A flood warning method for the lower reaches of the Yellow River

Abstract

The present invention provides a flood warning method for the lower reaches of the Yellow River, which is used to solve the technical problem of incomplete identification of water body information, resulting in low accuracy of flood warning for the lower reaches of the Yellow River. The steps of this method are as follows: First, collect remote sensing image data of different bands and superimpose them to obtain the superimposed remote sensing data; Second, mark the position of the middle reaches of the Yellow River, divide the local areas, and perform clustering on the local areas to obtain the Yellow River water body identification results; Then, according to the calculated water body identification rate of each local area, obtain the optimal band; Finally, reclassify the remote sensing data according to the optimal band to obtain the final Yellow River water body identification result; And use the final Yellow River water body identification result as warning data to achieve flood warning for the lower reaches of the Yellow River. In different geological regions of the Yellow River, the present invention can obtain good Yellow River water body identification results, improve the identification accuracy of the water body scale in the middle reaches of the Yellow River, and thus improve the accuracy of flood warning for the lower reaches of the Yellow River.

Description

Technical Field

[0001] The present invention relates to the technical field of image data processing, and particularly to a flood warning method for the lower reaches of the Yellow River. Background Art

[0002] The outbreak of floods will not only cause economic losses to society, but also result in casualties. Therefore, flood warnings can be issued in advance to the downstream areas of floods, thereby reducing the economic losses and casualties caused by floods. The floods in the lower reaches of the Yellow River mainly come from the middle reaches of the Yellow River. The scale of water bodies in the middle reaches of the Yellow River can be identified through remote sensing images, so as to realize flood warnings for the lower reaches of the Yellow River.

[0003] When a satellite collects remote sensing images, it conducts remote sensing imaging by collecting spectral information in different bands. Different bands of remote sensing information have different spectral value responses to water bodies. As a result, when using remote sensing images to identify the Yellow River water bodies, if the selected band range is not good, the extracted water body information will be incomplete. And due to the wide span of the Yellow River, different band ranges are adopted in different geological regions, resulting in different identification effects for the Yellow River water bodies. Summary of the Invention

[0004] In view of the deficiencies in the above-mentioned background art, the present invention proposes a flood warning method for the lower reaches of the Yellow River, which solves the technical problem of incomplete identification of water body information and low accuracy of flood warning for the lower reaches of the Yellow River.

[0005] The technical solution of the present invention is realized as follows:

[0006] A flood warning method for the lower reaches of the Yellow River, the steps are as follows:

[0007] Step 1: Collect remote sensing image data in different bands, and select the remote sensing image data in the red, green, and blue bands for superposition to obtain the superimposed remote sensing data; the remote sensing image data refers to the spectral values of ground object information.

[0008] Step 2: Use manual annotation to obtain the position of the middle reaches of the Yellow River in the superimposed remote sensing data; and divide the image of the superimposed remote sensing data into water body regions along the direction perpendicular to the extension of the Yellow River water body to obtain M local regions;

[0009] Step 3: Perform clustering on each local region to obtain K categories, and map the position of the middle reaches of the Yellow River to the categories in the K categories to obtain the Yellow River water body identification results of each local region;

[0010] Step 4: Calculate the water body smoothness of the water body categories in each local region, and combine the water body smoothness with the difference of the remote sensing data in the middle of all categories in the neighborhood to obtain the water body identification rate in each local region;

[0011] The calculation method of the water body smoothness is as follows: obtain the water body category g in the current i-th local area according to the Yellow River water body recognition results of each local area i ; use the edge detection algorithm to process the water body category g in the current i-th local area i to obtain the edge pixel points of the water body category g i ; use the method of polynomial fitting to fit the coordinates of the edge pixel points of the water body category g i to obtain the corresponding fitting function; take the first derivative of the fitting function to obtain the slope values corresponding to each coordinate point of the edge pixel points on the fitting function in the water body category g i ; obtain the variance value of all current slope values as the water body smoothness S i of the water body category g in the current i-th local area i .

[0012] The calculation method of the water body recognition rate is as follows:

[0013] L i = exp(-a*S i )*C i ;

[0014] where L i is the water body recognition rate in the i-th local area, C i is the average difference between the remote sensing data in the category where the Yellow River water body recognition result is located in the i-th local area and the remote sensing data in the neighboring category, a is a hyperparameter, and exp() represents the exponential function with the natural constant e as the base

[0015] The method for obtaining the average difference C i between the remote sensing data in the category where the Yellow River water body recognition result is located in the i-th local area and the remote sensing data in the neighboring category is as follows:

[0016] Calculate the mean H i of the remote sensing data in the category where the Yellow River water body recognition result is located in the i-th local area and the mean H ij of the remote sensing data in the j-th neighboring category of the i-th local area, and calculate the absolute value of the difference H' ij between H i and H ij ;

[0017] Calculate the difference degree value Y ij = E ij *H' ij ; where E ij is the weight

[0018] C is obtained by calculating the average of the difference degree values between all neighborhood classes and the class to which the i-th local region belongs i 。

[0019] Step Five: Obtain the optimal band according to the water body recognition rate in each local region;

[0020] The method for obtaining the optimal band according to the water body recognition rate in each local region is as follows: Set a threshold r. When the water body recognition rate of the i-th local region is less than the threshold r, while ensuring that the clustering result in the i-th local region remains unchanged, recombine the remote sensing data of different bands in the i-th local region, and calculate C i value of the remote sensing data under all combinations in the i-th local region; Select the band of the combination corresponding to the maximum Ci value as the optimal band.

[0021] Step Six: Reclassify and judge the remote sensing data corresponding to the optimal band to obtain the final Yellow River water body recognition result;

[0022] The method for reclassifying and judging the remote sensing data corresponding to the optimal band to obtain the final Yellow River water body recognition result is as follows:

[0023] S6.1. Reclassify the remote sensing data in the i-th local region corresponding to the optimal band to obtain the classification result of the i-th local region;

[0024] S6.2. Calculate the intersection-over-union ratio of the reclassification result of the i-th local region and the original water body category of the i-th local region respectively; And screen out the maximum intersection-over-union ratio to obtain the Yellow River water body recognition result in the reclassification result of the i-th local region;

[0025] S6.3. If the recognition rate of the Yellow River water body recognition result corresponding to the optimal band is less than the recognition rate threshold, use the particle swarm optimization algorithm to solve the weight of each band in the i-th local region, and weight the remote sensing data of each band to obtain the weighted superimposed remote sensing data;

[0026] The fitness function corresponding to the particle swarm optimization algorithm is:

[0027] M = exp(-L′ i )

[0028] where M is the fitness value, and L′ i is the water body recognition rate corresponding to the weighted remote sensing data of each band in the i-th local region.

[0029] S6.4. Reclassify the weighted superimposed remote sensing data according to the methods in steps S6.1 - S6.2, and obtain the final Yellow River water body recognition result through the intersection-over-union ratio.

[0030] Step 7: Use the final Yellow River water body identification result as the data for early warning and conduct flood early warning for the lower reaches of the Yellow River.

[0031] The method for judging whether to conduct flood early warning for the lower reaches of the Yellow River according to the size of the water body area is as follows: Set an area threshold t for the water body area in different regions. When the water body area in a local region is greater than the preset area threshold t in the corresponding region, it is considered that the water body in this local region is too large, which will cause flood disasters in the local area of the lower reaches of the Yellow River. Then, conduct downstream regional flood early warning for the local regions after the local region corresponding to the area threshold t.

[0032] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention can achieve good Yellow River water body identification results in different geological regions of the Yellow River, improve the identification accuracy of the water body scale in the middle reaches of the Yellow River, and thus improve the accuracy of flood early warning in the lower reaches of the Yellow River. Brief Description of the Drawings

[0033] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for description in the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0034] Figure 1 It is a flow chart of the present invention.

[0035] Figure 2 is an RGB three-band stacked remote sensing data image of a certain section of the Yellow River in this embodiment, as well as the overall corresponding classification effect (DBSCAN classification) and the local corresponding Yellow River water body classification effect; among them, (a) is the RGB three-band stacked remote sensing data image; (b) is the overall corresponding classification effect corresponding to (a); (c) is the local classification effect corresponding to (a).

[0036] Figure 3 It is the stacked remote sensing data image corresponding to the optimal band obtained in this embodiment and the Yellow River water body identification result; (a) is the stacked remote sensing data image; (b) is the Yellow River water body identification result corresponding to (a). Detailed Embodiment

[0037] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the protection scope of the present invention.

[0038] AsFigure 1 As shown in Figure 1 , an embodiment of the present invention provides a flood warning method for the lower reaches of the Yellow River. First, remote sensing image data of different bands in the middle reaches of the Yellow River are collected, and the remote sensing data of different bands are superimposed to obtain the remote sensing image of the current middle reaches of the Yellow River. Secondly, according to the remote sensing image of the current middle reaches of the Yellow River, the current Yellow River water body recognition rate is obtained, and based on the water body recognition rate of the current Yellow River area, the locally optimal spectral band is obtained. Finally, according to the locally optimal spectral band, the optimal Yellow River water body recognition result is obtained; according to the current Yellow River water body recognition result, flood warning for the lower reaches of the Yellow River is realized. The specific steps are as follows:

[0039] Step 1: Collect remote sensing image data of different bands in the middle reaches of the Yellow River, and superimpose the remote sensing image data of different bands to obtain the superimposed remote sensing data; obtain the remote sensing data of different bands in the Yellow River area through satellite collection, where the band data in the remote sensing data represents the spectral values of ground object information. The data difference between spectral values reflects the distinction between different ground object information on the earth's surface.

[0040] The data on the remote sensing image is generally the DN value. The DN value represents the pixel brightness value of the remote sensing image, which is an integer value without a unit, and its value is related to the radiation resolution of the sensor, the emissivity of the ground object, the atmospheric transmittance, and the scattering rate, etc.

[0041] Since the remote sensing data of different bands are collected by the same satellite and the satellite scanning method remains unchanged, at the same position in the remote sensing data, the data of different bands represent the same ground object information. Furthermore, by superimposing the remote sensing data of different bands, the superimposed remote sensing data can be obtained, which can highlight the greater difference between the water body and the surrounding ground object information.

[0042] The present invention pre-obtains the RGB imaging result of the remote sensing data by superimposing the remote sensing information corresponding to the ranges of the red, green, and blue bands, as shown in Fig. 2(a). Furthermore, through the coordinate calibration by relevant experienced personnel, a coordinate position in the remote sensing data that belongs to the water body in the middle reaches of the Yellow River can be obtained to determine the position of the water body in the middle reaches of the Yellow River. Only obtaining the same position is sufficient because due to the flow between water bodies, the spectral values between water bodies change little locally in the remote sensing data.

[0043] Step 2: Use manual annotation to obtain the position of the middle reaches of the Yellow River in the superimposed remote sensing data; and divide the image of the superimposed remote sensing data into water body regions along the direction perpendicular to the extension of the Yellow River water body to obtain M local regions.

[0044] Step 3: Perform clustering on each local region to obtain K categories, and map the position of the middle reaches of the Yellow River to the categories among the K categories to obtain the Yellow River water body recognition results (i.e., the Yellow River water body regions) of each local region.

[0045] Obtain the horizontal and vertical coordinate values of each position in the remote sensing data after the superposition of the current RGB three bands, as well as the RGB three-band spectral values, to form a five-dimensional vector. Use the DBSCAN algorithm to cluster the five-dimensional data at each position in the remote sensing data. A total of k categories can be obtained. Obtain the category where the above-mentioned manually marked water body position is located, and the remote sensing data corresponding to this category is the water body recognition result. Figure 2(b) shows the overall corresponding classification effect (DBSCAN classification), and Figure 2(c) shows the local corresponding classification effect of the Yellow River water body. When using the DBSCAN algorithm to cluster the five-dimensional data at each position in the remote sensing data, normalize the remote sensing data of each dimension. Set the clustering radius to 0.05 and the minimum number to 100. The clustering radius and the minimum number are hyperparameters, and the implementer can adjust them according to specific situations. The implementer can choose other water body segmentation methods to obtain the recognition result of the middle reaches of the Yellow River in the current band combination.

[0046] After obtaining the recognition result of the middle reaches of the current Yellow River water body, partition the superimposed remote sensing data. Horizontally, every 100 data is a local area. First, perform the recognition of the water body in the entire remote sensing data, and then divide the recognition result of the Yellow River water body into local areas.

[0047] Step 4: Calculate the smoothness of the water bodies belonging to the water body category in each local area. Combine the smoothness of the water body with the difference in the remote sensing data between all categories in the neighborhood to obtain the water body recognition rate in each local area. Since in different geological regions, the difference between the remote sensing data of the Yellow River water body and the remote sensing data around the water body may change, resulting in possible instability of the water body recognition effect. Furthermore, through the change in the water body recognition accuracy rate in the specific areas of different remote sensing data, recombine the remote sensing data of different bands to obtain a new remote sensing data superposition result, making the water body area obvious and improving the accuracy of water body recognition. Thus, ensure the stability of the water body recognition result and improve the accuracy of flood warning in the lower reaches of the Yellow River.

[0048] The calculation method of the water body smoothness is as follows: Obtain the water body category g in the current i-th local area according to the recognition result of the Yellow River water body in each local area i ; Use the edge detection algorithm to process the water body category g in the current i-th local area i to obtain the edge pixel points of the water body category g i ; Use the method of polynomial fitting to fit the coordinates of the edge pixel points of the water body category g i to obtain the corresponding fitting function; Take the first derivative of the fitting function to obtain the slope values corresponding to each coordinate point of the edge pixel points on the fitting function in the water body category g i ; Obtain the variance value of all current slope values as the water body category g in the current i-th local area iWater body smoothness S i 。

[0049] Since there is no sudden lane change during the flow of the Yellow River, the boundary of the Yellow River water body should be relatively smooth, which can be used to indicate that the current recognition result of the Yellow River water body is good.

[0050] Moreover, for the data within the category of the recognition result of the Yellow River water body in the current i-th local area, the greater the difference between it and the data in the neighboring category, the better the recognition rate of the current Yellow River water body, and the higher the accuracy rate in the process of recognizing the Yellow River water body. However, in the recognition of the Yellow River water body in more areas along the flow direction of the Yellow River, due to the change of ground object information, the difference between the water body and the surrounding ground object information changes, resulting in a deterioration of the water body recognition result. Therefore, the difference between the data in different bands and the water body data at this category position is obtained, and the superposition weight is adjusted according to the difference result.

[0051] The smoother the water body boundary and the greater the difference between the remote sensing data of the water body in the current i-th local area and the remote sensing data in the neighboring category, the higher the accuracy rate of water body recognition in the current i-th local area. Calculate the water body recognition rate L i :

[0052] L i =exp(-a*S i )*C i ;

[0053] Among them, the larger the value of S i , the worse the boundary smoothness of the middle reaches of the Yellow River water body in the current i-th local area. The negative correlation mapping is performed on S i using the exp(-x) function, so that the larger its s i value, the larger the L i value, indicating the better the water body recognition effect. C i is the average difference between the remote sensing data within the category of the recognition result of the Yellow River water body in the i-th local area and the remote sensing data in the neighboring category. The acquisition method of C i is as follows:

[0054] Calculate the mean value H i of the remote sensing data within the category of the recognition result of the Yellow River water body in the i-th local area (where the remote sensing data is a three-dimensional data, and the mean value is the mean of each three-dimensional data) and the mean value H ij of the remote sensing data in the j-th neighboring category of the i-th local area, and calculate the absolute value H′ ij of the difference between H ij and H i ; the difference H′ ijThe larger it is, the greater the difference between the remote sensing data within the i-th local area in the Yellow River water body recognition result category and the remote sensing data in its neighboring categories, indicating that the current water body recognition effect is better.

[0055] Calculate the difference degree value Y of the remote sensing data between the j-th neighboring category and the category to which the i-th local area belongs ij = E ij *H′ ij ; where E ij is the weight; for the H′ corresponding to the neighboring category closer to the Yellow River water body recognition result category ij is more important. Therefore, the minimum distance value from the center point coordinates of the j-th neighboring category in its neighboring categories to the Yellow River water body boundary can be obtained by the distance formula between a point and a curve. The smaller the distance value between the j-th neighboring category and the Yellow River water body category, the more important it is. Then, the distance values between all neighboring categories and the Yellow River water body category are normalized, and 1 minus the normalized result is used to obtain E ij， such that the H′ of the neighboring category closer to the Yellow River water body recognition result category ij has a greater weight.

[0056] C is obtained by calculating the average value of the difference degree values between all neighboring categories and the category to which the i-th local area belongs i .

[0057] Furthermore, the water body recognition rate L i The larger it is, the smoother the boundary of the Yellow River water body result, and the greater the difference between the remote sensing data of the water body category within the i-th local area and the remote sensing data within the neighboring categories, indicating that the accuracy of water body recognition within the current i-th local area is higher. Among them, a is a hyperparameter used to adjust the weight value of the smoothness of the water body segmentation result in the calculation of L. The larger its value, the more it emphasizes the smoothness of the water body segmentation result. In this embodiment, a = 2, and the implementer can adjust it according to the specific implementation scenario.

[0058] Step Five: Obtain the optimal band according to the water body recognition rate in each local area. The image corresponding to the optimal band is as Figure 3(as shown in (a)); after obtaining the water body recognition rate in the \(i\)-th local area, due to the change of the geological area, the water body recognition rate corresponding to the same local area under different bands will also change. Set the threshold \(r = 100\). When the water body recognition rate in the \(i\)-th local area is less than the threshold \(r\), it means that the water body recognition effect does not meet the standard, and the band needs to be changed when performing water body recognition (only the data of the water body category is replaced, that is, ensure that the classification result in the current local area remains unchanged, and select the best three-band data; for example, if the original data is the data of the red, green, and blue bands, at this time, at least one of the bands can be replaced). Under the condition of ensuring that the clustering result in the \(i\)-th local area remains unchanged, recombine the remote sensing data of different bands in the \(i\)-th local area, and calculate the \(C\) i value in the \(i\)-th local area for all combinations; select the band of the combination corresponding to the maximum \(C_i\) value as the optimal band. Among them, the larger the \(C\) i value of the remote sensing data corresponding to each spectral band in the \(i\)-th local area, the more obvious the recognition result will be when using the remote sensing data under this band for water body detection.

[0059] Step Six: Reclassify and determine the remote sensing data corresponding to the optimal band to obtain the final Yellow River water body recognition result, as Figure 3 (as shown in (b)); after obtaining the Yellow River water body recognition accuracy corresponding to the locally optimal spectral band, due to the poor quality of the remote sensing data collected by the satellite itself, the \(L\) of the new Yellow River water body segmentation result in the \(i\)-th local area will i still be less than the threshold \(r\), which means that the Yellow River segmentation result still does not meet the requirements. Therefore, in this embodiment, the optimal band is selected for optimization to make the water body recognition result more obvious, so as to improve the flood warning accuracy of the lower reaches of the Yellow River. The specific implementation method is:

[0060] S6.1. Reclassify the remote sensing data (for the local area) in the \(i\)-th local area corresponding to the optimal band to obtain the classification result of the \(i\)-th local area; here, the entire image can be processed. If it is the entire image, it is also necessary to divide the water body area according to the method in Step Two to obtain the local area, and then reclassify each local area.

[0061] S6.2. Calculate the intersection-over-union ratio of the reclassification result of the \(i\)-th local area and the original water body category of the \(i\)-th local area respectively; and screen out the maximum intersection-over-union ratio to obtain the Yellow River water body recognition result in the reclassification result of the \(i\)-th local area;

[0062] S6.3. Since there may be cases where the newly segmented water bodies in a local area in the recombined band segmentation results still do not meet the standards, it is necessary to perform weighted processing on the data of different bands. By weighting the data of different bands and recombining the weighted remote sensing data of the three bands, remote sensing data with a greater difference between the water body data and other ground object information data is obtained, and the optimal Yellow River water body recognition result is obtained under the existing band range. In this embodiment, the particle swarm optimization algorithm is selected to obtain the weights of the remote sensing data of different bands. If the recognition rate of the Yellow River water body recognition result corresponding to the optimal band is less than the recognition rate threshold, the particle swarm algorithm is used to solve the weights of each band in the i-th local area, and the remote sensing data of each band is weighted to obtain the weighted superimposed remote sensing data;

[0063] S6.4. According to the methods in steps S6.1 - S6.2, the DBSCAN algorithm is used to reclassify the weighted superimposed remote sensing data, and the final Yellow River water body recognition result is obtained through the intersection - union ratio.

[0064] By weighting the remote sensing data of the three bands, the finally superimposed data has a higher water body recognition accuracy rate, meeting the requirements of the water body recognition accuracy rate. The fitness function corresponding to the particle swarm algorithm is:

[0065] M = exp(-L′ i );

[0066] where M is the fitness value, and L′ i is the water body recognition rate corresponding to the weighted remote sensing data of each band in the i - th local area. When using the particle swarm algorithm to solve, after calculating the fitness, the fitness value M of the new solution (new weight) of each iteration is obtained according to the fitness function. By iteratively updating the weights, after meeting the number of iterations, the value of M reaches the minimum, and the corresponding optimal solution at this time is the weighted value required for the three - band data.

[0067] Among them, the weighting of the three - band data is overall weighting. In this embodiment, when solving the three weights, each weight selects 50 solution spaces, the acceleration upper limit is 0.5, both the individual learning rate and the group learning rate are 2, and the number of iterations is set to 100. The values of the above - mentioned parameters can be adjusted by the implementer according to the specific implementation scenario.

[0068] Step Seven: Use the final Yellow River water body recognition result as the data for early warning to conduct early warning of floods in the lower reaches of the Yellow River. According to the final Yellow River water body recognition result, complete the Yellow River water body recognition results of the corresponding local areas in different sections, set the area threshold of the Yellow River water body in the corresponding local areas of different sections, and complete the final Yellow River early warning.

[0069] Since the Yellow River water body recognition result is the boundary of the water body, the area of the water body is obtained by enclosing the boundary into a region. An area threshold t is set for the water body areas in different regions. When the water body area in a local region is greater than the preset area threshold t in the corresponding region, it is considered that the water body in this local region is too large, which will cause flood disasters in the downstream local regions of the Yellow River. Furthermore, flood warnings are issued for the downstream regions after the local regions corresponding to the area threshold t. Here, t is a sequence of hyperparameters, which are the preset upper limit values for the water body areas in different local regions; the area threshold is set according to manual experience, that is, a warning value is set. This can be obtained through historical data. Specifically, by collecting Yellow River image data during normal periods, extracting the connected regions in the image data of the water body areas, and calculating the areas of the connected regions, this can be achieved.

[0070] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection scope of the present invention.

Claims

1. A flood warning method for the lower reaches of the Yellow River, characterized in that, The steps are as follows: Step 1: Collect remote sensing image data in different bands, and select the remote sensing image data in the red, green, and blue bands for superposition to obtain the superimposed remote sensing data; Step 2: Use manual annotation to obtain the position of the middle reaches of the Yellow River in the superimposed remote sensing data; and divide the image of the superimposed remote sensing data into water body regions along the direction perpendicular to the extension of the Yellow River water body to obtain M local regions; Step 3: Perform clustering on each local region to obtain K categories, and map the position of the middle reaches of the Yellow River to the categories in the K categories to obtain the Yellow River water body recognition results for each local region; Step 4: Calculate the water body smoothness of the water body category in each local area, and combine the water body smoothness with the remote sensing data difference among all categories in the neighborhood to obtain the water body recognition rate in each local area; the calculation method of the water body smoothness is: obtain the water body category in the current i-th local area according to the Yellow River water body recognition result of each local area ; Use an edge detection algorithm to process the water body category in the current i-th local area to obtain the edge pixel points of the water body category ; Use the method of polynomial fitting to fit the coordinates of the edge pixel points of the water body category to obtain the corresponding fitting function; Take the first derivative of the fitting function to obtain the slope values corresponding to each coordinate point of the edge pixel points in the water body category on the fitting function; Obtain the variance value of all current slope values as the water body smoothness of the water body category in the current i-th local area ; The calculation method of the water body recognition rate is as follows: ​ ； Among them, is the water body recognition rate in the i-th local area, is the average difference between the remote sensing data within the category where the Yellow River water body recognition result is located and the remote sensing data within the neighboring category in the i-th local area, is a hyperparameter, represents the exponential function with the natural constant e as the base; Step 5: Obtain the optimal band according to the water body recognition rate in each local region; Step 6: Reclassify and determine the remote sensing data corresponding to the optimal band to obtain the final Yellow River water body recognition result; Step 7: Use the final Yellow River water body recognition result as the data for early warning to conduct flood early warning for the lower reaches of the Yellow River.

2. The flood warning method for the lower reaches of the Yellow River according to claim 1, wherein The remote sensing image data refers to the spectral values of ground object information.

3. The flood warning method for the lower reaches of the Yellow River according to claim 1, characterized in that The average difference between the remote sensing data within the category where the Yellow River water body recognition result is located in the i-th local area and the remote sensing data within the neighboring category is obtained as follows: Calculate the mean of the remote sensing data within the category where the Yellow River water body recognition result is located in the i-th local area and the mean of the remote sensing data within the j-th neighborhood category of the i-th local area , and calculate and the absolute value of the difference between ; Calculate the difference degree value of remote sensing data between the j-th neighborhood category and the category to which the i-th local area belongs ; where is the weight; Obtained by calculating the average value of the difference degree values between all neighborhood categories and the category to which the i-th local region belongs .

4. The flood warning method for the lower reaches of the Yellow River according to claim 1, wherein The method for obtaining the optimal band according to the water body recognition rate in each local area is as follows: Set a threshold r. When the water body recognition rate of the i-th local area is less than the threshold r, without changing the clustering result in the i-th local area, recombine the remote sensing data of different bands in the i-th local area, and calculate the value within the i-th local area for all combinations; Select the band of the combination corresponding to the maximum Ci value as the optimal band.

5. The flood warning method for the lower reaches of the Yellow River according to claim 4, wherein, The method of reclassifying and determining the remote sensing data corresponding to the optimal band to obtain the final Yellow River water body recognition result is as follows: S6.1: Reclassify the remote sensing data in the i-th local region corresponding to the optimal band to obtain the classification result of the i-th local region; S6.2: Calculate the intersection over union of the reclassification result of the i-th local region and the original water body category of the i-th local region respectively; and screen out the maximum intersection over union to obtain the Yellow River water body recognition result in the reclassification result of the i-th local region; S6.3: If the recognition rate of the Yellow River water body recognition result corresponding to the optimal band is less than the recognition rate threshold, use the particle swarm algorithm to solve the weights of each band in the i-th local region, and weight the remote sensing data of each band to obtain the weighted superimposed remote sensing data; S6.4: Reclassify the weighted superimposed remote sensing data according to the methods in steps S6.1 - S6.2, and obtain the final Yellow River water body recognition result through the intersection over union.

6. The method for flood warning in the lower reaches of the Yellow River according to claim 5, characterized in that The fitness function corresponding to the particle swarm algorithm is: ； where M is the fitness value, is the water body recognition rate corresponding to the remotely sensed data weighted by each band in the i-th local area.

7. The flood warning method for the lower reaches of the Yellow River according to claim 1, wherein The method of judging whether to conduct flood early warning for the lower reaches of the Yellow River according to the size of the water body area is as follows: Set an area threshold t for the water body areas in different regions. When there is a local region where the water body area is greater than the preset area threshold t in the corresponding region, it is considered that the water body in this local region is too large, which will cause flood disasters in the local area of the lower reaches of the Yellow River. Therefore, flood early warning is conducted for the local regions after the local region corresponding to the area threshold t greater than the area threshold t.

Images