Irrigation Canal System Extraction Method and System Based on Satellite Images

Abstract

The present invention discloses a method and system for extracting irrigation canal systems based on satellite images. The irrigation canal system extraction method includes: providing a first and a second satellite image, where the first satellite image is a synthetic aperture radar image and the second satellite image is an optical image; segmenting the canal system using the gray-scale segmentation method, and eliminating mis-segmentation in combination with elevation data to obtain the first canal system distribution; extracting the second canal system distribution using the maximum entropy model; and combining with the second satellite image for verification to obtain the final irrigation canal system. The technical solution provided by the present invention is based on the HH and HV single-band polarization images obtained from the dual-polarization image data of the synthetic aperture radar image. Using the gray-scale segmentation method, the part mis-extracted as the canal system is eliminated in combination with elevation data; and the maximum entropy model is used to effectively extract the irrigation canal system; finally, in combination with the interpretation result of the optical satellite image, the accuracy of the extraction results of the two irrigation canal systems is verified, and finally the high-precision irrigation canal system distribution is obtained.

Description

Technical Field

[0001] The present invention relates to the technical field of satellite remote sensing, and in particular to a method and system for extracting irrigation canal systems based on satellite images. Background Art[[ID=~]]

[0002] Irrigation canal systems play an important role in the growth of agricultural crops and agricultural development. Traditional canal system monitoring mainly relies on unmanned aerial vehicle technology for identification and extraction. Although it has high accuracy, it is time-consuming, laborious and costly, and cannot achieve all-weather real-time monitoring.

[0003] In the field of remote sensing monitoring, satellite remote sensing has more obvious advantages. For example, the Gaofen-3 (GF-3) satellite has the ability to observe the ground all day and all weather. Currently, the Gaofen-3 satellite has conducted more research on water body identification, and the influence of mountain shadows is relatively serious. There is little research on the identification of irrigation canal systems. And the currently commonly used algorithms mainly include threshold method, supervised classification and machine learning, etc., which are mostly used for the extraction of multi-band remote sensing image information and cannot effectively extract and utilize single-band image information.

[0004] Therefore, there is an urgent need to develop a method and system that can use the Gaofen-3 or similar synthetic aperture radar satellites to achieve high-precision and high-accuracy monitoring of irrigation canal systems, especially those in mountainous or hilly areas. Summary of the Invention

[0005] Aiming at the deficiencies of the prior art, the purpose of the present invention is to provide a method and system for extracting irrigation canal systems based on satellite images.

[0006] To achieve the foregoing invention purpose, the technical solutions adopted by the present invention include:

[0007] In a first aspect, the present invention provides a method for extracting an irrigation canal system based on satellite images, which includes:

[0008] Providing a first satellite image and a second satellite image, wherein the first satellite image is a synthetic aperture radar image and the second satellite image is an optical image;

[0009] Using the gray-scale segmentation method to segment the canal system from the first satellite image, and combining the elevation data of the corresponding geographical location to eliminate the mis-segmented canal system to obtain the first canal system distribution;

[0010] Using the maximum entropy model, introducing elevation factor, slope factor and aspect factor, to extract the second canal system distribution from the first satellite image;

[0011] Combining the interpretation result of the second satellite image, verifying the accuracy of the first canal system distribution and the second canal system distribution to obtain a verification result, and based on the verification result, screening the one with higher accuracy as the final irrigation canal system.

[0012] In a second aspect, the present invention further provides an irrigation canal system extraction system based on satellite images, which includes:

[0013] An image acquisition module for providing a first satellite image and a second satellite image, where the first satellite image is a synthetic aperture radar image and the second satellite image is an optical image;

[0014] A first canal system module for segmenting the canal system from the first satellite image using the gray-scale segmentation method and removing the mis-segmented canal system in combination with the elevation data of the corresponding geographical location to obtain a first canal system distribution;

[0015] A second canal system module for extracting a second canal system distribution from the first satellite image using the maximum entropy model and introducing elevation factors, slope factors, and aspect factors;

[0016] An accuracy verification module for verifying the accuracy of the first canal system distribution and the second canal system distribution in combination with the interpretation result of the second satellite image to obtain a verification result, and screening the one with higher accuracy based on the verification result to obtain the final irrigation canal system.

[0017] Based on the above technical solutions, compared with the prior art, the beneficial effects of the present invention at least include:

[0018] The technical solution provided by the present invention is based on the HH and HV single-band polarization images obtained from the dual-polarization image data of the synthetic aperture radar image. Using the gray-scale segmentation method, the part mis-extracted as the canal system is removed in combination with the elevation data; and using the maximum entropy model, factors such as elevation, slope, and aspect are introduced to reduce the influence of mountain shadows, effectively extracting the irrigation canal system; finally, in combination with the interpretation result of the optical satellite image, the accuracy of the extraction results of the two irrigation canal systems is verified, and finally a high-precision irrigation canal system distribution is obtained.

[0019] The above description is only an overview of the technical solution of the present invention. In order to enable those skilled in the art to more clearly understand the technical means of the present application and implement it in accordance with the content of the specification, the following takes a preferred embodiment of the present invention and combines detailed drawings for description as follows. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1a is an elevation map of the terrain factors provided by a typical embodiment of the present invention;

[0021] Figure 1b is a slope map of the terrain factors provided by a typical embodiment of the present invention;

[0022] Figure 1c is an aspect map of the terrain factors provided by a typical embodiment of the present invention;

[0023] Figure 2a It is the GF-3 HH image provided by a typical embodiment of the present invention;

[0024] Figure 2b It is the GF-3 HV image provided by a typical embodiment of the present invention;

[0025] Figure 3 It is the final irrigation canal system map extracted from a typical embodiment of the present invention;

[0026] Figure 4a It is the image before gray-scale segmentation provided by a typical embodiment of the present invention;

[0027] Figure 4b It is the image after gray-scale segmentation provided by a typical embodiment of the present invention;

[0028] Figure 5 It is the maximum entropy model prediction accuracy map provided by a typical embodiment of the present invention;

[0029] Figure 6 It is the feature factor importance ranking map obtained by the maximum entropy model provided by a typical embodiment of the present invention;

[0030] Figure 7 It is the true color image map of GF-2 provided by a typical embodiment of the present invention;

[0031] Figure 8a It is the GF-3 fine strip 2 (FSII) HH polarized image provided by a typical practical case of the present invention;

[0032] Figure 8b It is the GF-3 fine strip 2 (FSII) HV polarized image provided by a typical practical case of the present invention;

[0033] Figure 9a It is the HH polarized image obtained after post-processing of multi-look, filtering, geocoding, radiometric calibration and cropping provided by a typical practical case of the present invention;

[0034] Figure 9b It is the HH polarized image obtained after post-processing of multi-look, filtering, geocoding, radiometric calibration and cropping provided by a typical practical case of the present invention;

[0035] Figure 10 It is the first canal system distribution example map provided by a typical practical case of the present invention;

[0036] Figure 11 It is the second canal system distribution example map provided by a typical practical case of the present invention;

[0037] Figure 12 It is the GF-2 verification point distribution map provided by a typical practical case of the present invention. Detailed implementation manners

[0038] In view of the deficiencies in the prior art, through long-term research and a large number of practices, the inventors of this case have been able to propose the technical solution of the present invention. The following will further explain the technical solution, its implementation process, principles, etc.

[0039] In the following description, many specific details are set forth in order to provide a thorough understanding of the present invention. However, the present invention may be implemented in other ways different from those described herein. Therefore, the scope of protection of the present invention is not limited by the specific embodiments disclosed below.

[0040] Moreover, relational terms such as "first" and "second" are only used to distinguish one component or method step with the same name from another, and do not necessarily require or imply any actual relationship or order between these components or method steps.

[0041] An embodiment of the present invention provides an extraction method for irrigation canal systems based on satellite images, which includes the following steps:

[0042] Provide a first satellite image and a second satellite image, where the first satellite image is a synthetic aperture radar image, and the second satellite image is an optical image;

[0043] Use the gray-scale segmentation method to segment the canal systems from the first satellite image, and combine the elevation data of the corresponding geographical location to eliminate the mis-segmented canal systems to obtain the first canal system distribution; [[ID=2⃣0]]

[0044] Use the maximum entropy model, introduce the elevation factor, slope factor, and aspect factor, and extract the second canal system distribution from the first satellite image;

[0045] Combine the interpretation result of the second satellite image to verify the accuracy of the first canal system distribution and the second canal system distribution to obtain a verification result, and based on the verification result, select the one with higher accuracy to obtain the final irrigation canal system.

[0046] In the above technical solution, the first satellite image is, for example, an image obtained by the Gaofen-3 satellite, and the second satellite image is, for example, an image obtained by the Gaofen-2 satellite. Of course, within the scope of the technical solution of the present invention, those skilled in the art can adaptively select a suitable satellite to obtain the corresponding image, for example, replace it with a satellite at home and abroad with the same imaging method and approximate imaging accuracy.

[0047] Of course, any form of acquisition method for the final canal system distribution, as long as it refers to the verification result based on the second satellite image, falls within the scope of implementation of the technical solution provided by the present invention.

[0048] As some typical application examples of the above technical solutions, the above process can be implemented through the following specific procedures:

[0049] (1) Preprocessing of GF-3 data; (2) Preprocessing of GF-2 data; (3) Acquisition of terrain factors; (4) Extraction of irrigation canal systems by the gray-scale segmentation method; (5) Extraction of irrigation canal systems based on the maximum entropy model and obtaining high-precision extraction results of irrigation canal systems.

[0050] The numbers of the above steps are only for marking purposes and do not limit the implementation order to be necessarily in the above number order. Of course, the implementation order of the technical solutions in the present invention is implemented according to their logical relationships. The implementation of some steps (such as the above preprocessing and terrain factor acquisition steps) can be carried out sequentially, in reverse order, or synchronously, and is not limited to the specific time sequence.

[0051] The present invention adopts the combined means of gray-scale segmentation and maximum entropy model extraction. Since the gray-scale segmentation method does not require complex calculations on the image, it has a fast calculation speed and is not affected by interference factors such as illumination and clouds and can be preferably used in mountain areas. Although it can effectively extract irrigation canal systems, it does not consider the shape, texture and other features of the extraction target. As a non-parametric density estimation method, the maximum entropy model can not only consider multiple factors including texture, shape, etc., but also perform weighted processing on the importance of different feature factors, effectively avoiding the influence of some noise points and interference points on the extraction of irrigation canal systems. Considering complex environmental areas such as mountains, the gray-scale segmentation method and the maximum entropy model are used to extract irrigation canal systems, and finally, combined with the verification of the second satellite image, the final high-precision and high-accuracy irrigation canal system results are obtained.

[0052] Regarding the data preprocessing in the above step (1), in some implementation schemes, the first satellite image is the HH and HV single-band polarization images obtained after preprocessing the remote sensing images taken by the first satellite.

[0053] In some implementation schemes, the preprocessing includes multi-look processing and filtering processing.

[0054] Specifically, the purpose of multi-look processing is to make the image have a better visual effect, increase the estimation accuracy of the backscattering of each pixel to a certain extent, and average a single sample of the relevant target to achieve the purpose of multi-look processing, making the texture features of the image closer to the actual situation on the ground and also reducing the noise. If the multi-look parameters are set too large, it will lead to a reduction in spatial resolution and is not conducive to visual interpretation. In actual applications, appropriate parameters should be set, and the specific parameter settings can be appropriately adjusted based on experimental results and will not be elaborated here.

[0055] Regarding the filtering process, the specific implementation case of the present invention uses Frost filtering. The reason is that Frost filtering can reduce speckles and noise while well preserving the information of edge boundaries in SAR (Synthetic Aperture Radar) images. Of course, other forms of filtering can also be used. Multi-looking and filtering processes are common operations in the field for processing satellite images.

[0056] Frost filtering mainly performs filtering by calculating the minimum mean square error for the target pixel point through neighborhood information, which can largely preserve the boundary information of the irrigation canal system and is beneficial for subsequent analysis of the extraction of the irrigation canal system.

[0057] Specifically, for example, applying Frost filtering in the SARscape 5.6.2 software, the default window is 5*5 (the larger the window setting, the smoother the filtering effect but the longer the required time), and the filtered result is output. The involved calculation formula is as follows:

[0058]

[0059] The estimated value of a certain pixel point in the image is the weighted average of all pixel values within a certain window in the noisy image. Among them, (x, y) is the coordinate of the target denoising pixel point, and i and j are the offsets of (x, y) within the window range. Both m(x + i, y + j) and I(x + i, y + j) are the weighted values of pixel values, which decrease with the increase of distance.

[0060] The weight m(x + i, y + j) can be calculated by the following formula:

[0061] m(x + i, y + j) = K2αexp[-α|t|]

[0062]

[0063]

[0064]

[0065] Among them, K represents a constant, is the coefficient of variation in the image domain of this window.

[0066] Regarding the remaining processing process in the preprocessing, in some implementation schemes, the extraction method may specifically include:

[0067] Providing a digital elevation model.

[0068] Geocoding the remotely sensed image that has undergone multi-looking processing and filtering processing based on the digital elevation model to obtain an elevation composite image.

[0069] Radiometrically calibrate the elevation composite image, convert the intensity of the backscattered energy into the backscattering coefficient, and obtain the HH and HV single-band polarization images.

[0070] Participate Figure 1a - Figure 1c , as some typical application examples of the above technical solutions, this embodiment is based on a high-precision digital elevation model (Digital Elevation Model, DEM) with a spatial resolution of 12.5 m, and geocodes the SAR image provided by the first satellite. First, it is transferred from the slant range coordinate system to the geographic coordinate system. Secondly, through radiometric calibration, the intensity of the backscattered energy can be converted into the backscattering coefficient for comparison. Since different shooting methods and time phases cannot be directly compared, radiometric calibration is to place them on the same level for comparison. Finally, after cropping, the HH and HV polarization images of the area to be extracted for the irrigation canal system as shown in Figure 2a And Figure 2b are obtained. Of course, if the image area itself is equal to the area to be extracted, cropping may not be necessary.

[0071] In some embodiments, the radiometric calibration is expressed as:

[0072]

[0073] Where F d is the backscattering intensity received by the sensor of the first satellite; P t is the transmission power; P n is the additional power; And are the transmission and receiving antenna gains respectively; is the receiving antenna gain; is the current gain of the radar receiver; G p is the processor constant; R 3 is the distance propagation loss; θ el is the antenna elevation angle; θ az is the antenna azimuth angle; L a is the loss of the atmosphere; L s is the loss of the system; S is the scattering area; σ O is the backscattering coefficient.

[0074] In practical applications, not only the first satellite image needs to be preprocessed, but the second satellite image also needs to be preprocessed. For example:

[0075] See Figure 7 As shown, the GF-2 remote sensing data is mainly preprocessed through multi-spectral image preprocessing and panchromatic image preprocessing, and finally image fusion is performed. It mainly includes data preprocessing such as orthorectification, image registration, atmospheric correction, and image fusion.

[0076] Of course, the specific preprocessing method can refer to the preprocessing methods for optical images in the prior art, and it is subject to being able to provide appropriate references and verification criteria for the following steps. The specific preprocessing method will not be elaborated here.

[0077] Regarding the above step (3), in some embodiments, the elevation factor, slope factor, and aspect factor are obtained from the digital elevation model.

[0078] In some embodiments, the calculation methods of the slope factor and aspect factor are as follows:

[0079]

[0080]

[0081] In the formula, T1 represents the slope, T2 represents the aspect, and h a and h b are respectively the elevation change rates in the north-south and east-west directions in the digital elevation model.

[0082] Specifically, according to the DEM with a spatial resolution of 12.5 m above, the corresponding slopes and aspects of the pixel points are extracted. And in this example, the DEM, slope, and aspect also need to be resampled to a spatial resolution of 10 m to be consistent with the HH and HV images as topographic factors. Of course, if the resolution of the DEM itself is suitable for the HH and HV images, this resampling step can be omitted.

[0083] In some embodiments, the method for extracting the canal system by using the gray-scale segmentation method to segment the canal system from the first satellite image and combining the elevation data of the corresponding geographical location to eliminate the mis-segmented canal system to obtain the first canal system distribution specifically includes the following sub-steps:

[0084] Set the gray-scale threshold and slope threshold.

[0085] Based on the gray-scale threshold, a preliminary segmentation image is segmented from the HH and HV single-band polarization images.

[0086] Eliminate the pixel points with slopes above the slope threshold in the preliminary segmentation image to obtain the first canal system distribution.

[0087] In specific implementation, corresponding to the above step (4), in ENVI software, using HH and HV images, through the RasterColors Slices module, create an irrigation canal system category, adjust the threshold range, observe the DN value statistical distribution histogram, and observe the overall DN value distribution of the image, and the proportion of the irrigation canal system in the histogram can be initially judged. The threshold segmentation method is used to extract the irrigation canal system, which has a good effect in the plain area and almost no manual editing is required. However, for some areas with more mountains, the mountain shadows are all extracted as the irrigation canal system. Subsequently, by analyzing the characteristics of the terrain elevation in this area, the mis-extracted areas above the corresponding slope are removed, and the final irrigation canal system area as shown in Figure 3 is obtained.

[0088] See Figure 4a and Figure 4b As shown, setting the gray-scale segmentation threshold as P, the gray-scale segmentation process can be represented by the following functional relationship:

[0089] P = P[x, y, h(x, y), t(x, y)]

[0090] In the formula, (x, y) represents the coordinates of the pixel, h(x, y) represents the local attribute of the pixel, and t(x, y) represents the gray-scale value of the pixel.

[0091] The image after thresholding can be expressed as:

[0092]

[0093] In the formula, 1 and 0 represent the target object (representing the irrigation canal system) and the non-target object respectively. If P is determined by t(x, y), the threshold is global; if P is jointly determined by h(x, y) and t(x, y), the threshold is local; when P is determined by (x, y), it is adaptive.

[0094] Regarding the above step (5), see Figure 5 , in some embodiments, in the extraction method, using the maximum entropy model, introducing elevation factor, slope factor and aspect factor, extracting the second canal system distribution from the first satellite image specifically includes the following sub-steps:

[0095] Provide multiple irrigation canal system points as its overall distribution data.

[0096] Specifically as shown in Figure 6 , combining the overall distribution data, comprehensively using the elevation factor, slope factor, aspect factor and HH and HV single-band polarization images, predicting the irrigation canal system distribution through the maximum entropy model, and obtaining the second canal system distribution. Figure 5Among them, the AUC values of the training set and the test set are 0.920 and 0.997 respectively, which are much larger than the random prediction value of 0.5, indicating that the prediction accuracy of the model is relatively high.

[0097] In some embodiments, in the extraction method, the interpretation results of the second satellite image are combined to verify the accuracy of the first canal system distribution and the second canal system distribution to obtain a verification result, and based on the verification result, the first canal system distribution and the second canal system distribution are screened and / or combined to obtain the final irrigation canal system, which specifically includes the following sub-steps:

[0098] Perform cross-validation on the first canal system distribution and the second canal system distribution to obtain a cross-prediction result.

[0099] Select some sample points from the interpretation results of the second satellite image as the true canal system labels.

[0100] Construct a confusion matrix with the cross-prediction result and the true canal system label, and use the confusion matrix to verify the accuracy of the cross-prediction result to obtain the final irrigation canal system.

[0101] In a specific implementation case, a total of 5 factors including terrain factors (DEM, slope, aspect) and HH, HV images are comprehensively used as environmental factors, and a certain number of irrigation canal system points are used as their overall distribution data. The irrigation canal system distribution is predicted by the maximum entropy model (MAXENT). The maximum entropy model finally selects the result with the highest accuracy through the ROC curve as the irrigation canal system distribution map, performs cross-validation with the irrigation canal system extracted by the gray-scale segmentation method, and selects a certain number of sample points through the high-resolution No. 2 interpretation image to verify the accuracy through the confusion matrix.

[0102] Regarding the maximum entropy model, it is a statistical model based on probability distribution. The training samples are (a1, b1), (a2, b2),... (an, bn), and the empirical distributions of the joint distribution and the marginal distribution are respectively:

[0103]

[0104]

[0105] In the formula, t(X = a, Y = b) is the number of times the training sample (a, b) appears simultaneously, t(X = a) is the number of times a appears in the training sample, and n represents the sample size.

[0106] Use the function f(a, b) to represent the relationship between the input a and the output b:

[0107]

[0108]

[0109]

[0110] In the formula, represents the expected value of the function f(a, b) with respect to the empirical distribution E p (f) represents the expected value of the function f(a, b) with respect to the model p(b|a) and the empirical distribution When is equal to E p (f), it is the constraint condition for the learning calculation of the maximum entropy model.

[0111] Suppose there are n equal to E p (f), and there are also n constraint conditions. Then the maximum entropy model can be expressed as:

[0112]

[0113] Regarding the subsequent cross-validation, the embodiment of the present invention performs cross-validation based on the maximum entropy model and the irrigation canal system extracted by the gray-scale segmentation method, selects a certain number of sample points according to the high-resolution No. 2 interpretation image, and verifies the accuracy of the extraction result through a confusion matrix (Kappa coefficient, overall accuracy) to obtain a high-precision irrigation canal system extraction result.

[0114] The confusion matrix is shown in the following table:

[0115] Category 1 Category 2 Category 1 a11 a12 Category 2 a21 a22

[0116] In the table, a11, a12, a21, and a22 are the confusion matrices calculated from the true class and the predicted class.

[0117] In some embodiments, the accuracy verification is performed based on the calculation of the Kappa coefficient and the overall accuracy.

[0118] In some embodiments, the calculation method of the accuracy verification is:

[0119]

[0120]

[0121] Among them, K is the Kappa coefficient, OA is the overall accuracy, N represents the total number of samples, and a ij represents the number of times i is predicted as j in the confusion matrix; a i+ = ∑ j a ij , a +j = ∑ i a ij .

[0122] After the accuracy verification, the distributions of the first canal system and the second canal system are screened according to the Kappa coefficient and the overall accuracy. The specific selection is as follows:

[0123] By overlaying the irrigation canal system extraction result with the second satellite image, several sample points are randomly selected. The confusion matrix of the true class and the predicted class is calculated through the position and attributes of the points, and the Kappa coefficient and the overall accuracy are calculated to verify the accuracy of the irrigation canal system extracted by the above two methods. The value of the Kappa coefficient ranges from 0 to 1, and it measures the accuracy of the classification result by comparing whether the classification result is consistent with the actual result. Finally, the one with higher accuracy is selected as the final result of the extracted irrigation canal system.

[0124] Corresponding to the above irrigation canal system extraction method, the second aspect of the embodiment of the present invention also provides an irrigation canal system extraction system based on satellite images, which includes the following modules:

[0125] An image acquisition module, configured to provide a first satellite image and a second satellite image, where the first satellite image is a synthetic aperture radar image and the second satellite image is an optical image.

[0126] A first canal system module, configured to segment the canal system from the first satellite image by using the gray-scale segmentation method, and eliminate the mis-segmented canal system by combining the elevation data of the corresponding geographical location to obtain the distribution of the first canal system.

[0127] A second canal system module, configured to extract the distribution of the second canal system from the first satellite image by using the maximum entropy model and introducing elevation factors, slope factors, and aspect factors.

[0128] An accuracy verification module, configured to verify the accuracy of the distribution of the first canal system and the distribution of the second canal system in combination with the interpretation result of the second satellite image to obtain a verification result, and select the one with higher accuracy based on the verification result to obtain the final irrigation canal system.

[0129] The embodiment of the present invention also provides a very specific actual case of extracting an irrigation canal system, where Figure 8a and Figure 8b respectively represent the original GF-3 fine strip 2 (FSII) HH(a) and HV(b) polarization images, which are synthetic aperture radar images of Ganzhou District, Zhangye City, China. After multi-look, filtering, geocoding, radiometric calibration, and cropping post-processing, the Figure 9a and Figure 9b shown HH(a) and HV(a) polarization images are obtained. The distribution of the first canal system is extracted by gray-scale segmentation and eliminating areas with a slope of or above the threshold as shown in Figure 10 shown, and then the distribution of the second canal system is extracted by using the maximum entropy model based on elevation, slope, aspect, and HH and HV polarization images as shown in Figure 11As shown Figure 12 The figure shows the distribution of accuracy verification points selected in GF-2. These points can be selected manually, according to a specific recognition model, or randomly or in an array. The following table shows the final accuracy verification results. It can be seen from the table that the overall accuracy and kappa coefficient of the irrigation canal system extracted by the maximum entropy model are higher than those of the gray-scale segmentation method. Therefore, the results extracted by the maximum entropy model are used as the final irrigation canal system.

[0130] Method Overall accuracy (%) Kappa coefficient Gray-scale segmentation 84.00 0.6758 Maximum entropy model 91.00 0.8187

[0131] Based on the manual recognition marks at the corresponding positions, in the above-mentioned implementation case, the overall accuracy of the sample points of the irrigation canal system is 91%, and the kappa coefficient is 0.8187. In contrast, using the existing gray-scale segmentation method, the overall accuracy of the sample points in the irrigation canal system is 84%, and the kappa coefficient is 0.8187, which is significantly lower than that of the above-mentioned implementation case. Moreover, there are missed extractions for dense irrigation canal systems, and the extraction results near the mountains are inaccurate.

[0132] Based on the above-mentioned embodiments, it can be clearly seen that the embodiments of the present invention are based on the HH and HV single-band polarization images obtained from the dual-polarization image data of GF-3 (or other equivalent synthetic aperture radar satellites): (1) Using the gray-scale segmentation method and combining elevation data to eliminate the parts mis-extracted as canal systems, obtaining the distribution area of the irrigation canal system; (2) Using the maximum entropy model and introducing factors such as elevation, slope, and aspect to reduce the influence of mountain shadows and effectively extract the irrigation canal system; (3) Combining the interpretation results of GF-2 (or other equivalent optical imaging satellites) to verify the accuracy of the extraction results of the two methods, and finally obtaining a high-precision distribution area of the irrigation canal system, filling the technical gap of this project, and being particularly suitable for the remote sensing extraction of irrigation canal systems in mountainous or hilly terrains.

[0133] It should be understood that the above-mentioned embodiments are only used to illustrate the technical concept and characteristics of the present invention, and their purpose is to enable those familiar with this technology to understand the content of the present invention and implement it accordingly, and cannot be used to limit the protection scope of the present invention. Any equivalent changes or modifications made according to the spirit and essence of the present invention should be covered within the protection scope of the present invention.

Claims

1. An irrigation canal system extraction method based on satellite images, characterized in that, Including: Providing a first satellite image and a second satellite image, where the first satellite image is a synthetic aperture radar image, which is an HH and HV single-band polarization image obtained by preprocessing the remote sensing image taken by a first satellite. The preprocessing includes multi-look processing and filtering processing, and the second satellite image is an optical image; The specific preprocessing includes: Providing a digital elevation model; Geocoding the remote sensing image that has undergone multi-look processing and filtering processing based on the digital elevation model to obtain an elevation composite image; Performing radiometric calibration on the elevation composite image to convert the intensity of the backscattered energy into a backscattering coefficient, obtaining the HH and HV single-band polarization images; Setting a gray threshold and a slope threshold; Based on the gray threshold, segmenting a preliminary segmentation image from the HH and HV single-band polarization images; Removing the pixel points in the preliminary segmentation image whose corresponding slopes are above the slope threshold to obtain a first canal system distribution; Providing multiple irrigation canal system points as its overall distribution data; Combining the overall distribution data, and predicting the irrigation canal system distribution through a maximum entropy model by integrating elevation factors, slope factors, aspect factors, and HH and HV single-band polarization images to obtain a second canal system distribution; Performing cross-validation on the first canal system distribution and the second canal system distribution to obtain a cross-prediction result; Selecting some sample points from the interpretation result of the second satellite image as the true canal system labels; Constructing a confusion matrix from the cross-prediction result and the true canal system labels, and using the confusion matrix to verify the accuracy of the cross-prediction result to obtain the final irrigation canal system; The accuracy verification is based on the calculation of the Kappa coefficient and the overall accuracy, and the calculation method of the accuracy verification is: ； where K is the Kappa coefficient, OA is the overall accuracy, and N represents the total number of samples. represents the number of times that i is predicted as j in the confusion matrix. , ; Based on the verification result, screening out the one with high accuracy as the final irrigation canal system.

2. The irrigation canal system extraction method according to claim 1, wherein The radiometric calibration is expressed as: ； wherein, is the backscattering intensity received by the sensor of the first satellite; is the transmission power; is the additional power; and are the transmission and reception antenna gains respectively; is the current gain of the radar receiver; is the processor constant; is the distance propagation loss; is the antenna elevation angle; is the antenna azimuth angle; is the loss of the atmosphere; is the loss of the system; is the scattering area; is the backscattering coefficient.

3. The irrigation canal system extraction method according to claim 1, characterized in that The elevation factor, slope factor, and aspect factor are obtained from the digital elevation model; The calculation methods of the slope factor and the aspect factor are: ； In the formula, represents the slope, represents the aspect, and are respectively the elevation change rates in the north-south and east-west directions in the digital elevation model.

4. An irrigation canal system extraction system based on satellite images, which is used to execute the irrigation canal system extraction method described in any one of claims 1-3, and is characterized in that, Including: An image acquisition module for providing a first satellite image and a second satellite image, where the first satellite image is a synthetic aperture radar image and the second satellite image is an optical image; A first canal system module for segmenting the canal system from the first satellite image using the gray-scale segmentation method, and removing the mis-segmented canal system by combining the elevation data of the corresponding geographical location to obtain a first canal system distribution; A second canal system module for using the maximum entropy model and introducing elevation factors, slope factors, and aspect factors to extract a second canal system distribution from the first satellite image; An accuracy verification module for combining the interpretation result of the second satellite image to verify the accuracy of the first canal system distribution and the second canal system distribution to obtain a verification result, and screening out the one with higher accuracy based on the verification result to obtain the final irrigation canal system.

Images