Method for determining water body area of reservoir by using remote sensing image based on deep learning network

Abstract

The invention discloses a method for determining the area of a water body of a reservoir by using a remote sensing image based on a deep learning network. The method comprises the following steps: acquiring multi-source SAR image data and DEM data of a target area containing the reservoir; the multi-source SAR image data are preprocessed, and reservoir orthographic SAR image data are acquired in combination with the DEM data; determining a training sample set and a test sample set based on the reservoir orthographic SAR image data; constructing an initial water body extraction model based on a deep learning network, and training and testing the initial water body extraction model based on the training sample set and the test sample set to obtain an optimal water body extraction model; and determining a water body region based on the reservoir orthographic SAR image data and the optimal water body extraction model, and determining the water body area of the reservoir in combination with the boundary vector of the reservoir. According to the method, water body objects of different sizes and levels can be accurately recognized, the detection precision of small water bodies can be improved, and the problem of water body misrecognition caused by mountain shadows is solved.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and more specifically, to a method and system for determining the water area of ​​a reservoir by using remote sensing images based on a deep learning network. Background Art

[0002] Reservoir water area extraction and change monitoring is a key environmental monitoring field, especially in water resources management and disaster prevention. This task faces a series of challenges at the technical level, mainly because the reservoir environment is usually complex, and weather conditions and terrain factors can have a significant impact on the monitoring results. Therefore, the use of advanced technical means for accurate monitoring is the core of research in this field. At present, many scholars have conducted extensive research on SAR image extraction of water bodies and their beach change monitoring. Due to its unique imaging mechanism, SAR images have special advantages in monitoring water bodies and their surrounding environments. Compared with optical images, SAR images can obtain data at night and under cloudy and rainy weather conditions, so they are suitable for long-term and continuous environmental monitoring. However, the data characteristics of SAR images also bring certain processing challenges, especially for the accurate extraction of water areas. Traditional water body segmentation algorithms, such as threshold segmentation methods, edge detection methods, and clustering segmentation methods, are simple and easy to use, but they are very sensitive to coherent speckle noise. In practical applications, they often cause problems such as over-segmentation and edge discontinuity, and are difficult to apply to SAR images of complex scenes. These limitations have promoted the development of deep learning-based methods that can extract richer and deeper feature information from SAR images. The fully convolutional neural network (FCN) proposed in the literature marks the application of deep learning in semantic segmentation tasks, but the upsampling operation of FCN is too simple, resulting in the loss of detail information. Modern research has also explored the use of digital elevation model (DEM) data to help suppress the interference of mountain shadows in terrain-undulating areas on water extraction, as well as reducing the cost of new sample annotation through transfer learning and improving the timeliness of water extraction.

[0003] In order to overcome the technical challenges encountered by traditional SAR image water extraction methods in monitoring reservoir water area changes, a method and system for determining reservoir water area using remote sensing images based on a deep learning network is needed. Summary of the invention

[0004] The present invention proposes a method and system for determining the water area of ​​a reservoir using remote sensing images based on a deep learning network, so as to solve the problem of how to determine the water area of ​​a reservoir efficiently and accurately.

[0005] In order to solve the above problem, according to one aspect of the present invention, a method for determining the water area of ​​a reservoir using remote sensing images based on a deep learning network is provided, the method comprising:

[0006] Acquire multi-source SAR image data and DEM data of the target area including the reservoir;

[0007] Preprocessing the multi-source SAR image data, and acquiring reservoir ortho-SAR image data in combination with the DEM data;

[0008] Determine a training sample set and a test sample set based on the reservoir ortho-SAR image data;

[0009] Constructing an initial water extraction model based on a deep learning network, and training and testing the initial water extraction model based on the training sample set and the test sample set to obtain an optimal water extraction model;

[0010] The water body region is determined based on the ortho-SAR image data of the reservoir and the optimal water body extraction model, and the water body area of ​​the reservoir is determined in combination with the boundary vector of the reservoir.

[0011] Preferably, the multi-source SAR image data is preprocessed, including:

[0012] The multi-source SAR image data is subjected to image filtering, geocoding and radiation calibration processing.

[0013] Preferably, the radiation calibration process includes:

[0014] Quantify the radiation intensity of SAR image data to a range of 0-255;

[0015] Selecting a SAR image data as a reference image, and taking the scattering intensity of the water body in the reference image as a reference standard;

[0016] Select water areas in other SAR image data, count the average radiation intensity corresponding to each other SAR image data, and calculate the radiation intensity adjustment coefficient based on the average radiation intensity and the scattering intensity of the water body in the reference image;

[0017] The radiation intensity of each other SAR image data is corrected based on the radiation intensity adjustment coefficient.

[0018] Preferably, the determining of the training sample set and the test sample set based on the reservoir ortho-SAR image data comprises:

[0019] Samples containing water bodies are selected from the orthophoto SAR image data of the reservoir, and the selected samples are amplified by diagonal mirroring, horizontal flipping and vertical flipping, and the amplified sample data are divided into a training sample set and a test sample set.

[0020] Preferably, the water body extraction model adopts the U-Net framework and processes the input image based on the U-Net downsampling strategy. After three downsamplings, a shallow feature layer is obtained, and then the shallow feature map is divided into blocks and sent to an encoder containing 12 layers of Transformer. With the help of its related strategies, global information is obtained, and the image size is restored and the corresponding features are integrated through upsampling in the decoding stage, and finally the image segmentation result is output by the classifier.

[0021] According to another aspect of the present invention, a system for determining the water area of ​​a reservoir using remote sensing images based on a deep learning network is provided, the system comprising:

[0022] A data acquisition unit, used for acquiring multi-source SAR image data and DEM data of a target area including a reservoir;

[0023] A preprocessing unit, used for preprocessing the multi-source SAR image data, and obtaining reservoir ortho-SAR image data in combination with the DEM data;

[0024] A sample determination unit, used for determining a training sample set and a test sample set based on the reservoir ortho-SAR image data;

[0025] A water body extraction model determination unit, used to construct an initial water body extraction model based on a deep learning network, and train and test the initial water body extraction model based on the training sample set and the test sample set to obtain an optimal water body extraction model;

[0026] The area determination unit is used to determine the water body area based on the ortho-SAR image data of the reservoir and the optimal water body extraction model, and determine the water body area of ​​the reservoir in combination with the boundary vector of the reservoir.

[0027] Preferably, the preprocessing unit preprocesses the multi-source SAR image data, including:

[0028] The multi-source SAR image data is subjected to image filtering, geocoding and radiation calibration processing.

[0029] Preferably, the radiation calibration processing of the pre-processing unit includes:

[0030] Quantify the radiation intensity of SAR image data to a range of 0-255;

[0031] Selecting a SAR image data as a reference image, and taking the scattering intensity of the water body in the reference image as a reference standard;

[0032] Select water areas in other SAR image data, count the average radiation intensity corresponding to each other SAR image data, and calculate the radiation intensity adjustment coefficient based on the average radiation intensity and the scattering intensity of the water body in the reference image;

[0033] The radiation intensity of each other SAR image data is corrected based on the radiation intensity adjustment coefficient.

[0034] Preferably, the sample determination unit determines the training sample set and the test sample set based on the reservoir ortho-SAR image data, including:

[0035] Samples containing water bodies are selected from the orthophoto SAR image data of the reservoir, and the selected samples are amplified by diagonal mirroring, horizontal flipping and vertical flipping, and the amplified sample data are divided into a training sample set and a test sample set.

[0036] Preferably, the water body extraction model adopts the U-Net framework and processes the input image based on the U-Net downsampling strategy. After three downsamplings, a shallow feature layer is obtained, and then the shallow feature map is divided into blocks and sent to an encoder containing 12 layers of Transformer. With the help of its related strategies, global information is obtained, and the image size is restored and the corresponding features are integrated through upsampling in the decoding stage, and finally the image segmentation result is output by the classifier.

[0037] The present invention provides a method and system for determining the water area of ​​a reservoir using remote sensing images based on a deep learning network, including: obtaining multi-source SAR image data and DEM data of a target area including a reservoir; preprocessing the multi-source SAR image data, and obtaining the reservoir ortho-SAR image data in combination with the DEM data; determining a training sample set and a test sample set based on the reservoir ortho-SAR image data; constructing an initial water extraction model based on a deep learning network, and training and testing the initial water extraction model based on the training sample set and the test sample set to obtain an optimal water extraction model; determining a water area based on the reservoir ortho-SAR image data and the optimal water extraction model, and determining the water area of ​​the reservoir in combination with the boundary vector of the reservoir. The method of the present invention can accurately identify water objects of different sizes and levels, improve the detection accuracy of small water bodies, and reduce the problem of water body misidentification caused by mountain shadows. It is applied to the change monitoring of small-area water bodies such as reservoirs, which can not only accurately extract water bodies, but also analyze seasonal changes in reservoirs, which is conducive to better understanding the changes in water volume in reservoirs. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] A more complete understanding of exemplary embodiments of the present invention may be obtained by referring to the following drawings:

[0039] Figure 1A flowchart of a method 100 for determining a reservoir water area by using remote sensing images based on a deep learning network according to an embodiment of the present invention;

[0040] Figure 2 A schematic diagram of a method for determining a reservoir water area according to an embodiment of the present invention;

[0041] Figure 3 A schematic diagram of the structure of a deep learning network according to an embodiment of the present invention;

[0042] Figure 4 is an optical map of a reservoir region according to an embodiment of the present invention;

[0043] Figure 5 A schematic diagram of a reservoir water area extraction result according to an embodiment of the present invention;

[0044] Figure 6 A histogram showing changes in reservoir water area according to an embodiment of the present invention;

[0045] Figure 7 It is a structural schematic diagram of a system 700 for determining the water area of ​​a reservoir by utilizing remote sensing images based on a deep learning network according to an embodiment of the present invention. DETAILED DESCRIPTION

[0046] Now, exemplary embodiments of the present invention are described with reference to the accompanying drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. These embodiments are provided to disclose the present invention in detail and completely and to fully convey the scope of the present invention to those skilled in the art. The terms used in the exemplary embodiments shown in the accompanying drawings are not intended to limit the present invention. In the accompanying drawings, the same units / elements are marked with the same reference numerals.

[0047] Unless otherwise specified, the terms (including technical terms) used herein have the commonly understood meanings to those skilled in the art. In addition, it is understood that the terms defined in commonly used dictionaries should be understood to have the same meanings as those in the context of the relevant fields, and should not be understood as idealized or overly formal meanings.

[0048] Figure 1 1 is a flow chart of a method 100 for determining the water area of ​​a reservoir using remote sensing images based on a deep learning network according to an embodiment of the present invention. Figure 1As shown, the method for determining the water area of ​​a reservoir using remote sensing images based on a deep learning network provided by an embodiment of the present invention can accurately identify water objects of different sizes and levels, can improve the detection accuracy of small water bodies, and reduce the problem of misidentification of water bodies caused by mountain shadows. When applied to the change monitoring of small-area water bodies such as reservoirs, it can not only accurately extract the water body, but also analyze the seasonal changes of the reservoir, which is conducive to a better understanding of the changes in the water volume of the reservoir. The method 100 for determining the water area of ​​a reservoir using remote sensing images based on a deep learning network provided by an embodiment of the present invention starts from step 101. In step 101, multi-source SAR image data and DEM data of the target area including the reservoir are obtained.

[0049] In step 102, the multi-source SAR image data is preprocessed and combined with the DEM data to obtain reservoir ortho-SAR image data.

[0050] Preferably, the multi-source SAR image data is preprocessed, including:

[0051] The multi-source SAR image data is subjected to image filtering, geocoding and radiation calibration processing.

[0052] Preferably, the radiation calibration process includes:

[0053] Quantify the radiation intensity of SAR image data to a range of 0-255;

[0054] Selecting a SAR image data as a reference image, and taking the scattering intensity of the water body in the reference image as a reference standard;

[0055] Select water areas in other SAR image data, count the average radiation intensity corresponding to each other SAR image data, and calculate the radiation intensity adjustment coefficient based on the average radiation intensity and the scattering intensity of the water body in the reference image;

[0056] The radiation intensity of each other SAR image data is corrected based on the radiation intensity adjustment coefficient.

[0057] Combination Figure 2 As shown, in the present invention, multi-source remote sensing image data of the target area is first obtained and preprocessed. Specifically, the preprocessing includes:

[0058] 1) Image filtering. Image filtering is one of the important steps in SAR image processing and is used to suppress speckle noise. In the present invention, image filtering is performed by using a LEE filtering algorithm suitable for noise suppression of SAR images.

[0059] 2) Geocoding. In order to achieve high-precision geocoding, the present invention adopts a method based on the RD equation and combines image track information and digital elevation model (DEM) data to obtain ortho-SAR images. Among them, the DEM data used for geocoding uses TanDEM-X 90m data.

[0060] 3) Radiation calibration. First, the radiation intensity of the SAR image is quantified to a range of 0-255 by using the mean and variance method. Next, a SAR image is selected as a benchmark, and the scattering intensity of its water body is used as a reference standard. After that, water areas are selected in other SAR images, their average radiation intensity is counted, and the radiation intensity adjustment coefficient k compared with the benchmark image is calculated based on these means. Finally, the radiation intensity of the SAR image is corrected using this adjustment coefficient k to ensure that they are on the same measurement standard. Through such correction, it can be ensured that all SAR images have the same radiation intensity scale in the water body extraction algorithm, thereby improving the accuracy and reliability of the algorithm.

[0061] In step 103, a training sample set and a test sample set are determined based on the reservoir ortho-SAR image data.

[0062] Preferably, the determining of the training sample set and the test sample set based on the reservoir ortho-SAR image data comprises:

[0063] Samples containing water bodies are selected from the orthophoto SAR image data of the reservoir, and the selected samples are amplified by diagonal mirroring, horizontal flipping and vertical flipping, and the amplified sample data are divided into a training sample set and a test sample set.

[0064] In the present invention, samples containing water bodies are selected from reservoir ortho-SAR image data and expanded to obtain multiple samples, which are divided into a training sample set and a test sample set.

[0065] In an embodiment of the present invention, SAR image data acquired in July 2019 was selected, and 36 samples containing water bodies were randomly selected, each of which had a size of 1024×1024 pixels. In these samples, the water body area was marked as 1, and the non-water body area was marked as 0. In order to enhance the generalization ability of the model in a small number of sample data sets, a data expansion strategy was adopted, and the initial data set was implemented with diagonal mirroring, horizontal flipping, and vertical flipping data amplification techniques. Through these methods, the SAR image water body samples were increased from the original 36 to 136, and each sample maintained the size of 1024×1024 unchanged. Through a random sampling method, 123 samples were selected as training sets and 13 samples were selected as validation sets, keeping the ratio of training sets to validation sets at approximately 10:1. In addition, in order to evaluate the accuracy and generalization ability of the algorithm of the present invention, two areas in the plains and mountains were randomly selected from the SAR images in May 2020 as test sets. Among them, the SAR image size in the plain area is 2021×2375, and the SAR image size in the mountainous area is 1058×1189.

[0066] In step 104, an initial water body extraction model based on a deep learning network is constructed, and the initial water body extraction model is trained and tested based on the training sample set and the test sample set to obtain an optimal water body extraction model.

[0067] Preferably, the water body extraction model adopts the U-Net framework and processes the input image based on the U-Net downsampling strategy. After three downsamplings, a shallow feature layer is obtained, and then the shallow feature map is divided into blocks and sent to an encoder containing 12 layers of Transformer. With the help of its related strategies, global information is obtained, and the image size is restored and the corresponding features are integrated through upsampling in the decoding stage, and finally the image segmentation result is output by the classifier.

[0068] Combination Figure 3As shown, in the present invention, a network structure designed based on the traditional U-Net framework is adopted, which includes an encoder and a decoder, which form a U-shaped structure. In this structure, the encoder uses convolution operations to extract surface features of the image, which are then converted into labeled image blocks through linear embedding and position encoding. These image blocks are then passed to multiple Transformer modules, which can capture local spatial details in the convolution step and global context information in the Transformer step. In the decoding stage, the structure adjusts the spatial dimension of the output from the encoder to match the convolutional network module, thereby enhancing the model's ability to integrate spatial information locally and globally. In order to adapt to different spatial sizes, the decoder also includes multiple upsampling steps. This design combines the local feature extraction capability of the convolutional neural network with the global feature integration capability of the Transformer, aiming to provide richer context information for image segmentation tasks. Specifically, the model adopts the downsampling strategy of U-Net, downsampling the input image three times to generate feature maps of different scales, which constitute a shallow feature layer. These shallow feature maps are further divided into image blocks and sent to the encoder consisting of 12 layers of Transformer for processing. In the encoder, Transformer uses its unique normalization, multi-head attention, and deep perceptron strategies to process image blocks and obtain global context information. In the decoding stage, the model restores the original spatial size of the image by upsampling the encoded features three times, and in the process fuses the shallow features of the same resolution generated by the encoder, and finally produces high-quality image segmentation results through the classifier.

[0069] in,

[0070]

[0071]

[0072]

[0073] Among them, Q is the query matrix, K is the content you want to pay attention to, QK T is the dot product operation, d model is the dimension of the vector, i is the i-th dimension of the vector, and pos is the current absolute position.

[0074] During the network training process, an improved deep learning network model was used as the basic model for training. The learning rate of the model was set to 0.001, and the patch size was defined as 16 and the size of the hidden layer was 512. In addition, the number of attention heads of the Transformer model was set to 4, and the number of layers of the Transformer encoder and decoder were both configured to 4.

[0075] In the process of determining the model, in order to more accurately measure the performance of the algorithm, multiple evaluation indicators including accuracy, recall, total accuracy, F1 score and intersection-over-union ratio are used to quantify the accuracy of water body extraction. The specific formulas include:

[0076]

[0077]

[0078]

[0079]

[0080] Among them, TP is true positive, FP is false positive, TN is true negative, and FN is false negative.

[0081] In step 105, the water body region is determined based on the ortho-SAR image data of the reservoir and the optimal water body extraction model, and the water body area of ​​the reservoir is determined in combination with the boundary vector of the reservoir.

[0082] In the present invention, after determining the optimal water body extraction model, the required reservoir orthophoto SAR impact data is input into the model to determine the water body area, and then the area is calculated in combination with the boundary vector of the reservoir to determine the water body area of ​​the reservoir.

[0083] In this invention, in order to verify the effectiveness of the algorithm of the invention, several classic models HRNet, U-Net and FCN-VGG16 are selected for the task of water extraction on high-resolution SAR images. These models are used as a control group to compare and analyze the performance of the algorithm of the invention.

[0084] In the comparative study with U-Net, FCN-VGG16 and HRNet models, the model proposed in this invention shows a better performance in SAR image water body recognition. The model especially improves the detection accuracy of small water bodies and significantly reduces the problem of water body misidentification caused by mountain shadows. When applied to the change monitoring of small-area water bodies such as reservoirs, it can not only accurately extract water bodies, but also analyze the seasonal changes of reservoirs, which is conducive to a better understanding of the changes in the water volume of reservoirs. The evaluation indicators show that the model has an accuracy of 89.54%, a recall rate of 91.24%, an overall accuracy of 98.01%, an F1 score of 90.26%, and an intersection-over-union ratio of 82.28%. Compared with the comparison algorithm, the F1 score of this model has increased by 8% to 39%.

[0085] The method of the present invention combines the deep learning model and the high-resolution (3m) data of the COSMO-SkyMed satellite, makes full use of the global context capture capability of the Transformer model and the multi-scale feature extraction advantages of the U-Net model, and constructs a water body extraction model suitable for small-area SAR images, realizing accurate monitoring of changes in the water area of ​​the reservoir. In the model, the self-attention mechanism of the Transformer architecture plays a key role. It can process global information from SAR images, especially under the condition of small sample data. This global context capture capability greatly improves the generalization performance of the model and effectively reduces the overfitting phenomenon. The advantage of the U-Net model is that it can effectively extract multi-scale features in the image, capture context information through the encoder gradually downsampling, and then restore detail information through the decoder step by step upsampling. U-Net can accurately identify water objects of different sizes and levels. This allows the combined model to not only use deep features for semantic judgment, but also use shallow features to capture finer local details at the same time, thereby significantly improving the accuracy and efficiency of monitoring changes in the water area of ​​the reservoir.

[0086] In an embodiment of the present invention, the remote sensing data uses SAR images of 7 periods of COSMO-SkyMed satellites in 2019-07, 2019-09, 2019-11, 2019-12, 2020-02, 2020-05, and 2020-06, and the basic parameters of the images are shown in Table 1. The SAR images in July 2019 are used to make training samples, and the SAR images in May 2020 are used to verify the accuracy of water body extraction. At the same time, based on the water body algorithm proposed in the present invention, the reservoir water area extraction work is carried out on the SAR images of 6 periods in 2023 for the analysis of the evolution law of the reservoir area.

[0087] Table 1 Basic parameters

[0088] satellite Cosmo-SkyMed polarization HH Resolution / m 3 track Raise track Angle of incidence / (°) 26.6 Width / Km 40×40

[0089] During the analysis, the optical images of the reservoir area were obtained. Figure 4 As shown in Figure 2, the reservoir water area extraction work was carried out on the SAR images of the 6th phase in 2023, and the final reservoir water area extraction results are as follows Figure 5 Finally, combined with the boundary vector of the reservoir, the changes in the reservoir area at different times are extracted. The results of the changes in the reservoir water area are shown in Figure 6 shown.

[0090] Figure 7 FIG. 7 is a schematic diagram of a system 700 for determining the water area of ​​a reservoir using remote sensing images based on a deep learning network according to an embodiment of the present invention. Figure 7 As shown, the system 700 for determining the water area of ​​a reservoir using remote sensing images based on a deep learning network provided by an embodiment of the present invention includes: a data acquisition unit 701, a preprocessing unit 702, a sample determination unit 703, a water body extraction model determination unit 704 and an area determination unit 705.

[0091] Preferably, the data acquisition unit 701 is used to acquire multi-source SAR image data and DEM data of a target area including a reservoir.

[0092] Preferably, the preprocessing unit 702 is used to preprocess the multi-source SAR image data and obtain reservoir ortho-SAR image data in combination with the DEM data.

[0093] Preferably, the preprocessing unit 702 preprocesses the multi-source SAR image data, including:

[0094] The multi-source SAR image data is subjected to image filtering, geocoding and radiation calibration processing.

[0095] Preferably, the radiation calibration processing of the pre-processing unit 702 includes:

[0096] Quantify the radiation intensity of SAR image data to a range of 0-255;

[0097] Selecting a SAR image data as a reference image, and taking the scattering intensity of the water body in the reference image as a reference standard;

[0098] Select water areas in other SAR image data, count the average radiation intensity corresponding to each other SAR image data, and calculate the radiation intensity adjustment coefficient based on the average radiation intensity and the scattering intensity of the water body in the reference image;

[0099] The radiation intensity of each other SAR image data is corrected based on the radiation intensity adjustment coefficient.

[0100] Preferably, the sample determination unit 703 is used to determine a training sample set and a test sample set based on the reservoir ortho-SAR image data.

[0101] Preferably, the sample determination unit 703 determines the training sample set and the test sample set based on the reservoir ortho-SAR image data, including:

[0102] Samples containing water bodies are selected from the orthophoto SAR image data of the reservoir, and the selected samples are amplified by diagonal mirroring, horizontal flipping and vertical flipping, and the amplified sample data are divided into a training sample set and a test sample set.

[0103] Preferably, the water body extraction model determination unit 704 is used to construct an initial water body extraction model based on a deep learning network, and train and test the initial water body extraction model based on the training sample set and the test sample set to obtain an optimal water body extraction model;

[0104] Preferably, the area determination unit 705 is used to determine the water body region based on the ortho-SAR image data of the reservoir and the optimal water body extraction model, and determine the water body area of ​​the reservoir in combination with the boundary vector of the reservoir.

[0105] Preferably, the water body extraction model adopts the U-Net framework and processes the input image based on the U-Net downsampling strategy. After three downsamplings, a shallow feature layer is obtained, and then the shallow feature map is divided into blocks and sent to an encoder containing 12 layers of Transformer. With the help of its related strategies, global information is obtained, and the image size is restored and the corresponding features are integrated through upsampling in the decoding stage, and finally the image segmentation result is output by the classifier.

[0106] The system 700 for determining the water area of ​​a reservoir using remote sensing images based on a deep learning network in an embodiment of the present invention corresponds to the method 100 for determining the water area of ​​a reservoir using remote sensing images based on a deep learning network in another embodiment of the present invention, and will not be repeated here.

[0107] The invention has been described above with reference to a few embodiments. However, it is readily apparent to a person skilled in the art that other embodiments than the ones disclosed above are equally within the scope of the invention, as defined by the appended patent claims.

[0108] Generally, all terms used in the claims are to be interpreted according to their ordinary meaning in the technical field, unless explicitly defined otherwise therein. All references to "a / said / the [means, components, etc.]" are to be openly interpreted as at least one instance of said means, components, etc., unless explicitly stated otherwise. The steps of any method disclosed herein do not necessarily have to be performed in the exact order disclosed, unless explicitly stated otherwise.

[0109] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0110] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0111] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0112] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0113] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for determining the water area of ​​a reservoir using remote sensing images based on a deep learning network, characterized in that: The method comprises: Acquire multi-source SAR image data and DEM data of the target area including the reservoir; Preprocessing the multi-source SAR image data, and acquiring reservoir ortho-SAR image data in combination with the DEM data; Determine a training sample set and a test sample set based on the reservoir ortho-SAR image data; Constructing an initial water extraction model based on a deep learning network, and training and testing the initial water extraction model based on the training sample set and the test sample set to obtain an optimal water extraction model; The water body region is determined based on the ortho-SAR image data of the reservoir and the optimal water body extraction model, and the water body area of ​​the reservoir is determined in combination with the boundary vector of the reservoir.

2. The method according to claim 1, characterized in that Preprocessing the multi-source SAR image data includes: The multi-source SAR image data is subjected to image filtering, geocoding and radiation calibration processing.

3. The method according to claim 2, characterized in that The radiation calibration process includes: Quantify the radiation intensity of SAR image data to a range of 0-255; Selecting a SAR image data as a reference image, and taking the scattering intensity of the water body in the reference image as a reference standard; Select water areas in other SAR image data, count the average radiation intensity corresponding to each other SAR image data, and calculate the radiation intensity adjustment coefficient based on the average radiation intensity and the scattering intensity of the water body in the reference image; The radiation intensity of each other SAR image data is corrected based on the radiation intensity adjustment coefficient.

4. The method according to claim 1, characterized in that: The determining of the training sample set and the test sample set based on the reservoir ortho-SAR image data comprises: Samples containing water bodies are selected from the orthophoto SAR image data of the reservoir, and the selected samples are amplified by diagonal mirroring, horizontal flipping and vertical flipping, and the amplified sample data are divided into a training sample set and a test sample set.

5. The method according to claim 1, characterized in that The water body extraction model adopts the U-Net framework and processes the input image based on the U-Net downsampling strategy. After three downsamplings, a shallow feature layer is obtained, and then the shallow feature map is divided into blocks and sent to an encoder containing 12 layers of Transformer. With the help of its related strategies, global information is obtained. In the decoding stage, the image size is restored by upsampling and the corresponding features are fused. Finally, the classifier outputs the image segmentation result.

6. A system for determining the water area of ​​a reservoir using remote sensing images based on a deep learning network, characterized in that: The system comprises: A data acquisition unit, used for acquiring multi-source SAR image data and DEM data of a target area including a reservoir; A preprocessing unit, used for preprocessing the multi-source SAR image data, and obtaining reservoir ortho-SAR image data in combination with the DEM data; A sample determination unit, used for determining a training sample set and a test sample set based on the reservoir ortho-SAR image data; A water body extraction model determination unit, used to construct an initial water body extraction model based on a deep learning network, and train and test the initial water body extraction model based on the training sample set and the test sample set to obtain an optimal water body extraction model; The area determination unit is used to determine the water body area based on the ortho-SAR image data of the reservoir and the optimal water body extraction model, and determine the water body area of ​​the reservoir in combination with the boundary vector of the reservoir.

7. The system according to claim 6, characterized in that The preprocessing unit preprocesses the multi-source SAR image data, including: The multi-source SAR image data is subjected to image filtering, geocoding and radiation calibration processing.

8. The system according to claim 7, characterized in that The radiation calibration processing of the pre-processing unit includes: Quantify the radiation intensity of SAR image data to a range of 0-255; Selecting a SAR image data as a reference image, and taking the scattering intensity of the water body in the reference image as a reference standard; Select water areas in other SAR image data, count the average radiation intensity corresponding to each other SAR image data, and calculate the radiation intensity adjustment coefficient based on the average radiation intensity and the scattering intensity of the water body in the reference image; The radiation intensity of each other SAR image data is corrected based on the radiation intensity adjustment coefficient.

9. The system according to claim 6, characterized in that The sample determination unit determines a training sample set and a test sample set based on the reservoir ortho-SAR image data, including: Samples containing water bodies are selected from the orthophoto SAR image data of the reservoir, and the selected samples are amplified by diagonal mirroring, horizontal flipping and vertical flipping, and the amplified sample data are divided into a training sample set and a test sample set.

10. The system according to claim 6, characterized in that The water body extraction model adopts the U-Net framework and processes the input image based on the U-Net downsampling strategy. After three downsamplings, a shallow feature layer is obtained, and then the shallow feature map is divided into blocks and sent to an encoder containing 12 layers of Transformer. With the help of its related strategies, global information is obtained. In the decoding stage, the image size is restored by upsampling and the corresponding features are fused. Finally, the classifier outputs the image segmentation result.

Images