Water body extraction method under complex terrain conditions in mountainous areas based on multi-source data fusion

Abstract

The present invention provides a method for extracting water bodies under complex terrain conditions in mountainous areas based on multi-source data fusion. A comprehensive radar data image obtained by fusion of synthetic aperture radar images is determined according to terrain distribution and time sequence, so as to improve the recognition accuracy of water bodies from the aspect of image acquisition, establish a grayscale co-occurrence matrix of an initial data image, obtain image texture feature index values ​​based on the grayscale co-occurrence matrix, and remove paddy field features in the initial data image based on the texture feature index values ​​to obtain a first data image. Dual-polarization fusion and false color synthesis are performed on the initial data image to obtain a second data image. The slope and shadow of DEM data are analyzed, and mountain shadows in the DEM data are removed to obtain a third data image. The data images are used as training samples, a binary classification model is established, and subsequent water body extraction is performed based on the binary classification model to achieve accurate recognition and extraction of water body features.

Description

Technical Field

[0001] The invention relates to the technical field of image data processing, and in particular to a method for extracting water bodies under complex terrain conditions in mountainous areas based on multi-source data fusion. Background Art

[0002] At present, relevant scholars have carried out a lot of research on water extraction based on different sensors in complex terrain conditions in mountainous areas. Complex terrain, such as cold and arid areas, has complex terrain, many mountains and valleys, and remote sensing images are easily affected by interference factors such as mountain shadows and paddy fields on water bodies. These factors increase the difficulty of water body identification, especially in the absence of training sets for deep learning network training, it becomes more difficult to accurately extract water body features and perform refined semantic segmentation.

[0003] With the development of satellite remote sensing technology, synthetic aperture radar (SAR) has become an important tool for surface monitoring due to its unique imaging advantages, such as the ability to obtain surface information all day and all night. In SAR images, water bodies have a low backscatter coefficient, appear dark, and have a smooth surface. Applying synthetic aperture radar (SAR) to extract water bodies in complex mountainous terrain conditions provides a new perspective. Summary of the invention

[0004] The present invention provides a method for extracting water under complex terrain conditions in mountainous areas based on multi-source data fusion, so as to solve the problems raised in the background technology.

[0005] A method for extracting water bodies in complex mountainous terrain based on multi-source data fusion, comprising:

[0006] S1: determining a comprehensive radar data image obtained by fusion of synthetic aperture radar images according to terrain distribution and time sequence, and preprocessing the comprehensive radar data image to obtain an initial data image;

[0007] S2: establishing a gray level co-occurrence matrix of the initial data image, obtaining an image texture feature index value based on the gray level co-occurrence matrix, removing paddy field features in the initial data image based on the texture feature index value, obtaining a first data image, performing dual polarization fusion and false color synthesis on the initial data image, obtaining a second data image;

[0008] S3: Analyze the slope and shadow of the DEM data, remove the mountain shadow in the DEM data, and obtain the third data image;

[0009] S4: performing multi-source data fusion on the first data image, the second data image and the third data image, using the obtained fused image as a training sample, establishing a binary classification model, and performing subsequent water body extraction based on the binary classification model.

[0010] Preferably, in S1, determining the multi-source data image obtained by fusion of synthetic aperture radar images according to terrain distribution and time sequence includes:

[0011] Determine the lateral fusion characteristics of synthetic aperture radar images based on terrain distribution;

[0012] Based on the time sequence, determine the longitudinal fusion characteristics of synthetic aperture radar images;

[0013] Based on the horizontal fusion features and the vertical fusion features, the synthetic aperture radar images are fused to obtain a comprehensive radar data image.

[0014] Preferably, in S1, preprocessing the comprehensive radar data image to obtain an initial data image includes:

[0015] Performing multi-view processing and image filtering on the comprehensive radar data image to obtain a denoised data image;

[0016] The denoised data image is geocoded, radiometrically calibrated and decibelized to obtain an initial data image.

[0017] Preferably, in S2, a gray level co-occurrence matrix of the initial data image is established, an image texture feature index value is obtained based on the gray level co-occurrence matrix, and paddy field features in the initial data image are removed based on the texture feature index value to obtain a first data image, including:

[0018] Obtaining a gray level co-occurrence matrix from the initial data image;

[0019] Based on the gray level co-occurrence matrix, the image texture feature index value of the gray level co-occurrence matrix is ​​calculated according to the following formula;

[0020]

[0021]

[0022]

[0023]

[0024] in, Represents the uniform index value of the image texture feature, The angular second-order moment index value representing the image texture feature, The entropy index value representing the image texture feature, represents the gray level, It represents the number of times the pixel pairs with horizontal gray level i and vertical gray level j appear in the gray level co-occurrence matrix. Indicates that the gray level co-occurrence matrix is ​​normalized;

[0025] Based on the preset paddy field texture features and in combination with the image texture feature index value of the gray level co-occurrence matrix, the paddy field features in the initial data image are determined and removed to obtain the first data image.

[0026] Preferably, in S2, performing dual-polarization fusion and false color synthesis on the initial data image to obtain a second data image includes:

[0027] Acquire a radar echo signal of a synthetic aperture radar image, perform polarization decomposition on the radar echo signal to obtain an HH component and an HV component, and perform dual-polarization fusion on the initial data image according to the following formula to obtain an intermediate data image;

[0028]

[0029] in, Indicates the dual-polarization fusion process of the initial data image. represents the HH component, and HV represents the HV component;

[0030] Perform false color synthesis on the intermediate data image to highlight its features, so as to obtain a second data image.

[0031] Preferably, in S3, the slope and shadow of the DEM data are analyzed, and the mountain shadows in the DEM data are removed to obtain a third data image, including:

[0032] Obtain slope characteristics and shadow characteristics from DEM data;

[0033] Based on the preset slope characteristic range and shadow characteristic range of the mountain shadow, combined with the slope characteristics and shadow characteristics in the DEM data, the mountain shadow position is determined;

[0034] The slope features and shadow features of the mountain shadow position are removed to obtain a third data image.

[0035] Preferably, in S4, performing multi-source data fusion on the first data image, the second data image and the third data image includes:

[0036] Acquire at least one characteristic attribute of the first data image, the second data image, and the third data image, and acquire the characteristic similarity under the characteristic attribute, and acquire the data image matching result when the characteristic similarity is greater than a preset similarity, and determine the fusion position matching feature between the first data image, the second data image, and the third data image based on the matching result;

[0037] Acquire the paddy field removal area in the first data image, and mark the paddy field for the corresponding position areas in the second data image and the third data image based on the position matching feature;

[0038] Based on the hill shadow removal area in the third data image, and based on the position matching feature, the hill shadow marking is performed on the corresponding position area in the second data image and the third data image;

[0039] Acquire false color features of local objects in the second data image, obtain local object features based on the false color features, and mark local object features at corresponding positions of the first data image and the second data image based on the position matching features;

[0040] Based on the above three times of labeling results, a first labeled data image, a second labeled data image and a third labeled data image are obtained, and the same type of data features in the first labeled data image, the second labeled data image and the third labeled data image are obtained as important data features, and different types of data features are obtained as first basic data features, second basic data features and third basic data features;

[0041] The first basic data feature, the second basic data feature, and the third basic data feature are respectively fused as hierarchical features to obtain intermediate fusion features, and the important data features are secondary fused with the intermediate fusion features to obtain a fused image.

[0042] Preferably, in S4, the obtained fused image is used as a training sample to establish a binary classification model, including:

[0043] Based on the training samples, the support vector machine classification method is used to calculate the optimal hyperplane;

[0044] Based on the optimal hyperplane, a binary classification model is established.

[0045] Preferably, after establishing a binary classification model based on the optimal hyperplane, the method further includes:

[0046] Acquire a verification sample, verify the binary classification model based on the verification sample, and establish a confusion matrix based on the verification output result;

[0047] Based on the confusion matrix, the overall accuracy and Kappa coefficient are obtained, and based on the preset accuracy requirements, the model accuracy of the binary classification model is determined to meet the preset requirements;

[0048] If so, it is determined that the binary classification model has passed the verification;

[0049] Otherwise, it is determined that the binary classification model fails the validation and is retrained.

[0050] Preferably, in S4, subsequent water body extraction is performed based on a binary classification model, including:

[0051] Acquire real-time radar data and real-time DEM data, and process the real-time radar data and real-time DEM data according to steps S1-S3, and fuse the processed real-time data according to step S4 to obtain target real-time data;

[0052] The target real-time data is input into the binary classification model, the location of the water body area is determined based on the classification result, and the water body is extracted based on the location of the water body area.

[0053] Compared with the prior art, the present invention has achieved the following beneficial effects:

[0054] The comprehensive radar data image obtained by fusion of synthetic aperture radar images is determined according to terrain distribution and time sequence, so as to improve the recognition accuracy of water bodies from the aspect of image acquisition, pre-process the comprehensive radar data image to obtain an initial data image, establish a gray level co-occurrence matrix of the initial data image, obtain image texture feature index values ​​based on the gray level co-occurrence matrix, and remove paddy field features in the initial data image based on the texture feature index values ​​to obtain a first data image, perform dual polarization fusion and false color synthesis on the initial data image to obtain a second data image, analyze the slope and shadow of DEM data, remove mountain shadows in DEM data to obtain a third data image, process the image to exclude images of interference factors such as mountain shadows and paddy fields on water bodies, and provide a basis for the recognition and extraction of water bodies. Finally, multi-source data fusion is performed on the first data image, the second data image and the third data image, and the obtained fused image is used as a training sample to establish a binary classification model, and subsequent water body extraction is performed based on the binary classification model to achieve accurate recognition and extraction of water body features.

[0055] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures specifically pointed out in this application document.

[0056] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0058] Figure 1 It is a flow chart of a method for extracting water under complex terrain conditions in mountainous areas based on multi-source data fusion in an embodiment of the present invention;

[0059] Figure 2 is a display diagram of the HH image described in an embodiment of the present invention;

[0060] Figure 3 It is a display diagram of the HV image described in the embodiment of the present invention. DETAILED DESCRIPTION

[0061] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0062] Embodiment 1:

[0063] The embodiment of the present invention provides a method for extracting water bodies under complex terrain conditions in mountainous areas based on multi-source data fusion. Figure 1 As shown, including:

[0064] S1: determining a comprehensive radar data image obtained by fusion of synthetic aperture radar images according to terrain distribution and time sequence, and preprocessing the comprehensive radar data image to obtain an initial data image;

[0065] S2: establishing a gray level co-occurrence matrix of the initial data image, obtaining an image texture feature index value based on the gray level co-occurrence matrix, removing paddy field features in the initial data image based on the texture feature index value, obtaining a first data image, performing dual polarization fusion and false color synthesis on the initial data image, obtaining a second data image;

[0066] S3: Analyze the slope and shadow of the DEM data, remove the mountain shadow in the DEM data, and obtain the third data image;

[0067] S4: performing multi-source data fusion on the first data image, the second data image and the third data image, using the obtained fused image as a training sample, establishing a binary classification model, and performing subsequent water body extraction based on the binary classification model.

[0068] In this embodiment, the synthetic aperture radar image is specifically ALOS-2 data. ALOS-2 is an advanced earth observation satellite launched by the Japan Aerospace Exploration Agency (JAXA). The PALSAR-2 sensor carried by ALOS-2 can provide high-resolution radar images, providing a new data source for surface cover type identification.

[0069] In this embodiment, the data used in the present invention is the image data of the ALOS PALSAR-2 dual polarization mode (HH and HV polarization) acquired on April 27, 2023. The incident angle is 32.8°, the azimuth viewing number is 1, the azimuth resolution is 2.21m, the range viewing number is 1, and the range resolution is 2.23m.

[0070] In this embodiment, the data preprocessing steps include multi-look processing, image filtering, geocoding, and radiometric calibration and decibelization.

[0071] In this embodiment, since the grayscale value of the mountain shadow caused by the complex and undulating terrain is close to the grayscale value of the water body, it needs to be removed. The difference between the mountain shadow and the water body is the slope value. The surface slope of the water body is basically 0, while the mountain shadow has a larger slope.

[0072] In this embodiment, the DEM data is 30m ASTER DEM data acquired based on the geospatial data cloud, which is data elevation model data.

[0073] In this embodiment, the synthetic aperture radar image is different from the optical image in that it lacks color information, so false color is needed to highlight the features of the objects.

[0074] The beneficial effects of the above design scheme are: by determining the comprehensive radar data image obtained by fusion of synthetic aperture radar images according to terrain distribution and time sequence, the recognition accuracy of water bodies is improved from the aspect of image acquisition, the comprehensive radar data image is preprocessed to obtain an initial data image, a grayscale co-occurrence matrix of the initial data image is established, image texture feature index values ​​are obtained based on the grayscale co-occurrence matrix, and paddy field features in the initial data image are removed based on the texture feature index values ​​to obtain a first data image, dual-polarization fusion and false color synthesis are performed on the initial data image to obtain a second data image, the slope and shadow of the DEM data are analyzed, the mountain shadow in the DEM data is removed to obtain a third data image, the image is processed to exclude the images of interference factors such as mountain shadows and paddy fields on the water body, and a basis is provided for the recognition and extraction of water bodies. Finally, the first data image, the second data image and the third data image are subjected to multi-source data fusion, the obtained fused image is used as a training sample, a binary classification model is established, and subsequent water body extraction is performed based on the binary classification model to achieve accurate recognition and extraction of water body features.

[0075] Embodiment 2:

[0076] Based on Example 1, the embodiment of the present invention provides a method for extracting water under complex terrain conditions in mountainous areas based on multi-source data fusion, characterized in that in S1, a multi-source data image obtained by fusion of synthetic aperture radar images is determined according to terrain distribution and time sequence, including:

[0077] Determine the lateral fusion characteristics of synthetic aperture radar images based on terrain distribution;

[0078] Based on the time sequence, determine the longitudinal fusion characteristics of synthetic aperture radar images;

[0079] Based on the horizontal fusion features and the vertical fusion features, the synthetic aperture radar images are fused to obtain a comprehensive radar data image.

[0080] In this embodiment, the horizontal fusion features of the synthetic aperture radar image are determined based on the terrain distribution, the vertical fusion features of the synthetic aperture radar image are determined based on the time sequence, and the synthetic aperture radar images are fused based on the horizontal fusion features and the vertical fusion features to obtain a comprehensive radar data image, thereby ensuring the comprehensiveness of the obtained comprehensive radar data image and providing a rich data image basis for ladder recognition and extraction.

[0081] Embodiment 3:

[0082] Based on Example 1, the embodiment of the present invention provides a method for extracting water under complex terrain conditions in mountainous areas based on multi-source data fusion. In S1, the comprehensive radar data image is preprocessed to obtain an initial data image, including:

[0083] Performing multi-view processing and image filtering on the comprehensive radar data image to obtain a denoised data image;

[0084] The denoised data image is geocoded, radiometrically calibrated and decibelized to obtain an initial data image.

[0085] In this embodiment, in order to suppress the speckle noise of the comprehensive radar data image and improve the image quality, multi-view processing is performed on the single-view compound data. In addition to multi-view, enhanced filtering is also used for denoising, and the filtering can also well preserve the boundary characteristics of various objects.

[0086] The beneficial effect of the above design scheme is: by performing multi-view processing and image filtering on the comprehensive radar data image, a denoised data image is obtained, and the denoised data image is geocoded and radiometrically calibrated and decibelized to obtain an initial data image, thereby achieving preprocessing of the data image and providing a basis for subsequent further analysis and processing of the image.

[0087] Embodiment 4:

[0088] Based on Example 1, the embodiment of the present invention provides a method for extracting water bodies under complex terrain conditions in mountainous areas based on multi-source data fusion. In S2, a gray level co-occurrence matrix of the initial data image is established, an image texture feature index value is obtained based on the gray level co-occurrence matrix, and paddy field features in the initial data image are removed based on the texture feature index value to obtain a first data image, including:

[0089] Obtaining a gray level co-occurrence matrix from the initial data image;

[0090] Based on the gray level co-occurrence matrix, the image texture feature index value of the gray level co-occurrence matrix is ​​calculated according to the following formula;

[0091]

[0092]

[0093]

[0094]

[0095] in, Represents the uniform index value of the image texture feature, The angular second-order moment index value representing the image texture feature, The entropy index value representing the image texture feature, represents the gray level, It represents the number of times the pixel pairs with horizontal gray level i and vertical gray level j appear in the gray level co-occurrence matrix. Indicates that the gray level co-occurrence matrix is ​​normalized;

[0096] Based on the preset paddy field texture features and in combination with the image texture feature index value of the gray level co-occurrence matrix, the paddy field features in the initial data image are determined and removed to obtain the first data image.

[0097] In this embodiment, under the complex surface environment, there are many types of cultivated land, among which paddy fields have a relatively weak backscattering characteristic on SAR images due to their high water content, and are dark in color, which is difficult to distinguish from water bodies in grayscale tones. Texture is a basic feature of images, which can express the surface and structural properties of objects, such as the roughness of the surface and the neatness of the structure. Gray-level co-occurrence matrix is ​​a texture feature extraction method based on statistical analysis, and is also one of the common methods to describe texture by studying the spatial correlation characteristics of image grayscale.

[0098] In this embodiment, the paddy field and the water body have great differences in three characteristic indicators: homogeneity (hom), angular second moment (ASM) and entropy (ENT), so the study selected these three indicators as texture features.

[0099] In this embodiment, uniformity hom reflects the homogeneity of image texture, and the more uniform the local area is, the larger its value is; the angular second moment ASM reflects the uniformity of image grayscale distribution and texture coarseness, and the more uniform the local area is, the larger its value is; entropy ENT reflects the amount of image information and the randomness of element values, and the greater the local change, the greater the entropy.

[0100] The beneficial effect of the above design scheme is: by establishing the grayscale co-occurrence matrix of the initial data image, obtaining the image texture feature index value based on the grayscale co-occurrence matrix, and removing the paddy field features in the initial data image based on the texture feature index value, a first data image is obtained, and the paddy field features that are difficult to identify are removed, thereby providing a basis for the subsequent accurate identification of water bodies.

[0101] Embodiment 5:

[0102] Based on Example 1, the embodiment of the present invention provides a method for extracting water under complex terrain conditions in mountainous areas based on multi-source data fusion. In S2, the initial data image is subjected to dual-polarization fusion and false color synthesis to obtain a second data image, including:

[0103] Acquire a radar echo signal of a synthetic aperture radar image, perform polarization decomposition on the radar echo signal to obtain an HH component and an HV component, and perform dual-polarization fusion on the initial data image according to the following formula to obtain an intermediate data image;

[0104]

[0105] in, Indicates the dual-polarization fusion process of the initial data image. represents the HH component, and HV represents the HV component;

[0106] Perform false color synthesis on the intermediate data image to highlight its features, so as to obtain a second data image.

[0107] In this embodiment, the HH component is a horizontal polarization component, and the HV component is a horizontal-vertical polarization component. Figure 2 Shown is the HH image, Figure 3 For HV images.

[0108] The beneficial effect of the above design scheme is: by performing dual-polarization fusion and false color synthesis on the initial data image, a second data image is obtained, which further highlights the characteristics of the image and provides a basis for the subsequent accurate identification of water bodies.

[0109] Embodiment 6:

[0110] Based on Example 1, the embodiment of the present invention provides a method for extracting water under complex terrain conditions in mountainous areas based on multi-source data fusion. In S3, the slope and shadow of the DEM data are analyzed, and the mountain shadow in the DEM data is removed to obtain a third data image, including:

[0111] Obtain slope characteristics and shadow characteristics from DEM data;

[0112] Based on the preset slope characteristic range and shadow characteristic range of the mountain shadow, combined with the slope characteristics and shadow characteristics in the DEM data, the mountain shadow position is determined;

[0113] The slope features and shadow features of the mountain shadow position are removed to obtain a third data image.

[0114] The beneficial effect of the above design scheme is: by analyzing the slope and shadow of the DEM data, the mountain shadow in the DEM data is removed to obtain a third data image, thereby achieving the purpose of eliminating the influence of the mountain shadow on water body extraction.

[0115] Embodiment 7:

[0116] Based on Example 1, the embodiment of the present invention provides a method for extracting water under complex terrain conditions in a mountainous area based on multi-source data fusion. In S4, multi-source data fusion is performed on the first data image, the second data image and the third data image, including:

[0117] Acquire at least one characteristic attribute of the first data image, the second data image, and the third data image, and acquire the characteristic similarity under the characteristic attribute, and acquire the data image matching result when the characteristic similarity is greater than a preset similarity, and determine the fusion position matching feature between the first data image, the second data image, and the third data image based on the matching result;

[0118] Acquire the paddy field removal area in the first data image, and mark the paddy field for the corresponding position areas in the second data image and the third data image based on the position matching feature;

[0119] Based on the hill shadow removal area in the third data image, and based on the position matching feature, the hill shadow marking is performed on the corresponding position area in the second data image and the third data image;

[0120] Acquire false color features of local objects in the second data image, obtain local object features based on the false color features, and mark local object features at corresponding positions of the first data image and the second data image based on the position matching features;

[0121] Based on the above three times of labeling results, a first labeled data image, a second labeled data image and a third labeled data image are obtained, and the same type of data features in the first labeled data image, the second labeled data image and the third labeled data image are obtained as important data features, and different types of data features are obtained as first basic data features, second basic data features and third basic data features;

[0122] The first basic data feature, the second basic data feature, and the third basic data feature are respectively fused as hierarchical features to obtain intermediate fusion features, and the important data features are secondary fused with the intermediate fusion features to obtain a fused image.

[0123] The beneficial effect of the above design scheme is: by fusing the first basic data feature, the second basic data feature, and the third basic data feature as hierarchical features respectively, an intermediate fusion feature is obtained, and the important data feature is secondary fused with the intermediate fusion feature to obtain a fused image, ensuring that the obtained fused image can comprehensively and accurately represent each feature, and focus on representing important features, thereby providing accurate training samples for model training.

[0124] Embodiment 8:

[0125] Based on Example 1, the embodiment of the present invention provides a method for extracting water under complex terrain conditions in mountainous areas based on multi-source data fusion. In S4, the obtained fused image is used as a training sample to establish a binary classification model, including:

[0126] Based on the training samples, the support vector machine classification method is used to calculate the optimal hyperplane;

[0127] Based on the optimal hyperplane, a binary classification model is established.

[0128] In this embodiment, the scheme uses a kernel function to map samples from low-dimensional space to high-dimensional space to find the optimal classification hyperplane, find the points closest to the separation hyperplane, that is, the support vectors, and ensure that they are as far away from the separation plane as possible. The algorithm is based on the principle of structural risk minimization. First, an optimal hyperplane is constructed in the feature space, and then the global optimal solution is solved, that is, the expected fit of a certain probability upper bound in the entire sample space, and all points in the classification data set are separated as much as possible under the constraints of the classification, and the points are as far away from the hyperplane as possible.

[0129] The beneficial effect of the above design scheme is: by using the support vector machine classification method based on training samples, the optimal hyperplane is calculated, and based on the optimal hyperplane, a binary classification model is established to provide a model basis for water body identification and extraction.

[0130] Embodiment 9:

[0131] Based on Example 8, the embodiment of the present invention provides a method for extracting water under complex terrain conditions in mountainous areas based on multi-source data fusion. After establishing a binary classification model based on the optimal hyperplane, the method further includes:

[0132] Acquire a verification sample, verify the binary classification model based on the verification sample, and establish a confusion matrix based on the verification output result;

[0133] Based on the confusion matrix, the overall accuracy and Kappa coefficient are obtained, and based on the preset accuracy requirements, the model accuracy of the binary classification model is determined to meet the preset requirements;

[0134] If so, it is determined that the binary classification model has passed the verification;

[0135] Otherwise, it is determined that the binary classification model fails the validation and is retrained.

[0136] In this embodiment, the confusion matrix is ​​a two-dimensional matrix in which rows represent actual categories and columns represent predicted categories. Each element in the matrix represents the number of samples between the actual category and the predicted category. The confusion matrix can be used to calculate other accuracy evaluation indicators. Overall accuracy: The overall accuracy is the number of correctly classified samples divided by the total number of samples. It measures the average performance of the classifier over all categories. Kappa coefficient: The Kappa coefficient is an indicator that measures the difference between a classifier and a random classifier. Its value is between -1 and 1. The closer the value is to 1, the better the performance of the classifier.

[0137] The beneficial effect of the above design scheme is: by obtaining the overall accuracy and Kappa coefficient based on the confusion matrix, and judging whether the model accuracy of the binary classification model meets the preset requirements based on the preset accuracy requirements, if so, it is determined that the binary classification model verification has passed, otherwise, it is determined that the binary classification model verification has failed, and retraining is performed to ensure the model accuracy of the binary classification model.

[0138] Embodiment 10:

[0139] Based on Example 1, the embodiment of the present invention provides a method for extracting water bodies under complex terrain conditions in mountainous areas based on multi-source data fusion. In S4, subsequent water body extraction is performed based on a binary classification model, including:

[0140] Acquire real-time radar data and real-time DEM data, and process the real-time radar data and real-time DEM data according to steps S1-S3, and fuse the processed real-time data according to step S4 to obtain target real-time data;

[0141] The target real-time data is input into the binary classification model, the location of the water body area is determined based on the classification result, and the water body is extracted based on the location of the water body area.

[0142] The beneficial effect of the above design scheme is: by acquiring real-time radar data and real-time DEM data, and processing the real-time radar data and real-time DEM data according to steps S1-S3, the real-time data after data processing is fused according to step S4 to obtain target real-time data, the target real-time data is input into the binary classification model, the water body area position is determined based on the classification result, and the water body is extracted based on the water body area position, so as to realize accurate recognition and extraction of water body characteristics.

[0143] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention belong to the scope of this application document and its equivalent technology, the present invention is also intended to include these changes and variations.

Claims

1. A method for extracting water bodies in complex mountainous terrain conditions based on multi-source data fusion, characterized in that: include: S1: determining a comprehensive radar data image obtained by fusion of synthetic aperture radar images according to terrain distribution and time sequence, and preprocessing the comprehensive radar data image to obtain an initial data image; S2: establishing a gray level co-occurrence matrix of the initial data image, obtaining an image texture feature index value based on the gray level co-occurrence matrix, removing paddy field features in the initial data image based on the texture feature index value, obtaining a first data image, performing dual polarization fusion and false color synthesis on the initial data image, obtaining a second data image; S3: Analyze the slope and shadow of the DEM data, remove the mountain shadow in the DEM data, and obtain the third data image; S4: performing multi-source data fusion on the first data image, the second data image and the third data image, using the obtained fused image as a training sample, establishing a binary classification model, and performing subsequent water body extraction based on the binary classification model; In S4, multi-source data fusion is performed on the first data image, the second data image and the third data image, including: Acquire at least one characteristic attribute of the first data image, the second data image, and the third data image, and acquire the characteristic similarity under the characteristic attribute, and acquire the data image matching result when the characteristic similarity is greater than a preset similarity, and determine the fusion position matching feature between the first data image, the second data image, and the third data image based on the matching result; Acquire the paddy field removal area in the first data image, and mark the paddy field for the corresponding position areas in the second data image and the third data image based on the position matching feature; Based on the hill shadow removal area in the third data image, and based on the position matching feature, the hill shadow marking is performed on the corresponding position areas in the second data image and the third data image; Acquire false color features of local objects in the second data image, obtain local object features based on the false color features, and mark local object features at corresponding positions of the first data image and the second data image based on the position matching features; Based on the marking results of paddy field marking, mountain shadow marking and local feature marking, a first marked data image, a second marked data image and a third marked data image are obtained respectively, and the same type of data features in the first marked data image, the second marked data image and the third marked data image are obtained as important data features, and different types of data features are obtained as first basic data features, second basic data features and third basic data features; The first basic data feature, the second basic data feature, and the third basic data feature are respectively fused as hierarchical features to obtain intermediate fusion features, and the important data features are secondary fused with the intermediate fusion features to obtain a fused image.

2. The method for extracting water bodies in complex mountainous terrain based on multi-source data fusion according to claim 1 is characterized in that: In S1, the comprehensive radar data image obtained by fusion of synthetic aperture radar images is determined according to the terrain distribution and time sequence, including: Determine the lateral fusion characteristics of synthetic aperture radar images based on terrain distribution; Based on the time sequence, determine the longitudinal fusion characteristics of synthetic aperture radar images; Based on the horizontal fusion features and the vertical fusion features, the synthetic aperture radar images are fused to obtain a comprehensive radar data image.

3. The method for extracting water bodies in complex mountainous terrain based on multi-source data fusion according to claim 1 is characterized in that: In S1, the comprehensive radar data image is preprocessed to obtain an initial data image, including: Performing multi-view processing and image filtering on the comprehensive radar data image to obtain a denoised data image; The denoised data image is geocoded, radiometrically calibrated and decibelized to obtain an initial data image.

4. The method for extracting water bodies in complex mountainous terrain based on multi-source data fusion according to claim 1, characterized in that: In S2, a gray level co-occurrence matrix of the initial data image is established, an image texture feature index value is obtained based on the gray level co-occurrence matrix, and paddy field features in the initial data image are removed based on the texture feature index value to obtain a first data image, including: Obtaining a gray level co-occurrence matrix from the initial data image; Based on the gray level co-occurrence matrix, the image texture feature index value of the gray level co-occurrence matrix is ​​calculated according to the following formula; in, Represents the uniform index value of the image texture feature, The angular second-order moment index value representing the image texture feature, The entropy index value representing the image texture feature, represents the gray level, It represents the number of times the pixel pairs with horizontal gray level i and vertical gray level j appear in the gray level co-occurrence matrix. Indicates that the gray level co-occurrence matrix is ​​normalized; Based on the preset paddy field texture features and in combination with the image texture feature index value of the gray level co-occurrence matrix, the paddy field features in the initial data image are determined and removed to obtain the first data image.

5. The method for extracting water bodies in complex mountainous terrain based on multi-source data fusion according to claim 1, characterized in that: In S2, dual-polarization fusion and false color synthesis are performed on the initial data image to obtain a second data image, including: Acquire a radar echo signal of a synthetic aperture radar image, perform polarization decomposition on the radar echo signal to obtain an HH component and an HV component, and perform dual-polarization fusion on the initial data image according to the following formula to obtain an intermediate data image; in, Indicates the dual-polarization fusion process of the initial data image. represents the HH component, and HV represents the HV component; Perform false color synthesis on the intermediate data image to highlight its features, so as to obtain a second data image.

6. The method for extracting water bodies in complex mountainous terrain based on multi-source data fusion according to claim 1, characterized in that: In S3, the slope and shadow of the DEM data are analyzed, and the mountain shadow in the DEM data is removed to obtain a third data image, including: Obtain slope characteristics and shadow characteristics from DEM data; Based on the preset slope characteristic range and shadow characteristic range of the mountain shadow, combined with the slope characteristics and shadow characteristics in the DEM data, the mountain shadow position is determined; The slope features and shadow features of the mountain shadow position are removed to obtain a third data image.

7. The method for extracting water bodies in complex mountainous terrain based on multi-source data fusion according to claim 1, characterized in that: In S4, the obtained fused image is used as a training sample to establish a binary classification model, including: Based on the training samples, the support vector machine classification method is used to calculate the optimal hyperplane; Based on the optimal hyperplane, a binary classification model is established.

8. The method for extracting water bodies in complex mountainous terrain based on multi-source data fusion according to claim 7, characterized in that: After establishing a binary classification model based on the optimal hyperplane, the following steps are also included: Acquire a verification sample, verify the binary classification model based on the verification sample, and establish a confusion matrix based on the verification output result; Based on the confusion matrix, the overall accuracy and Kappa coefficient are obtained, and based on the preset accuracy requirements, the model accuracy of the binary classification model is determined to meet the preset requirements; If yes, it is confirmed that the binary classification model has passed the verification; Otherwise, it is determined that the binary classification model fails the validation and is retrained.

9. The method for extracting water bodies in complex mountainous terrain based on multi-source data fusion according to claim 1, characterized in that: In S4, subsequent water body extraction is performed based on a binary classification model, including: Acquire real-time radar data and real-time DEM data, and process the real-time radar data and real-time DEM data according to steps S1-S3, and fuse the processed real-time data according to step S4 to obtain target real-time data; The target real-time data is input into the binary classification model, the location of the water body area is determined based on the classification result, and the water body is extracted based on the location of the water body area.

Images