Irrigation area irrigation water demand prediction method based on unmanned aerial vehicle remote sensing

Abstract

The invention relates to the technical field of irrigation water demand prediction, and discloses an irrigation area irrigation water demand prediction method based on unmanned aerial vehicle remote sensing, and the method comprises the following steps: S1, data collection; s2, clustering optimization calculation is carried out; and S3, predicting the irrigation water demand based on a clustering result. According to the method, an optimization factor # imgabs0 # is constructed, different weights are given to different features so as to reflect different indication effects of the different features on irrigation water demand, and correlation between spectral features is considered. Specifically, the weight of each wave band is determined by using the correlation between the spectral characteristics and # imgabs 1 #, and the wave band with higher correlation with # imgabs 2 # is endowed with higher weight; a high weight is given to the # imgabs3 # due to the close relation between the # imgabs3 # and the crop growth condition and the moisture condition of the # imgabs3 #. The optimization factor # imgabs4 # constructed in this way can more accurately reflect the importance of different features in irrigation water demand prediction, and the application precision of the K-means clustering algorithm in the scene is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of prediction of irrigation water demand, and specifically relates to a method for predicting irrigation water demand in irrigation districts based on unmanned aerial vehicle (UAV) remote sensing. Background Art

[0002] Precision agriculture is an inevitable trend in the development of modern agriculture. Its core lies in the refined management of the agricultural production process according to the growth status of crops and environmental conditions, so as to improve crop yield and quality, while reducing resource waste and environmental pollution. Irrigation is one of the most important links in agricultural production. Accurately predicting irrigation water demand is crucial for achieving precise irrigation. UAV remote sensing technology has been widely applied in the agricultural field due to its advantages such as fast speed, flexibility, and high resolution. By carrying multi-spectral or hyperspectral cameras, UAVs can obtain high-resolution images of farmland areas, extract information such as spectral features and texture features of crops from them, and combined with ground observation data, crop growth models and irrigation water demand prediction models can be established.

[0003] The K-means clustering algorithm is a commonly used unsupervised machine learning algorithm, which can divide data into different categories according to the inherent similarity of the data. In agricultural applications, clustering is used to identify different crop types, crop growth stages, soil types, etc. Clustering is applied to the prediction of irrigation water demand, and the farmland area is divided into different irrigation zones according to the spectral features of crops.

[0004] In the scenario of predicting irrigation water demand in irrigation districts based on UAV remote sensing, the traditional K-means clustering algorithm based on Euclidean distance, due to not considering the indicative differences of different features on water demand and the correlations between features, leads to low accuracy of clustering results when using multi-dimensional remote sensing data for the division of irrigation water demand status areas, and it is difficult to accurately reflect the actual irrigation needs of different regions, thus affecting the formulation of precise irrigation strategies. Specifically, Euclidean distance treats all features equally, ignoring the differences in the strength of the indicative effects of different spectral bands, vegetation indices, and environmental factors on irrigation water demand. For example, usually can reflect the water status of crops better than simple visible light bands; at the same time, Euclidean distance regards each feature dimension as independent, ignoring the possible correlations between spectral bands, between vegetation indices, and between environmental factors. For example, there is a strong correlation between the reflectance of the near-infrared band and the red light band, and this correlation is closely related to the growth status and water stress degree of crops. The neglect of these factors results in Euclidean distance being unable to accurately capture the complex information contained in the data, reducing the reliability and practicality of the clustering results.

[0005] In view of this, the present invention is specifically proposed. Summary of the Invention

[0006] To solve the above technical problems, the basic concept of the technical solution adopted in the present invention is:

[0007] A method for predicting irrigation water demand in irrigation areas based on unmanned aerial vehicle remote sensing, comprising the following steps:

[0008] Step S1: Data collection. Using an unmanned aerial vehicle equipped with a multi-spectral or hyperspectral camera, aerial photography is carried out on the target irrigation area under the conditions of clear weather and good lighting to obtain high-resolution image data, and GPS information is recorded synchronously; data preprocessing. Radiometric correction and geometric correction are carried out on the unmanned aerial vehicle images to eliminate errors caused by factors such as the atmosphere, lighting, and unmanned aerial vehicle attitude, and the images are registered into a unified geographic coordinate system; then the normalized difference vegetation index is calculated ; the preprocessed spectral data, the reflectance of the red light band 、the reflectance of the green light band 、the reflectance of the blue light band 、the reflectance of the near-infrared light band and the vegetation index are fused to construct a data set: each pixel is used as a data point, and its attributes include spectral reflectance and vegetation index value, and finally a two-dimensional matrix is formed, where is the number of pixels, is the number of spectral bands, is the number of vegetation indices, and this matrix will be used as the input data for K-means clustering in the subsequent steps;

[0009] Step S2: Clustering optimization calculation based on multi-feature indicative weighting and spatial information adaptive adjustment, specifically including the following steps:

[0010] a. Construct an optimization factor according to the difference in indicative effects and correlation between features, and perform preliminary optimization on the distance metric;

[0011] b. Based on the fusion of local spatial autocorrelation and global variation information, construct an optimization factor to perform secondary optimization on the distance metric;

[0012] c. Use the distance metric optimized twice to perform K-means clustering calculation;

[0013] Step S3: Prediction of irrigation water demand based on the clustering results.

[0014] As a preferred embodiment of the present invention, the construction of the optimization factor in step S2 aims to optimize the distance metric according to the indicative difference of different features on the irrigation water demand and the correlation between features; the specific steps are as follows:

[0015] First of all, all The features are divided into two groups: spectral features and vegetation indices ; among them, the spectral features include the spectral reflectance of four bands, namely the red-band reflectance , the green-band reflectance , the blue-band reflectance and the near-infrared band reflectance ; the vegetation index only includes a normalized difference vegetation index .

[0016] As a preferred embodiment of the present invention, then calculate the indicative weights of each group of features respectively; for the spectral features , calculate the Pearson correlation coefficient between each band and the normalized difference vegetation index ( ); ;

[0017] wherein, represents calculating the correlation coefficient between and , ; is an indicative factor for vegetation growth status and water status; normalize the calculated correlation coefficient to obtain the indicative weight vector of the spectral features:

[0018] ;

[0019] The normalization ensures that the sum of the weights of all spectral bands is .

[0020] As a preferred embodiment of the present invention, next, it is necessary to integrate the weight vector of the spectral features and the weight of the vegetation index into a unified weight vector; since has a length of , corresponding to four spectral bands, expand into a vector with a length of , and its only element is the value of ; that is ; finally, calculate the optimization factor ; splice and into a vector and convert it into a diagonal matrix : ​

[0021] ;

[0022] Among them, means concatenating and into a vector and converting it into a diagonal matrix; the optimization factor is a diagonal matrix, and the elements on its diagonal are the indicative weights of each feature; when calculating the distance, the optimization factor will be multiplied by the feature vector to achieve weighting of different features; features with strong indication of irrigation water demand will be given greater weights, while features with weak indication will be given smaller weights.

[0023] As a preferred embodiment of the present invention, the construction of the optimization factor mainly considers the local spatial correlation of the data and the global data distribution characteristics. By calculating the spatial autocorrelation index of each data point in its local neighborhood and the global coefficient of variation of each feature, the optimization factor is constructed;

[0024] First, determine the local neighborhood: for each data point , select its nearest neighbor data points to form the local neighborhood ; the local neighborhood reflects the spatial information around the data point ; the value of needs to be cross-validated according to the specific research area and data characteristics to select an appropriate value;

[0025] Next, calculate the local spatial autocorrelation index: for each feature in the local neighborhood , calculate its local spatial autocorrelation index ; The index is a commonly used spatial autocorrelation index for measuring the similarity between a certain attribute value in a region and its surrounding area; its calculation formula is:

[0026] ;

[0027] Among them: is the number of pixels in the local neighborhood ; is an element of the spatial weight matrix, indicating the spatial relationship between pixel and pixel ; represents the value of pixel on feature ; Represents a pixel The value taken on the feature ; Represents the feature Global mean value of

[0028] As a preferred embodiment of the present invention, The larger the value of, the more similar the data point is to the data points around it on the feature , that is, the stronger the local spatial correlation; The index measures the spatial autocorrelation of the feature in this region by calculating the product of the differences between each data point and the data points around it on a certain feature; calculate the global coefficient of variation: calculate the global coefficient of variation of each feature to measure the degree of dispersion of the feature values across the entire data set; The calculation formula of is:

[0029] ;

[0030] Among them, and respectively represent the global standard deviation and the global mean value of the feature ; The larger the value of, the greater the global variation degree of the feature ; finally, calculate the optimization factor and obtain the final optimized distance metric: The optimization factor is a diagonal matrix, and the elements on its diagonal are jointly determined by the local spatial autocorrelation index and the global coefficient of variation of each feature; The calculation formula of

[0031] ;

[0032] Among them, and are two adjustable parameters, which are respectively used to control the influence degrees of the local spatial autocorrelation index and the global coefficient of variation; these two parameters are determined in a data-driven manner; the median or mean value of the local spatial autocorrelation indices of all data points is used as the value, and the median or mean value of the coefficients of variation of all features is used as the value; represents element-wise exponentiation; represents element-wise multiplication; represents converting a vector into a diagonal matrix.

[0033] The present invention has the following beneficial effects compared with the prior art:

[0034] The present invention constructs an optimization factor , assigns different weights to different features to reflect their different indicative effects on irrigation water requirements, and considers the correlation between spectral features. Specifically, the correlation between spectral features and is used to determine the weight of each band. The band with a higher correlation with will be assigned a higher weight; for , due to its close relationship with crop growth status and water status, it will also be assigned a higher weight. The optimization factor constructed in this way can more accurately reflect the importance of different features in predicting irrigation water requirements and improve the application accuracy of the K-means clustering algorithm in this scenario.

[0035] The following further describes in detail the specific implementation manners of the present invention with reference to the accompanying drawings. Description of the Drawings

[0036] In the accompanying drawings:

[0037] Figure 1 is a flowchart of a method for predicting irrigation water requirements in an irrigation area based on unmanned aerial vehicle (UAV) remote sensing. Specific Embodiment

[0038] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. The following embodiments are used to illustrate the present invention.

[0039] A method for predicting irrigation water requirements in an irrigation area based on UAV remote sensing includes the following steps:

[0040] Step S1: Data collection and preprocessing.

[0041] Using a UAV equipped with a multi-spectral or hyperspectral camera, aerial photography is carried out on the target irrigation area under clear weather and good lighting conditions to obtain high-resolution image data, and GPS information is recorded synchronously. Then, radiometric correction and geometric correction are performed on the UAV images to eliminate errors caused by factors such as the atmosphere, lighting, and UAV attitude, and the images are registered into a unified geographic coordinate system. Next, the normalized difference vegetation index ( ) is calculated.

[0042] Finally, the preprocessed spectral data (red-band reflectance , green-band reflectance , blue-band reflectance , near-infrared band reflectance ) and vegetation index ( ) are fused to construct a dataset: Each pixel serves as a data point, and its attributes include spectral reflectance and vegetation index values, finally forming a two-dimensional matrix, where is the number of pixels, is the number of spectral bands (in this example ), is the number of vegetation indices (in this example ), and this matrix will be used as the input data for K-means clustering in the subsequent steps.

[0043] Step S2: Optimized calculation of K-means clustering based on multi-feature indicative weighting and adaptive adjustment of spatial information.

[0044] a. Construct an optimization factor according to the difference and correlation in the indicative effects between features, and preliminarily optimize the distance metric.

[0045] Detailed logic:

[0046] In the scenario of predicting irrigation water requirements based on UAV remote sensing, there is a significant problem with the traditional K-means clustering algorithm based on Euclidean distance: It treats all features participating in clustering equally, ignoring the differences in the strength of the indicative effects of these features on irrigation water requirements and the correlations between features. This indiscriminate treatment leads to the clustering results not being able to accurately reflect the true differences in irrigation water requirements in different regions. Specifically, UAV remote sensing data usually contains multiple spectral bands and one or more vegetation indices. These features have different effects on indicating the irrigation water requirements of crops. For example, the normalized difference vegetation index ( ) calculated from the reflectance combination of the near-infrared band ( ) and the red band ( ) is generally considered a good indicator factor for crop growth and water status. A higher value usually means lush vegetation growth, strong transpiration, and relatively large water requirements; while a lower value indicates poor vegetation growth or water stress, and relatively small water requirements. In contrast, the reflectance of the blue band ( ) is more affected by atmospheric scattering, and its direct relationship with crop water requirements is not as close as that of . However, when calculating the distance between two pixel points using the traditional Euclidean distance metric, it will equally calculate the reflectance differences between them in , and bands and simply sum up these differences. This results in the reflectance difference of the and have the same band, but in fact the indication of water requirement by the band should be weaker than that of and bands.

[0047] In addition, the traditional Euclidean distance also ignores the correlation between features. For example, in multispectral data, there is usually a strong negative correlation between the reflectance of the near-infrared band and the red band, and this negative correlation is the basis for indicating the vegetation growth status. In hyperspectral data, there is also often a strong positive correlation between the spectral reflectance of adjacent bands. All these correlation information contains information about crop growth status and water requirement. The traditional Euclidean distance treats each feature dimension as independent when calculating the distance, and ignoring this correlation information leads to the loss of useful information.

[0048] To sum up, since the traditional Euclidean distance metric treats all features equally, ignoring the differences in the indication of irrigation water requirement by different features and the correlation between features, when applying it to the scenario of predicting irrigation water requirement based on UAV remote sensing, it will lead to inaccurate clustering results and unable to effectively distinguish regions with different irrigation water requirement states.

[0049] To solve the above problems, an optimization factor is constructed to assign different weights to different features to reflect their different indication effects on irrigation water requirement and consider the correlation between spectral features. Specifically, the correlation between spectral features and is used to determine the weight of each band. The band with higher correlation with will be assigned a higher weight; for , due to its close relationship with crop growth status and water status, it will also be assigned a higher weight. The optimization factor constructed in this way can more accurately reflect the importance of different features in predicting irrigation water requirement and improve the application accuracy of the K-means clustering algorithm in this scenario.

[0050] Formula logic:

[0051] The construction of the optimization factor aims to optimize the distance metric according to the indication differences of different features for irrigation water requirement and the correlation between features. The specific steps are as follows:

[0052] First, all features are divided into two groups: spectral features and vegetation indices . Among them, the spectral features include The spectral reflectance of each band. In this example, , which are the reflectance of the red band , the reflectance of the green band , the reflectance of the blue band and the reflectance of the near-infrared band ; The vegetation index only contains one normalized difference vegetation index .

[0053] Then, calculate the indicative weights of each group of features. For spectral features , calculate the Pearson correlation coefficient between each band and the normalized difference vegetation index ( ):

[0054]

[0055] where, represents calculating the correlation coefficient between and , . is a widely recognized indicator factor for vegetation growth and water status. If the correlation coefficient between a spectral band and is high, it indicates that this band is also closely related to the vegetation growth and water status. Therefore, it is considered that this band has a strong indicative effect on irrigation water demand. Normalize the calculated correlation coefficient to obtain the indicative weight vector of spectral features:

[0056]

[0057] Normalization ensures that the sum of the weights of all spectral bands is . For the vegetation index , due to the close relationship between and crop growth and water status, and there is only one vegetation index in this example, so we set the weight of to a fixed value. According to experience, set to the maximum value in the spectral feature weight vector .

[0058] Next, it is necessary to integrate the weight vector of spectral features and the weight of the vegetation index into a unified weight vector. Since the length of is (corresponding to four spectral bands), we will Expand to a vector of length . The only element of this vector is , that is . Namely .

[0059] Finally, calculate the optimization factor . Concatenate and into a vector and convert it to a diagonal matrix :

[0060]

[0061] where means concatenating and into a vector and converting it to a diagonal matrix. The optimization factor is a diagonal matrix, and the elements on its diagonal are the indicative weights of each feature. When calculating the distance, the optimization factor will be multiplied by the feature vector to achieve weighting of different features. Features with strong indication of irrigation water requirement will be given greater weights, while features with weak indication will be given smaller weights.

[0062] b. Complete the construction of the optimization factor based on the fusion of local spatial autocorrelation and global variation information, and perform a secondary optimization on the distance metric.

[0063] Detailed logic:

[0064] In step a, the optimization factor is constructed to solve the problem that the traditional Euclidean distance treats all features equally, ignoring the difference in the indication of irrigation water requirement of different features and the correlation between features. However, even considering the indicative weights and correlations of features, there is still a problem: the spatial distribution characteristics of the data are ignored, especially the spatial correlation in local regions and the variation degree of global data, resulting in discontinuous and unreasonable phenomena in the clustering results in space, such as the "salt and pepper" phenomenon.

[0065] Specifically, in the scenario of predicting irrigation water requirement in irrigation areas, adjacent plots often have similar crop planting patterns and irrigation conditions, so their irrigation water requirements are often relatively close, showing strong spatial correlation. For example, in an area with the same crop and consistent irrigation conditions, the irrigation water requirements of different plots inside are usually relatively close, with strong spatial correlation. And the traditional The algorithm only considers the distance of data points in the feature space during the clustering process, without considering the spatial position relationship of data points. This may lead to a "salt-and-pepper" phenomenon in the clustering results in space, that is, the areas of the same category are divided into multiple scattered small pieces, or the areas that are adjacent in space but have slightly larger feature differences are wrongly classified into different categories, resulting in poor spatial continuity of the clustering results.

[0066] In addition, the data distribution characteristics in different regions may also vary. For example, in some regions, the growth conditions of crops are relatively uniform, and the variation degrees of features such as spectral reflectance and vegetation index in each band are relatively small; while in other regions, due to differences in field management measures or other factors, even for the same type of crops, there may be significant differences in their growth conditions, resulting in a large variation degree of features. The traditional K-means algorithm does not consider this difference in data distribution characteristics, which may lead to over-clustering in regions with a large variation degree and under-clustering in regions with a small variation degree.

[0067] For example, suppose there is a cornfield. Due to differences in field management measures, the corn growth in some parts of the field is better, while in other parts, the corn growth is worse. There are certain differences in the spectral characteristics and vegetation index of the corn in these two parts, but since they are adjacent in space and both belong to the state of "medium irrigation water requirement", they should be classified into the same category when conducting irrigation zoning. However, if clustering is carried out only based on the optimization factor without considering spatial information, it is very likely that some scattered pixel points with slightly larger spectral feature differences in these two parts will be wrongly classified into other categories, resulting in a "salt-and-pepper" phenomenon in the clustering results, that is, the area of "medium irrigation water requirement" is divided into multiple discontinuous small pieces, which will affect the formulation and implementation of the irrigation plan.

[0068] To solve the above problems, an optimization factor is constructed to further adjust the distance metric. It can reflect the spatial correlation of data in the local area and the variation degree of global data. Specifically, for each data point, calculate the spatial autocorrelation index of each feature within a local neighborhood around it to measure the spatial correlation of the feature in this local area; at the same time, it is also necessary to calculate the global coefficient of variation of each feature to measure the variation degree of the feature in the entire study area. By combining the local spatial autocorrelation index and the global coefficient of variation, an optimization factor is constructed to make the distance metric more refined, which can better reflect the spatial distribution characteristics of data, thereby improving the application accuracy of the clustering algorithm in the prediction scenario of irrigation water requirement in irrigation areas and reducing the occurrence of the "salt-and-pepper" phenomenon.

[0069] Formula Logic:

[0070] Optimization Factor The construction of the optimization factor mainly considers the local spatial correlation of data and the global data distribution characteristics. By calculating the spatial autocorrelation index of each data point within its local neighborhood and the global coefficient of variation of each feature, the optimization factor is constructed .

[0071] First, determine the local neighborhood: For each data point , select its nearest neighbor data points to form the local neighborhood . This local neighborhood reflects the spatial information around the data point . The value of needs to be cross-validated according to the specific research area and data characteristics to select an appropriate

[0072] value Next, calculate the local spatial autocorrelation index: For each feature (in this example, the features include four spectral bands: and a vegetation index: ) within the local neighborhood , calculate its local spatial autocorrelation index . The

[0073]

[0074] index is a commonly used spatial autocorrelation index for measuring the similarity between a certain attribute value in a region and its surrounding region. Its calculation formula is:

[0075] where: is the number of pixels within the local neighborhood

[0076] is an element of the spatial weight matrix, representing the spatial relationship between pixel and pixel . In this method, if pixel and pixel are one of the nearest neighbor points to each other (i.e., belongs to and belongs to ), then , otherwise .

[0077] represents pixel The value on the feature .

[0078] represents the pixel The value on the feature .

[0079] represents the global mean of the feature .

[0080] The larger the value of , the more similar the data point is to the surrounding data points on the feature, that is, the stronger the local spatial correlation. The index measures the spatial autocorrelation of the feature in this region by calculating the product of the differences between each data point and its surrounding data points on a certain feature. If a data point is close to its surrounding data points on this feature (i.e., the difference is small), then the sum of their products will be large, and the index will also be large, indicating that this region has strong spatial correlation on this feature.

[0081] Then calculate the global coefficient of variation: Calculate the global coefficient of variation of each feature , which is used to measure the dispersion degree of the feature values in the entire data set. The calculation formula of

[0082]

[0083] is: and respectively represent the global standard deviation and global mean of the feature . The larger the value of , the greater the global variation degree of the feature is the ratio of the standard deviation to the mean value, which eliminates the influence of the dimension and can be used to compare the variation degrees between different features. The larger the

[0084] value, the greater the variation degree of this feature, that is, the more dispersed the data is distributed on this feature. Finally, calculate the optimization factor and obtain the final optimized distance metric: The optimization factor is a diagonal matrix, and the elements on its diagonal are jointly determined by the local spatial autocorrelation index and the global coefficient of variation of each feature.

[0085]

[0086] Wherein:

[0087] is a local spatial autocorrelation index vector, corresponding respectively to and five features.

[0088] is a coefficient of variation vector, corresponding respectively to and five features.

[0089] and are two adjustable parameters, used respectively to control the influence degrees of the local spatial autocorrelation index and the global coefficient of variation. These two parameters are determined in a data-driven manner. In the present invention, the median or mean value of the local spatial autocorrelation indices of all data points is used as the value, and the median or mean value of the coefficients of variation of all features is used as the value.

[0090] represents element-wise exponentiation.

[0091] represents element-wise multiplication.

[0092] represents converting a vector into a diagonal matrix.

[0093] Optimization factor Each element on the diagonal of is the product of the power of the local spatial autocorrelation index of the corresponding feature and the power of the global coefficient of variation. This design enables to consider both the local spatial correlation and the global variation degree of the data simultaneously. For features with stronger local spatial correlation (i.e., is larger), their corresponding weights will increase; for features with larger global variation degree (i.e.,

[0094] After constructing the optimization factor and it is applied to the distance metric in K-means clustering to obtain an optimized distance metric formula:

[0095]

[0096] Wherein:

[0097] and are two data points, each containing spectral features and vegetation indices.

[0098] is an optimization factor constructed based on feature indicative weights and correlations. It is a diagonal matrix, and the elements on its diagonal reflect the indicative weights of different features for irrigation water requirements and consider the correlations between spectral features.

[0099] is an optimization factor constructed based on local spatial correlations and data distribution characteristics. It is also a diagonal matrix, and the elements on its diagonal reflect the local spatial correlations around each data point and the global variation degree of each feature.

[0100] represents the square root.

[0101] represents a vector transpose.

[0102] c. Perform K-means clustering calculation using the distance metric optimized by quadratic optimization.

[0103] Detailed logic:

[0104] After obtaining the new distance metric optimized according to the above two steps , perform K-means clustering calculation according to this distance metric. First, we need to determine the number of clustering categories . It can be determined according to prior knowledge or the elbow method value. The elbow method is a commonly used heuristic method for determining the value. Briefly speaking, as the value increases, the sum of squared errors of the clustering result ( ) will gradually decrease; when the value increases to a certain critical point, the rate of decline of SSE will slow down significantly, forming an inflection point similar to an "elbow", and the value corresponding to this inflection point is usually considered the optimal number of clusters. Since the elbow method and traditional K-means clustering calculation are existing methods, the detailed process and formulas will not be introduced here.

[0105] Summary of major steps:

[0106] So far, the optimization construction of the distance metric is completed and clustering division is performed using the optimized distance metric.

[0107] Step S3: Prediction of irrigation water requirements based on the clustering results.

[0108] Based on the K-means clustering results of Step S2, that is, within the irrigation area For different category divisions, irrigation water demand prediction is carried out. Each category area corresponds to a set of clustering centers of specific spectral features and vegetation indices, representing the unique crop growth conditions in that area. We will directly use the data of the clustering centers and combine expert experience to predict the irrigation water demand. The specific steps are as follows:

[0109] 1. Interpretation of clustering center data: For each category , the data of its clustering center contains the average values of this category in various features, directly reflecting the typical features of this category. For example : Category 's average reflectance in the red, green, blue, and near-infrared bands. : Category 's mean value, reflecting the average growth condition of the crops in this category.

[0110] 2. Expert evaluation and determination of irrigation water demand: According to the clustering center data of each category ( ), combined with the local historical irrigation experience and the water demand laws of different crops, evaluate the irrigation water demand status of each category. Based on the evaluation results, give the recommended value of the irrigation water demand per unit area for each category.

[0111] 3. Calculate the total irrigation water demand: Calculate the area of each category, then multiply the irrigation water demand per unit area of each category by its area, and finally sum up the irrigation water demands of all categories to obtain the total irrigation water demand of the entire irrigation area:

[0112]

[0113] Through the above steps, based on the optimized K-means clustering results and combined with expert experience, we have achieved the prediction of the irrigation water demand in the irrigation area. Since the optimized distance metric more accurately reflects the irrigation water demand status of different regions, the irrigation water demand prediction based on this clustering result is also more accurate and reliable.

Claims

1. A method for predicting irrigation water demand in irrigation areas based on unmanned aerial vehicle remote sensing, characterized in that, It includes the following steps: Step S1: Data collection. Use a drone equipped with a multispectral or hyperspectral camera to conduct an aerial survey of the target irrigation area under clear weather and good lighting conditions to obtain high-resolution image data, and synchronously record GPS information; data preprocessing. Perform radiometric correction and geometric correction on the drone images to eliminate errors caused by factors such as the atmosphere, lighting, and drone attitude, and register the images into a unified geographic coordinate system; then calculate the normalized difference vegetation index ; For the preprocessed spectral data, the reflectance in the red band , the reflectance in the green band , the reflectance in the blue band , the reflectance in the near-infrared band and the vegetation index are fused to construct a dataset: Each pixel serves as a data point, and its attributes include spectral reflectance and vegetation index values, and finally a two-dimensional matrix is formed, where is the number of pixels, is the number of spectral bands, is the number of vegetation indices, and this matrix will be used as the input data for K-means clustering in the subsequent steps; Step S2: Clustering optimization calculation based on multi-feature indicative weighted sum and spatial information adaptive adjustment, specifically including the following steps: a. Optimize the factor based on the difference and correlation of the indication effects between features, and preliminarily optimize the distance metric; b. Complete the construction of the optimization factor based on the fusion of local spatial autocorrelation and global variation information, and perform a secondary optimization on the distance metric; ​ c. Perform K-means clustering calculation using a quadratic optimized distance metric; Step S3: Prediction of irrigation water requirement based on the clustering results.

2. The method for predicting irrigation water demand in an irrigation area based on UAV remote sensing according to claim 1, wherein The construction of the optimization factor in Step S2 aims to optimize the distance metric according to the indicative differences of different features for irrigation water requirement and the correlation between features; the specific steps are as follows: First, all features are divided into two groups: spectral features and vegetation indices ; Wherein, Spectral characteristics Include The spectral reflectance of the band, which are the reflectance of the red light band , the reflectance of the green light band , the reflectance of the blue light band And the reflectance of the near-infrared band ; Vegetation index Only includes a normalized difference vegetation index .

3. The method for predicting irrigation water demand in an irrigation area based on UAV remote sensing according to claim 2, wherein, Then, calculate the indicative weights of the features in each group; for spectral features , calculate the Pearson correlation coefficient between each band and the normalized difference vegetation index ; ； Among them, represents calculating and the correlation coefficient between ; is an indicator factor for the vegetation growth status and the moisture status; the calculated correlation coefficient is normalized to obtain the indicative weight vector of the spectral characteristics: ； Normalization ensures that the sum of the weights of all spectral bands is .

4. The method for predicting irrigation water demand in an irrigation area based on UAV remote sensing according to claim 3, wherein Next, the weight vector of the spectral features and the weight of the vegetation index need to be integrated into a unified weight vector; Since has a length of , corresponding to four spectral bands, expand to a vector with a length of ; The only element of the vector is the value of ; That is ; Finally, calculate the optimization factor ; Concatenate and into a vector and convert it into a diagonal matrix : Among them, means concatenating and into a vector and converting it into a diagonal matrix; The optimization factor is a diagonal matrix, and the elements on its diagonal are the indicative weights of each feature; When calculating the distance, the optimization factor will be multiplied by the feature vector to achieve weighting of different features; Features with strong indication of irrigation water requirement will be given greater weights, while features with weak indication will be given smaller weights.

5. The method for predicting irrigation water demand in an irrigation area based on UAV remote sensing according to claim 1, characterized in that Optimization factor The construction of the optimization factor mainly considers the local spatial correlation of data and the global data distribution characteristics. By calculating the spatial autocorrelation index of each data point in its local neighborhood and the global coefficient of variation of each feature, the optimization factor is constructed ; First, determine the local neighborhood: For each data point , select its nearest neighbor data points to form the local neighborhood ; The local neighborhood reflects the spatial information around the data point ; The value of needs to be selected by cross - validation according to the specific research area and data characteristics to choose an appropriate value; Next, calculate the local spatial autocorrelation index: For each feature within the local neighborhood , calculate its local spatial autocorrelation index ; The index is a commonly used spatial autocorrelation indicator, which is used to measure the similarity between a certain attribute value in a region and its surrounding areas; Its calculation formula is: ； Wherein: is the number of pixels in the local neighborhood ; is an element of the spatial weight matrix, representing the spatial relationship between pixel and pixel ; represents the value of pixel on feature ; represents the value of pixel on feature ; represents the global mean of feature 6. The method for predicting irrigation water demand in an irrigation area based on UAV remote sensing according to claim 5, characterized in that, The larger the value, the more similar the data point is to the data points around it in the feature , that is, the stronger the local spatial correlation; The index measures the spatial autocorrelation of the feature in the region by calculating the product of the differences between each data point and the data points around it in a certain feature; calculate the global coefficient of variation: calculate the global coefficient of variation of each feature to measure the dispersion of the feature values over the entire dataset; The calculation formula of is as follows: ； Among them, and represent the global standard deviation and the global mean of feature respectively; The larger the value of , the greater the global variation degree of feature ; Finally, calculate the optimization factor and obtain the final optimized distance metric: The optimization factor is a diagonal matrix, and the elements on its diagonal are jointly determined by the local spatial autocorrelation index and the global variation coefficient of each feature; The calculation formula of ； Among them, and are two adjustable parameters, which are used to control the influence degrees of the local spatial autocorrelation index and the global coefficient of variation respectively; these two parameters are determined by a data-driven method; the median or mean of the local spatial autocorrelation indices of all data points is used as the value, and the median or mean of the coefficients of variation of all features is used as the value; denotes element-wise exponentiation; denotes element-wise multiplication; denotes converting a vector into a diagonal matrix.

Images