015] According to a specific implementation of an embodiment of the present invention, step S2 is specifically: setting the size of the time series data set to m rows × n columns × t dimensions, calculating the median of the phase dimension pixel by pixel, and the median Z of the phase dimension is calculated as follows:

[0016] Z=median(i1,i2,…,i t );

[0017] Wherein, i is a pixel, and finally the median image with a size of m rows×n columns is obtained.

[0018] According to a specific implementation of an embodiment of the present invention, in step S3, the median water index MWI is calculated as follows:

[0019] MWI=a·lnZ-b+k;

[0020] Wherein, a is the median multiplied by a constant, b is the median plus a constant, k is a harmonic parameter, and Z is the median of the time dimension of each pixel in the median image. MWI>0 is judged as a water body, and MWI<0 is judged as a non-water body.

[0021] According to a specific implementation of the embodiment of the present invention, in step S3, a is 2, b is 6, and k is 0.1. The automatic recognition effect is more accurate.

[0022] According to a specific implementation of the embodiment of the present invention, in step S4, the connected domain labeling algorithm is used to construct the object, and the steps are as follows:

[0023] 4.1: Traverse each pixel in the water body classification image from left to right and from top to bottom;

[0024] 4.2: If the pixel has been traversed, continue traversing, otherwise go to the next step;

[0025] 4.3: If the pixel has not been traversed, it is used as the starting seed and put into a set marked as label, and then all the pixels that meet the conditions are pushed into the stack according to the specified kernel neighborhood relationship;

[0026] 4.4: Pop the top pixel of the stack, put the pixel into the set marked with label, and push all the pixels that meet the conditions into the stack again according to the specified kernel neighborhood relationship;

[0027] 4.5: Repeat the previous step until the stack is empty, thereby constructing a connected region in the water body classification image, that is, the object, and the label of the object is label;

[0028] 4.6: Repeat step 4.2 until the entire water body classification image is traversed, and finally the pixel-based water body classification image is transformed into the object-based water body classification image, wherein the connectivity is determined by a given kernel, and the object size can be specified, and objects exceeding a certain value are regarded as background.

[0029] According to a specific implementation of the embodiment of the present invention, in step S5, the shape features include perimeter, area, aspect ratio and compactness.

[0030] According to a specific implementation method of an embodiment of the present invention, in step S6, based on the water body classification image, the median image and the four indicators of perimeter, area, aspect ratio and compactness are input into a random forest classifier together, and training samples are selected for large open water areas and the breeding ponds respectively to perform supervised classification, so as to realize pre-extraction of each breeding pond.

[0031] According to a specific implementation of the embodiment of the present invention, in step S7, the boundary direction change index calculation method is as follows:

[0032] 7.1: Determine the boundary pixels of the object;

[0033] 7.2: Construct convolution kernels in four directions:

[0034] 7.3: Perform convolution operations on the boundary pixels of each object using the four convolution kernels, and add the four convolution results. If the boundary direction changes, the sum of the four convolutions is 1, otherwise it is 0;

[0035] 7.4: For each object, the sum of the convolution results of all the boundary pixels corresponding to the object is calculated as the final boundary direction change index of the object. Based on the calculated boundary direction change index, the aquaculture ponds pre-extracted in step S6 are filtered, and the objects with boundary direction change index higher than a specific threshold are filtered out, and the objects with boundary direction change index lower than the specific threshold are retained as the final extracted aquaculture pond results.

[0036] In a second aspect, an embodiment of the present invention provides a target automatic recognition device based on SAR time-series remote sensing images, comprising:

[0037] The acquisition module is used to obtain all Sentinel-1 data of the area to be studied within a time period and sort them in chronological order to form a time series data set;

[0038] A median module is used to calculate the median of the phase dimension of each pixel in the time series data set to obtain a median image;

[0039] A classification module, used to calculate and obtain a median water index MWI based on the median image, and to extract surface water data based on the median water index MWI to obtain a water classification image based on pixels;

[0040] A construction module is used to construct an object using a connected domain labeling algorithm for the pixel-based water body classification image to obtain an object-based water body classification image;

[0041] A calculation module, used for calculating the shape feature of each object in the object-based water body classification image;

[0042] An extraction module, for pre-extracting each aquaculture pond object based on the object-based water body classification image according to the median image and the shape feature based on a random forest;

[0043] The screening module is used to calculate the boundary direction change index of each of the breeding pond objects, filter out unqualified breeding pond objects according to the boundary direction change index, and obtain the final data of the breeding pond objects.

[0044] In a third aspect, an embodiment of the present invention further provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the automatic target identification method in the aforementioned first aspect or any implementation method of the first aspect are implemented.

[0045] The method, device and electronic device for automatic target recognition based on SAR time-series remote sensing images provided by the embodiments of the present invention have at least the following technical effects:

[0046] First, this target automatic recognition method can automatically identify water bodies based on Sentinel-1 time series images through the median water index, which improves the performance of radar image recognition.

[0047] Second, this automatic target recognition method improves the recognition accuracy of aquaculture ponds through the boundary direction change index, which can better reflect the regularity of the boundary shape of the object in the image.

[0048] Third, compared with traditional methods, this automatic target recognition method reduces human intervention and improves the degree of automation and recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following is a brief introduction to the drawings required for the specific embodiments or the description of the prior art. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn according to the actual scale.

[0050] Figure 1 A flow chart of a method for automatic target recognition based on SAR time-series remote sensing images provided by an embodiment of the present invention is shown;

[0051] Figure 2 The structure block diagram of a target automatic recognition device based on SAR time-series remote sensing images provided by an embodiment of the present invention is shown;

[0052] Figure 3 A schematic structural diagram of an electronic device provided by an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0053] The following embodiments of the technical solution of the present invention are described in detail in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and are therefore only used as examples, and cannot be used to limit the protection scope of the present invention.

[0054] It should be noted that, unless otherwise specified, the technical terms or scientific terms used in this application should have the common meanings understood by those skilled in the art to which the present invention belongs.

[0055] Figure 1 A flowchart of a method for automatic target recognition based on SAR time-series remote sensing images provided by an embodiment of the present invention is shown in FIG. Figure 1 , the method comprises the following steps:

[0056] S1: Obtain all Sentinel-1 data of the area to be studied within a time period and sort them in chronological order to form a time series dataset.

[0057] Specifically, all Sentinel-1 data (including Sentinel-1A and 1B) of the area to be studied within a time period (such as a natural year) are obtained, and the ground range detected (GRD) product in the Interferometric Wide Swath (IW) mode is selected, and the VV single polarization images are sorted in chronological order to form a time series data set. Conventional preprocessing is performed on the data set, including orbit correction, elimination of GRD boundary noise, thermal noise removal, radiation calibration, filtering, terrain correction and decibelization, and finally the backscattering coefficient is obtained.

[0058] S2: Calculate the median of the phase dimension of each pixel in the time series dataset to obtain the median image.

[0059] Specifically, the size of the Sentinel-1 time series data set is m rows × n columns × t dimensions (phase dimension). The median of the phase dimension is calculated for each pixel. Taking pixel i as an example, the median Z of the phase dimension is calculated as follows:

[0060] Z=median(i1,i2,…,i t );

[0061] Finally, a median image with a size of m rows × n columns is obtained.

[0062] S3: The median water index MWI is calculated based on the median image, and the surface water data is extracted based on the median water index MWI to obtain a water body classification image.

[0063] This application proposes a new water index, called the Median Water Index (MWI). The median image obtained in step S2 is used to more conveniently and automatically accurately extract surface water bodies based on the median water index MWI. The calculation method of the median water index MWI is as follows:

[0064] MWI=a·lnZ-b+k;

[0065] Among them, a is the median multiplied by a constant, b is the median plus a constant, k is the harmonic parameter, and Z is the median of the time dimension of each pixel in the median image. After experimental calculation and analysis, a is set to 2, b is set to 6, and k is set to 0.1 to achieve the best effect. If MWI>0, it is judged as a water body, and MWI<0 is judged as a non-water body.

[0066] S4: For the water body classification image, the connected domain labeling algorithm is used to construct the object to obtain the object-based water body classification image.

[0067] For the extracted surface water bodies, i.e., the pixel-based water body classification image, the connected domain labeling algorithm (ConnectedComponents) is used to construct objects, that is, to find connected components with the same value as the input band and label them with unique values. Connectivity is determined by a given kernel, and the object size can be specified. Objects exceeding a certain value are considered background. The specific steps are as follows:

[0068] 4.1: Traverse each pixel in the water classification image from left to right and from top to bottom;

[0069] 4.2: If the pixel has been traversed, continue traversing, otherwise go to the next step;

[0070] 4.3: If the pixel has not been traversed, it is used as the starting seed and put into a set marked with label, and then all pixels that meet the conditions are pushed into the stack according to the specified kernel neighborhood relationship;

[0071] 4.4: Pop the top pixel of the stack, put the pixel into the set marked with label, and push all the pixels that meet the conditions into the stack again according to the specified kernel neighborhood relationship;

[0072] 4.5: Repeat the previous step until the stack is empty. Thus, a connected region in the water body classification image is constructed, that is, an object, and the label of the object is label;

[0073] 4.6: Repeat step 4.2 until the entire water body classification image is traversed, and finally the pixel-based water body classification image is converted into an object-based water body classification image.

[0074] S5: Calculate the shape features of each object in the object-based water classification image, the shape features including perimeter, area, aspect ratio, and compactness.

[0075] For each water object, count its perimeter, area, aspect ratio and compactness. The calculation method is as follows:

[0076] 5.1: The perimeter can be obtained by finding the boundary pixels and then counting the number of boundary pixels;

[0077] 5.2: The area is calculated by counting the number of pixels in each water body object and multiplying it by the actual area corresponding to each pixel;

[0078] 5.3:

[0079] 5.4:

[0080] S6: Based on the water body classification image, the pre-extraction of each aquaculture pond object is achieved based on the random forest according to the median image and shape features.

[0081] Based on the acquired water body classification images, the median image and four indicators, namely perimeter, area, aspect ratio and compactness, are input into the random forest classifier. A small number of training samples are selected for large open water areas and aquaculture ponds respectively to perform supervised classification and realize the pre-extraction of aquaculture ponds.

[0082] S7: Calculate the boundary direction change index of each aquaculture pond object, filter out unqualified aquaculture pond objects according to the boundary direction change index, and obtain the final aquaculture pond object data.

[0083] The pre-extraction of aquaculture ponds has been achieved through step S6, but there are still some smaller water surfaces (non-aquaculture ponds) included in the pre-extraction results. Since the boundaries of aquaculture ponds are mostly relatively regular quadrilaterals, while the boundaries of non-aquaculture ponds are often irregular curved curves, by constructing the boundary direction change index, non-aquaculture ponds can be effectively filtered out, thereby significantly improving the recognition accuracy of aquaculture ponds. The boundary direction change index is achieved by calculating the convolution kernel from four directions, as follows:

[0084] 7.1: Determine the boundary pixels of the object;

[0085] 7.2: Construct the convolution kernel. Taking the 3×3 convolution kernel as an example, construct the convolution kernels in four directions as shown in Table 1:

[0086]

[0087] 7.3: Perform convolution operations on the boundary pixels of each object using the four convolution kernels in step 7.2, and add the results of the four convolutions. If the boundary direction changes, the sum of the four convolutions is 1, otherwise it is 0.

[0088] 7.4: For each object, calculate the sum of the convolution results of all boundary pixels as the final boundary direction change index of the object.

[0089] Based on the calculated boundary direction change index, the aquaculture ponds pre-extracted in step S6 are filtered, objects with boundary direction change index higher than a specific threshold are filtered out, and objects with boundary direction change index lower than the specific threshold are retained as the final extracted aquaculture pond results.

[0090] It should be noted that the modules are arranged in a stream layout, which is only one embodiment of the present invention. Other arrangements may also be adopted, and the present invention is not limited thereto.

[0091] The embodiments of the present invention have the following technical effects:

[0092] First, this target automatic recognition method can automatically identify water bodies based on Sentinel-1 time series images through the median water index, which improves the performance of radar image recognition.

[0093] Second, this automatic target recognition method improves the recognition accuracy of aquaculture ponds through the boundary direction change index, which can better reflect the regularity of the boundary shape of the object in the image.

[0094] Third, compared with traditional methods, this automatic target recognition method reduces human intervention and improves recognition accuracy.

[0095] Figure 2 The present invention provides a block diagram of a target automatic recognition device based on SAR time-series remote sensing images, the device comprising:

[0096] The acquisition module is used to obtain all Sentinel-1 data of the area to be studied within a time period and sort them in chronological order to form a time series data set;

[0097] The median module is used to calculate the median of the phase dimension of each pixel in the time series data set to obtain the median image;

[0098] The classification module is used to calculate the median water index MWI based on the median image, extract the surface water data based on the median water index MWI, and obtain a water classification image based on pixels;

[0099] A construction module is used to construct objects for pixel-based water body classification images by using a connected domain labeling algorithm to obtain object-based water body classification images;

[0100] A calculation module for calculating the shape feature of each object in the object-based water body classification image;

[0101] An extraction module is used to pre-extract each aquaculture pond object based on the random forest according to the median image and shape features based on the object-based water body classification image;

[0102] The screening module is used to calculate the boundary direction change index of each aquaculture pond object, filter out unqualified aquaculture pond objects according to the boundary direction change index, and obtain the final aquaculture pond object data.

[0103] Figure 2 The functions of each module in the embodiment correspond to the contents in the corresponding method embodiment and will not be repeated here.

[0104] Figure 3 The structural schematic diagram of the electronic device 30 provided by the embodiment of the present invention is shown, and the electronic device 30 includes at least one processor 301 (such as a CPU), at least one input and output interface 304, a memory 302, and at least one communication bus 303, which is used to realize the connection and communication between these components. At least one processor 301 is used to execute the computer instructions stored in the memory 302, so that the at least one processor 301 can execute any of the aforementioned embodiments of the automatic target recognition method. The memory 302 is a non-transitory memory, which may include a volatile memory, such as a high-speed random access memory (RAM: Random Access Memory), and may also include a non-volatile memory (non-volatile memory), such as at least one disk storage. The communication connection with at least one other device or unit is realized through at least one input and output interface 304 (which may be a wired or wireless communication interface).

[0105] In some implementations, the memory 302 stores a program 3021, and the processor 301 executes the program 3021 to execute the content of any of the aforementioned table partitioning method embodiments.

[0106] The electronic device may exist in a variety of forms, including but not limited to:

[0107] (1) Mobile communication equipment: This type of equipment is characterized by having mobile communication functions and its main purpose is to provide voice and data communications. This type of terminal includes: smart phones, multimedia phones, functional phones, and low-end phones.

[0108] (2) Ultra-mobile personal computer devices: These devices fall into the category of personal computers, have computing and processing functions, and generally also have mobile Internet access features. These terminals include: PDA, MID and UMPC devices, etc.

[0109] (3) Portable entertainment devices: These devices can display and play multimedia content. They include audio and video players, handheld game consoles, e-books, smart toys, and portable car navigation devices.

[0110] (4) Specific server: A device that provides computing services. The server consists of a processor, hard disk, memory, system bus, etc. The server is similar to a general computer architecture, but because it needs to provide highly reliable services, it has higher requirements in terms of processing power, stability, reliability, security, scalability, and manageability.

[0111] (5) Other electronic devices with data interaction functions.

[0112] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A method for automatic target recognition based on SAR time-series remote sensing images, characterized in that: The following steps are involved: S1: Obtain all Sentinel-1 data of the area to be studied within a time period and sort them in chronological order to form a time series dataset; S2: Calculate the median of the phase dimension of each pixel in the time series data set to obtain a median image; S3: Calculate the median water index MWI according to the median image, extract surface water data according to the median water index MWI, and obtain a pixel-based water classification image. The median water index MWI is calculated as follows: MWI=a·lnZ-b+k; Wherein, a is the median multiplied by a constant, b is the median plus a constant, k is a harmonic parameter, and Z is the median of the time dimension of each pixel in the median image; S4: constructing an object using a connected domain labeling algorithm for the pixel-based water body classification image to obtain an object-based water body classification image; S5: Calculating the shape feature of each object in the object-based water body classification image; S6: Based on the object-based water body classification image, pre-extract each aquaculture pond object based on the median image and the shape feature based on random forest; S7: Calculate the boundary direction change index of each aquaculture pond object. The boundary direction change index calculation method is as follows: S7.1: Determine the boundary pixels of the object; S7.2: Construct convolution kernels in four directions: S7.3: performing convolution operations on the boundary pixels of each object using the four convolution kernels respectively, and adding the four convolution results. If the boundary direction changes, the sum of the four convolutions is 1, otherwise it is 0; S7.4: Calculate the sum of the convolution results of all boundary pixels corresponding to each object as the final boundary direction change index of the object; The unqualified aquaculture pond objects are filtered out according to the boundary direction change index to obtain the final data of the aquaculture pond objects.

2. The target automatic recognition method according to claim 1, characterized in that: In step S1, the time series data set needs to be preprocessed, including orbit correction, elimination of GRD boundary noise, thermal noise removal, radiation calibration, filtering, terrain correction and decibel processing, and finally the backscatter coefficient is obtained.

3. The target automatic recognition method according to claim 1, characterized in that: The step S2 is specifically as follows: setting the size of the time series data set to m rows×n columns×t dimensions, calculating the median of the time phase dimension pixel by pixel, and the median Z of the time phase dimension is calculated as follows: Z=median(i1,i2,…,i t ); Wherein, i is a pixel, and finally the median image with a size of m rows×n columns is obtained.

4. The target automatic recognition method according to claim 1, characterized in that: In step S3, a is 2, b is 6, and k is 0.

1.

5. The target automatic recognition method according to claim 1, characterized in that: In step S4, the connected domain labeling algorithm is used to construct the object, and the steps are as follows: S4.1: traverse each pixel in the water body classification image from left to right and from top to bottom; S4.2: If the pixel has been traversed, continue traversing, otherwise proceed to the next step; S4.3: If the pixel has not been traversed, it is used as a starting seed and put into a set marked as label, and then all the pixels that meet the conditions are pushed into the stack according to the specified kernel neighborhood relationship; S4.4: Pop the top pixel of the stack, put the pixel into a set marked with label, and push all the pixels that meet the conditions into the stack again according to the specified kernel neighborhood relationship; S4.5: Repeat the previous step until the stack is empty, thereby constructing a connected region in the water body classification image, that is, the object marked as label; S4.6: Repeat step S4.2 until the entire water body classification image is traversed, and finally the pixel-based water body classification image is converted into the object-based water body classification image.

6. The target automatic recognition method according to claim 1, characterized in that: In step S5, the shape features include perimeter, area, aspect ratio and compactness.

7. The target automatic recognition method according to claim 6, characterized in that: The step S6 is specifically as follows: based on the water body classification image, the median image and the four indicators of perimeter, area, aspect ratio and compactness are input into a random forest classifier, and training samples are selected for large open water areas and the breeding ponds respectively to perform supervised classification, so as to realize pre-extraction of each breeding pond.

8. A target automatic recognition device based on SAR time-series remote sensing images, using the target automatic recognition method according to any one of claims 1 to 7, characterized in that: include: The acquisition module is used to obtain all Sentinel-1 data of the area to be studied within a time period and sort them in chronological order to form a time series data set; A median module is used to calculate the median of the phase dimension of each pixel in the time series data set to obtain a median image; A classification module, used to calculate and obtain a median water index MWI based on the median image, and to extract surface water data based on the median water index MWI to obtain a water classification image based on pixels; A construction module is used to construct an object using a connected domain labeling algorithm for the pixel-based water body classification image to obtain an object-based water body classification image; A calculation module, used for calculating the shape feature of each object in the object-based water body classification image; An extraction module, for pre-extracting each aquaculture pond object based on the object-based water body classification image according to the median image and the shape feature based on a random forest; The screening module is used to calculate the boundary direction change index of each of the breeding pond objects, filter out unqualified breeding pond objects according to the boundary direction change index, and obtain the final data of the breeding pond objects.

Images