Unmanned aerial vehicle remote sensing assisted intelligent water-saving irrigation method and system

Abstract

The invention provides an intelligent water-saving irrigation method and system assisted by remote sensing of an unmanned aerial vehicle, and relates to the field of farmland irrigation, the method comprises the following steps: acquiring a soil humidity image, a crop growth stage image and climate condition data, and classifying through a random forest algorithm to obtain classified farmland region labels; when the farmland area label displays that the soil humidity is lower than a preset humidity threshold value and crops are in a water demand peak period, calculating a water demand priority score, and determining a high-priority irrigation area; the high-priority irrigation area is partitioned through a K-means clustering algorithm, and sub-partition attribute description is obtained; according to the sub-partition attribute description, matching a pre-established irrigation model database, and obtaining an irrigation amount parameter; and updating climate condition distribution of the sub-partitions, and adjusting the irrigation volume parameters through a linear regression model to obtain adjusted irrigation volume parameters. According to the method, dynamic intelligent zoning of the farmland irrigation area can be improved, and the water resource utilization efficiency and the crop yield are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of farmland irrigation, and particularly relates to an unmanned aerial vehicle remote sensing assisted intelligent water-saving irrigation method and system. BACKGROUND

[0002] Agricultural irrigation is an important link to ensure food security and sustainable development of agriculture. In particular, under the background of increasingly scarce water resources, precise and efficient irrigation technology is of key significance to improve water resource utilization and promote crop growth. In recent years, unmanned aerial vehicle remote sensing technology has been gradually applied to agriculture to provide support for irrigation decision-making by obtaining real-time data of farmland.

[0003] The existing unmanned aerial vehicle remote sensing irrigation method has limitations, and the data collection and analysis process lacks comprehensive consideration of multi-dimensional dynamic factors of farmland. Many methods only focus on single soil moisture data, ignoring the interactive influence of crop growth stage, soil type and climate conditions. Therefore, how to integrate multi-dimensional information such as soil, crops and climate on the basis of unmanned aerial vehicle remote sensing data collection, and realize dynamic intelligent zoning of farmland irrigation area accordingly, has become a key problem to improve water resource utilization efficiency and crop yield.

[0004] It should be noted that the information disclosed in the above background section is only used to strengthen the understanding of the background of the present application, and therefore can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0005] Therefore, the present application provides an unmanned aerial vehicle remote sensing assisted intelligent water-saving irrigation method and system to solve the problem of lack of comprehensive consideration of multi-dimensional dynamic factors of farmland in the prior art, realize dynamic intelligent zoning of farmland irrigation area, and improve water resource utilization efficiency and crop yield. The embodiment of the present application provides an unmanned aerial vehicle remote sensing assisted intelligent water-saving irrigation method, comprising:

[0006] scanning the farmland area by a multi-spectral sensor to obtain soil moisture images, crop growth stage images and climate condition data;

[0007] According to the soil moisture images, crop growth stage images and climate condition data, classification processing is performed by a random forest algorithm to obtain classified farmland area labels;

[0008] When the farmland area label shows that the soil moisture is lower than a preset moisture threshold and the crop growth stage is in a high water demand peak period, a decision tree algorithm is used to calculate a water demand priority score, and based on the water demand priority score, a high-priority irrigation area is determined;

[0009] The K-means clustering algorithm is used to partition the high-priority irrigation area, and the average soil moisture value and the average crop growth stage value of the sub-partition are calculated to obtain the sub-partition attribute description.

[0010] According to the sub-partition attribute description, the pre-established irrigation model database is matched to obtain the irrigation amount parameter;

[0011] The climate condition distribution of the sub-partition is updated through real-time weather data, and based on the climate condition distribution, the irrigation amount parameter is adjusted through a linear regression model to obtain the adjusted irrigation amount parameter.

[0012] In some optional embodiments, the classification processing is performed according to the soil moisture image, the crop growth stage image, and the climate condition data through a random forest algorithm to obtain a classified farmland area label, including:

[0013] According to the soil moisture image, K-means clustering is used for classification to obtain a soil moisture category;

[0014] According to the crop growth stage image, the information entropy is calculated by an entropy method to obtain a growth stage entropy value;

[0015] According to the climate condition data, principal components are extracted by principal component analysis to obtain climate principal components;

[0016] According to the soil moisture category, the growth stage entropy value, and the climate principal components, a random forest algorithm is used for classification processing to obtain a classified farmland area label.

[0017] In some optional embodiments, when the farmland area label shows that the soil moisture is lower than a preset humidity threshold and the crop growth stage is in a high water demand peak period, a decision tree algorithm is used to calculate a water demand priority score, and based on the water demand priority score, a high-priority irrigation area is determined, including:

[0018] When the farmland area label shows that the soil moisture is lower than a preset humidity threshold and the crop growth stage is in a high water demand peak period, a decision tree algorithm is used to calculate a water demand priority score, and based on the water demand priority score, a preliminary irrigation area is determined.

[0019] According to the crop growth data and irrigation demand data of the preliminary irrigation area, a weighted score is obtained to obtain a ranking list of the high-priority irrigation area;

[0020] According to the ranking list of the high-priority irrigation area, corresponding geographic location data is obtained to generate a spatial distribution map of the irrigation area;

[0021] According to the spatial distribution map and the soil moisture, a K-means clustering algorithm is used to divide the priority partition of the irrigation area to obtain a high-priority irrigation area.

[0022] In some optional embodiments, the high-priority irrigation area is divided into sub-regions using the K-means clustering algorithm. The sub-region attribute descriptions are obtained by calculating the average soil moisture value and average crop growth stage value of each sub-region, including:

[0023] The soil moisture and crop growth stage of the high-priority irrigation area are obtained and integrated to obtain a standardized data set.

[0024] The standardized dataset was subjected to cluster analysis using the K-means clustering algorithm to determine sub-partitions;

[0025] The sub-region attribute description is obtained by calculating the average soil moisture value and average crop growth stage value of the sub-region.

[0026] In some optional embodiments, the step of matching a pre-established irrigation model database with the sub-partition attribute description to obtain irrigation quantity parameters includes:

[0027] Obtain soil type data and terrain slope data for sub-regions;

[0028] Key features that directly affect irrigation volume are extracted from the sub-region attribute descriptions, and the soil type data and terrain slope data are integrated to obtain a key feature set;

[0029] Based on the key feature set, irrigation parameters are obtained by matching a pre-established irrigation model database using a cosine similarity algorithm.

[0030] In some optional embodiments, after updating the climate condition distribution of the sub-region through real-time weather data, and adjusting the irrigation quantity parameter based on the climate condition distribution using a linear regression model to obtain the adjusted irrigation quantity parameter, the method further includes:

[0031] Recalculate the water demand priority score of the sub-partition. If the water demand priority score of the sub-partition meets the dynamic perception requirements, then use the rasterization method to generate the partition decision map.

[0032] Based on the zoning decision map, a linear regression algorithm is used to predict the water demand of sub-zoning areas, and an irrigation water allocation scheme is obtained.

[0033] This invention provides an intelligent water-saving irrigation system assisted by unmanned aerial vehicle (UAV) remote sensing, comprising:

[0034] The data acquisition module is used to scan farmland areas using a multispectral sensor to acquire soil moisture images, crop growth stage images, and climate condition data;

[0035] The classification module is used to classify farmland areas based on the soil moisture image, crop growth stage image and climate condition data using a random forest algorithm to obtain classified farmland area labels.

[0036] The priority confirmation module is used to calculate the water demand priority score using a decision tree algorithm when the farmland area label shows that the soil moisture is lower than the preset moisture threshold and the crop growth stage is in the peak water demand period. Based on the water demand priority score, the high priority irrigation area is determined.

[0037] The partitioning module is used to partition the high-priority irrigation area using the K-means clustering algorithm. By calculating the average soil moisture value and average crop growth stage value of the sub-partitions, the attribute description of the sub-partitions is obtained.

[0038] The parameter matching module is used to match the pre-established irrigation model database according to the sub-partition attribute description to obtain irrigation quantity parameters;

[0039] The adjustment module is used to update the climate condition distribution of the sub-region through real-time weather data, and adjust the irrigation quantity parameters based on the climate condition distribution through a linear regression model to obtain the adjusted irrigation quantity parameters.

[0040] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention.

[0041] The intelligent water-saving irrigation method and system assisted by unmanned aerial vehicle (UAV) remote sensing of the present invention has the following beneficial effects:

[0042] This invention uses a multispectral sensor to scan farmland areas and acquire soil moisture images, crop growth stage images, and climate condition data. Based on these data, a random forest algorithm is used for classification to obtain categorized farmland area labels. When the farmland area labels indicate that the soil moisture is below a preset moisture threshold and the crop growth stage is at its peak water demand period, a decision tree algorithm is used to calculate a water demand priority score. Based on this score, high-priority irrigation areas are identified. A K-means clustering algorithm is used to partition these high-priority irrigation areas. By calculating the average soil moisture value and average crop growth stage value of each sub-partition, a sub-partition attribute description is obtained. Based on the sub-partition attribute description, a pre-established irrigation model database is matched to obtain irrigation parameters. The climate condition distribution of the sub-partitions is updated using real-time weather data. Based on this climate condition distribution, a linear regression model is used to adjust the irrigation parameters, resulting in adjusted irrigation parameters. This invention enables dynamic intelligent partitioning of farmland irrigation areas, improving water resource utilization efficiency and crop yield. Attached Figure Description

[0043] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings.

[0044] Figure 1 This is a flowchart illustrating the unmanned aerial vehicle (UAV) remote sensing-assisted intelligent water-saving irrigation method disclosed in an embodiment of the present invention.

[0045] Figure 2 This is a schematic diagram of the structure of the UAV remote sensing-assisted intelligent water-saving irrigation system disclosed in an embodiment of the present invention. Detailed Implementation

[0046] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the invention will be more comprehensive and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0047] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0048] The flowchart shown in the attached diagram is merely an illustrative example and does not necessarily include all steps. For example, some steps may be broken down, while others may be combined or partially combined. Therefore, the actual execution order may change depending on the specific circumstances.

[0049] like Figure 1 As shown in the figure, this embodiment of the invention provides an intelligent water-saving irrigation method assisted by unmanned aerial vehicle (UAV) remote sensing. This method can realize dynamic intelligent zoning of farmland irrigation areas, improve water resource utilization efficiency and crop yield.

[0050] S10. Data Acquisition: The drone, equipped with a 5-band multispectral camera, scans 100 hectares of farmland daily from 8:00 to 10:00 AM at a flight altitude of 100 meters and a resolution of 0.1 meters per pixel, simultaneously acquiring three types of data. Soil moisture images are acquired using the near-infrared band (850nm) and calculated using the formula NDWI = (NIR - G) / (NIR + G), where NIR represents the reflectance of the target in the near-infrared band, G represents the reflectance of the target in the green band, and NDWI is the Normalized Difference Water Index, used to indirectly reflect the moisture content of the earth's surface or objects. In agriculture, it is often used to generate soil moisture images to determine the degree of soil moisture. Crop growth stage images are acquired using the red-edge band (730nm) and calculated using NDVI = (NIR - R) / (NIR + R), where R is the reflectance of the target in the red band; NIR is the reflectance of the target in the near-infrared band; and NDVI is the Normalized Difference Vegetation Index, used to quantify crop growth status. Climate data is collected by airborne sensors, including temperature ranging from -10°C to 40°C (accuracy 0.1°C), relative humidity from 0 to 100% (accuracy 1%), and wind speed from 0 to 20 m / s (accuracy 0.1 m / s).

[0051] S20. Classification: Based on soil moisture images, soil moisture sensors are deployed in the corresponding areas of the soil moisture images, recording moisture data hourly, ranging from 10% to 50%, with an average of 30% and a standard deviation of 5%. The moisture data is uniformly normalized to the [-1, 1] interval using Z-score standardization. K-means clustering is then used, with the number of clusters set to K = 3, to classify the soil moisture data according to its distribution pattern, thus obtaining soil moisture categories. Growth stage entropy values ​​are calculated from crop growth stage images. First, NDVI thresholds (0.2-0.4 for germination, 0.4-0.7 for growth, and >0.7 for maturity) are used to match plot colors to determine the stage of each plot. Then, the entropy method is used to calculate the information entropy of different growth stages. Climate principal components are extracted from climate condition data. The climate condition data is first standardized, then principal component analysis (PCA) is used for dimensionality reduction. By analyzing the variance contribution of each climate indicator, principal components that retain core information are selected. Using the aforementioned soil moisture categories, growth stage entropy values, and climate principal components as the three main input features, a random forest classification model is constructed. The model classifies farmland areas into different labels based on the combined effect of these three features.

[0052] S30. Priority Confirmation: When farmland humidity is below a preset threshold (e.g., preset threshold 30%, actual humidity 20%) and crops are in their peak water demand period, a water demand priority assessment is triggered. A decision tree algorithm is used to calculate a score: the first-level nodes calculate the humidity difference and divide the area into different branches based on the magnitude of the difference; the second-level nodes introduce crop water demand weights; the third-level nodes process area factors; and the fourth-level nodes introduce historical irrigation weights. The terminal node of the decision tree outputs the final priority score, which is compared with a preset threshold of 8 to determine whether the area is a preliminary irrigation area. After determining the preliminary irrigation area, further data on crop growth and irrigation demand are collected for each area. Indicators such as duration of water shortage, soil water retention capacity, and water loss rate are quantified and weighted according to "water demand urgency 60% and irrigation feasibility 40%" to calculate a comprehensive score for each preliminary irrigation area. The areas are then sorted from highest to lowest score to form a high-priority irrigation area ranking list. Subsequently, the geographic location data, such as latitude and longitude and plot boundaries, of each region in the GIS database were linked. GIS tools were used to overlay the priority-location data onto the farmland base map, using color and symbol size to mark priorities, displaying core information in pop-up windows, and overlaying irrigation equipment and water source layers to generate a spatial distribution map. Then, using latitude and longitude and soil moisture difference as feature variables, K-means clustering was performed with K=2. After iterating until the groups stabilized, higher priority zones were determined by calculating the average comprehensive score within each zone, thus completing the selection of high-priority irrigation areas.

[0053] S40. Zoning: Collect soil moisture (percentage) and crop growth stage (0-1 numerical value) for each farmland in the high-priority irrigation area (e.g., containing 1000 farmland plots). Standardize the data to the [0,1] interval to form a standardized dataset. Use K-means clustering (set K=3). First, randomly initialize 3 centroids in the feature space. Calculate the distance between each farmland and the centroid using Euclidean distance, assign farmland to the nearest centroid, iteratively update the centroids until stable, and determine sub-zoning (e.g., 3 sub-zoning plots containing 400, 350, and 250 farmland plots respectively). Then, denormalize the centroid data of each sub-zoning, calculate the average soil moisture and average crop growth stage, and generate sub-zoning attribute descriptions based on these values ​​(e.g., "medium soil moisture, early crop growth stage"), clearly reflecting the core characteristics of each sub-zoning.

[0054] S50. Parameter Matching: Extract key features that directly affect irrigation amount from the sub-region attribute descriptions. For "soil type" and "topography slope" not present in the attribute descriptions but present in the pre-established irrigation model database, supplement them from the sub-region's basic database. For non-numerical key features such as soil type and crop type, a "text-number" mapping table needs to be established in advance, and the mapping rules must be completely consistent with the coding in the irrigation model database. Convert these texts into numbers according to the mapping table. For numerical features (quantified indicators such as slope and humidity), no coding is required, but the units or precision must be consistent with the database format. Finally, arrange the processed classification feature numbers and numerical feature quantification values ​​in a fixed order (consistent with the feature order recorded in the database) to form a feature vector. Subsequently, use the cosine similarity algorithm to match the pre-established irrigation model database, find the record with the highest matching degree (e.g., similarity 0.98), and its corresponding irrigation amount parameter (e.g., 30 mm) is the preliminary irrigation amount parameter for the current sub-region.

[0055] S60. Adjustment: After obtaining the preliminary irrigation parameters for the sub-regions, key weather indicators affecting soil moisture consumption are collected in real time, including real-time temperature, air humidity, rainfall, and evaporation in the sub-regions. These data are then associated with the spatial identifiers of the sub-regions to update the "climate condition distribution" specific to each sub-region. Next, using a pre-built linear regression model, the various indicators in the climate condition distribution are transformed into quantitative impacts on irrigation, and the adjusted irrigation parameters are obtained by substituting them into the formula.

[0056] Through the above steps, this embodiment can realize dynamic intelligent zoning of farmland irrigation areas, thereby improving water resource utilization efficiency and crop yield.

[0057] In some embodiments, based on the above embodiments, the process of scanning farmland areas using a multispectral sensor to acquire soil moisture images, crop growth stage images, and climate condition data includes:

[0058] To acquire soil moisture images, crop growth stage images, and climate condition data for farmland areas, a multispectral sensor was mounted on a drone for scanning operations. The drone was equipped with a 5-band multispectral camera covering the 400-900nm wavelength range, including blue, green, red, and near-infrared bands. During operations, the drone's flight altitude was set at 100 meters to ensure an image resolution of 0.1 meters per pixel, covering 100 hectares of farmland. Data collection was conducted daily between 8:00 and 10:00 AM, when stable lighting conditions ensured data accuracy.

[0059] In acquiring soil moisture images, water has very weak reflectivity in the near-infrared (NIR) band, while the reflectivity of soil or crops in the green (G) band is relatively stable and does not fluctuate significantly with humidity. For example, wet soil, with more moisture, absorbs more near-infrared light, so its NIR reflectivity will be lower; conversely, dry soil, with less moisture, absorbs less near-infrared light, so its NIR reflectivity will be higher. The normalized difference water index can be calculated using the near-infrared (850nm) reflectivity from a multispectral camera to indirectly invert the moisture content of the land surface or objects.

[0060] Its formula is

[0061] NDWI = (NIR - G) / (NIR + G)

[0062] NIR represents the reflectivity of the target in the near-infrared band, and G represents the reflectivity of the target in the green band. The reflectivity range of NIR and G is controlled between 0 and 1. NDWI is the normalized difference water index, and an NDWI value of 0.3 or higher is used as the criterion for judging high humidity areas.

[0063] An algorithm uses color mapping to transform the NDWI values ​​of different areas of the field into an intuitive image. For example, NDWI values ​​of 0-0.1 (extremely dry) are marked as light yellow, 0.1-0.3 (slightly dry) as yellow, 0.3-0.5 (moist) as light green, and values ​​above 0.5 (extremely moist) as dark green. Different colored areas in the image directly correspond to soil moisture levels, allowing for quick identification of arid, moderately moist, or waterlogged areas, thus obtaining a soil moisture image.

[0064] Acquiring images of crop growth stages relies heavily on the red band (R, typically around 650 nm) and the near-infrared band (NIR, typically around 850 nm). This is because chlorophyll in crop leaves absorbs a large amount of red light while strongly reflecting near-infrared light. Different growth stages result in varying chlorophyll content and leaf structure, leading to differences in the reflectance of NIR and R. The normalized vegetation index (NDI) is used to quantify crop growth status, as shown in the formula:

[0065] NDVI = (NIR - R) / (NIR + R)

[0066] Where R is the reflectance of the target in the red band, NIR is the reflectance of the target in the near-infrared band, and NDVI is the normalized vegetation index.

[0067] Crop growth stages are defined based on NDVI values, with 0.2-0.4 representing the germination stage, 0.4-0.7 representing the growth stage, and values ​​above 0.7 representing the maturity stage.

[0068] By combining NDVI values ​​with images, crop growth stage images are generated. An algorithm maps NDVI values ​​for different areas of the field to different colors; for example, germination areas with NDVI of 0.2-0.4 are marked with light yellow, growth areas with NDVI of 0.4-0.7 are marked with light green, and maturity areas with NDVI above 0.7 are marked with dark green, forming an intuitive growth stage distribution map.

[0069] Climate condition data was collected by temperature and humidity sensors carried by the drone, specifically including temperature, rainfall and wind speed. The temperature collection range was -10 to 40℃ with an accuracy of 0.1℃, and the wind speed collection range was 0-20m / s with an accuracy of 0.1m / s.

[0070] In some embodiments, based on the above embodiments, the step of obtaining classified farmland area labels by performing classification processing using a random forest algorithm based on the soil moisture image, crop growth stage image, and climate condition data includes:

[0071] Based on the soil moisture images, K-means clustering is used to classify the soil moisture categories.

[0072] Based on the crop growth stage images, the information entropy is calculated using the entropy method to obtain the growth stage entropy value;

[0073] Based on the climate condition data, principal component analysis was used to extract principal components to obtain climate principal components.

[0074] Based on the soil moisture category, the growth stage entropy value, and the climate principal component, a random forest algorithm is used for classification to obtain the classified farmland area labels.

[0075] Based on soil moisture images, soil moisture sensors are deployed within the corresponding areas. Referring to the differences in moisture distribution in the images, the farmland is divided into 100 grids of 100m × 100m. The center point of each grid is located using GPS to ensure coverage of all moisture types. Assuming the farmland area is 100 hectares, 100 soil moisture sensors are deployed, recording moisture data hourly, ranging from 10% to 50%, with an average of 30% and a standard deviation of 5%.

[0076] These humidity data were Z-score standardized using the following formula:

[0077] x'=(x-μ) / σ

[0078] Where x' is the normalized value, x is the original value, μ is the mean of the original data, and σ is the standard deviation of the original data.

[0079] The formula for the mean μ is:

[0080] Where, x i is the i-th original data, and n is the total number of valid data (i.e., the sample size).

[0081] The standard deviation σ is the square root of the variance. Variance reflects the degree to which each data point deviates from the mean. The formula is:

[0082] Where, x i Let be the i-th original data point, n be the total number of valid data points, and μ be the mean of the original data points.

[0083] Z-score standardization was used to normalize the humidity data to the range of [-1, 1] to eliminate dimensional differences and avoid the impact of numerical range differences on subsequent analysis.

[0084] Next, the K-means clustering algorithm was used, with the number of clusters K=3. The algorithm will automatically divide all plots into low humidity (humidity <20%), medium humidity (humidity 20%-40%), and high humidity (humidity >40%) according to the distribution pattern of humidity data.

[0085] The entropy value of the growth stage is calculated from the crop growth stage image. First, the color of the plot is matched by NDVI threshold (0.2-0.4 for germination stage, 0.4-0.7 for growth stage, and >0.7 for maturity stage) to determine the stage to which each plot belongs.

[0086] Suppose 100 hectares of farmland are divided into 1000 plots, and then into n sub-regions (e.g., each sub-region contains 10 plots). Count the number of plots in each sub-region at a certain stage (e.g., the growing season).

[0087] The information entropy at different growth stages is then calculated using the entropy method, with the following formula:

[0088] H = -∑q i ×ln(q i )

[0089] Where, q i H represents the proportion of a plot of land at a certain growth stage in the i-th sub-region, and H is the information entropy.

[0090] The core function of information entropy is to quantify the heterogeneity of the distribution of growth stages. The higher the entropy value, the more dispersed and diverse the plots at that growth stage are spatially. These entropy values ​​transform the classification information of crop growth stage images into quantitative features, which can more accurately reflect the spatial differences in crop growth status.

[0091] Climate principal components are extracted from climate condition data, which typically includes multi-dimensional data such as temperature, rainfall, and wind speed. First, these data are standardized using Z-score to eliminate the dimensional differences between different climate indicators. Then, principal component analysis (PCA) is used to reduce dimensionality and select principal components that can retain core information. For example, temperature and rainfall are finally extracted as the two principal components, forming the core input variable of "climate principal components".

[0092] Finally, the random forest algorithm was used to classify farmland areas. The soil moisture category, growth stage entropy value, and climate principal component obtained above were used as the three main input features of the model to construct a random forest classification model. The model parameters were set to 100 trees and a maximum depth of 10 (to ensure that the model has classification ability while avoiding overfitting). Then, all farmland sample data were divided into a training set (70%) and a test set (30%) in a 7:3 ratio. The model was trained using the training set and the model performance was verified using the test set. This completed the acquisition of farmland area labels after classification.

[0093] In some embodiments, based on the above embodiments, when the farmland area label shows soil moisture below a preset moisture threshold and the crop growth stage is at its peak water demand period, a decision tree algorithm is used to calculate a water demand priority score, and based on the water demand priority score, a high-priority irrigation area is determined, including:

[0094] When the farmland area label shows that the soil moisture is lower than the preset moisture threshold and the crop growth stage is in the peak water demand period, the decision tree algorithm is used to calculate the water demand priority score, and the preliminary irrigation area is determined based on the water demand priority score.

[0095] Based on the crop growth data and irrigation demand data of the preliminary irrigation area, a weighted score is used to obtain a ranking list of high-priority irrigation areas;

[0096] Based on the sorted list of high-priority irrigation areas, obtain the corresponding geographical location data and generate a spatial distribution map of the irrigation areas;

[0097] Based on the spatial distribution map and soil moisture, the K-means clustering algorithm is used to divide the irrigation area into priority zones to obtain high-priority irrigation zones.

[0098] When the farmland area label indicates that the soil moisture is below a preset threshold (e.g., preset threshold 30%, actual moisture 20%) and the crop is in its peak water demand stage, a water demand priority assessment is triggered first. At this point, a decision tree algorithm is used to calculate the water demand priority score. The decision tree uses key factors affecting water demand (moisture difference, crop growth stage weight, farmland area factor, historical irrigation weight) as classification nodes, gradually breaking down and quantifying the contribution of each factor to the priority.

[0099] First, the root node of the decision tree will verify the basic triggering conditions, and determine whether "soil moisture is lower than the preset threshold" and "crops are in the peak water demand period". Only if both conditions are met will the subsequent calculation be carried out; otherwise, it will be determined that no priority irrigation is needed.

[0100] After entering the calculation process, the first layer of nodes focuses on the humidity difference, which is the core input of the decision tree. Different branches will be divided according to the size of the difference: for example, the preset "difference ≤ 5%" corresponds to the basic coefficient 1, "5% < difference ≤ 15%" corresponds to the basic coefficient 1.2 (since 10% falls within this range, it is initially assigned an initial weight of 1.2), and "difference > 15%" corresponds to a higher coefficient, so as to reflect that the more severe the water shortage, the higher the basic weight.

[0101] The second-layer node introduces crop water requirement weights, adjusting scores based on the weight values ​​during peak water demand periods. The core of setting crop water requirement weights is to quantify the urgency of water needs for different crops at different growth stages. This requires combining crop physiological characteristics, critical growth periods, yield impact thresholds, and practical agricultural production experience, supported by scientific data and calibrated through practice. Ultimately, it is mapped to a numerical range of 0-1 (1 representing the highest water urgency, 0 representing the lowest). The specific setting logic can be elaborated from the following aspects:

[0102] First and foremost, the core basis is the difference in physiological needs at different stages of crop growth. Crops have drastically different water dependencies at different stages of their life cycle. Water requirement weights are prioritized for critical water-demand periods, such as the flowering and grain-filling stages of grain crops, and the flowering, boll-forming, and fruit-bearing stages of cash crops. These stages are when crop photosynthesis is strongest and nutrient transport is most active. Water shortages directly lead to poor pollination and hindered fruit development, resulting in yield losses of 20%-50% or more. Therefore, the water requirement weight for these critical stages is typically set at 0.7-0.9. However, during the sowing stage (where only moist soil is needed for germination, and replanting can be done if water is lacking), the seedling stage (where growth is slow, and short-term water shortages can be recovered from), or the maturity stage (where crops have accumulated most of their dry matter, reducing water requirements), water shortages have a smaller impact on yield (the loss rate is usually less than 5%), and the water requirement weight is adjusted to 0.2-0.5.

[0103] Secondly, it is necessary to consider the inherent water requirements of different crop species. The root depth and leaf transpiration efficiency vary greatly among different crops, determining their basic water requirements. For example, rice is a typical water-loving crop, requiring flooded or moist environments throughout its growth cycle. Even water shortages during non-critical periods (such as the tillering stage) will affect the number of tillers. Therefore, its water requirement weight at each stage is generally higher than that of dryland crops (e.g., 0.5 for rice tillering, while only 0.3 for corn tillering). Drought-resistant crops such as millet and sorghum have root systems that can reach depths of 1-2 meters, enabling them to absorb deep soil moisture. Even during short-term water shortages at the heading stage (the theoretical critical period), they can self-regulate, further reducing their water requirement weight (around 0.6 during the heading stage). Among cash crops, vegetables have large leaves, rapid transpiration, and high water requirement frequency, so their weight setting tends to be at a medium-to-high level (0.8-0.9 during the fruiting stage).

[0104] Furthermore, calibration is required, taking into account agricultural scientific experimental data and regional production experience. The water requirement weight is not a fixed value and will be adjusted according to local climate and soil conditions. For example, in arid and semi-arid regions in the north (such as Hebei and Gansu), evaporation is high and soil water retention is poor, and crops may have difficulty recovering from water shortages during non-critical periods. Therefore, the non-critical weight will be appropriately increased (e.g., the wheat seedling weight is increased from 0.3 to 0.4). In contrast, in rainy regions in the south, soil moisture is consistently high, and only the critical period of water shortage needs to be focused on. The non-critical weight can be maintained at a lower level (e.g., the rice grain-filling weight is 0.8, and below 0.3 at other times).

[0105] Finally, dynamic fine-tuning of irrigation decisions needs to be considered in light of actual needs. For example, when multiple crops are planted in farmland at the same time, the water demand weight of high-value crops will be appropriately increased to avoid conflicts in water resource allocation. If extreme weather occurs and crop transpiration increases sharply, the short-term water demand will exceed the norm. In this case, the water demand weight of the current stage will be temporarily increased (e.g., the original flowering demand weight is 0.8, which can be increased to 0.95 on hot days) to ensure that emergency water demand is met first.

[0106] The decision tree multiplies the base score obtained from the previous layer with the water demand weight to obtain the score for the second layer before moving to the next layer node.

[0107] The third layer of nodes processes the area factor. The decision tree will adjust according to the ratio of farmland area to the benchmark area (1000 square meters): if the area is greater than 1000 square meters (e.g., 2000 square meters), the score will be multiplied by 2 (increased priority, as large areas of water shortage have a greater impact); if the area is less than 1000 square meters (e.g., 500 square meters), the score will be multiplied by 0.5 (decreased priority).

[0108] The fourth-layer node incorporates historical irrigation weights and determines whether there are irrigation records in the last 7 days by judging the branch: if there is no irrigation, it is multiplied by an additive coefficient of 1.2; if there is irrigation (e.g., irrigation was carried out 3 days ago), it is multiplied by a decay coefficient of 0.8, thus reflecting the cumulative water loss.

[0109] Finally, the terminal node of the decision tree will output the final priority score and compare it with the preset threshold of 8 to determine whether the area is a preliminary irrigation area. Areas with a score higher than the preset threshold of 8 are considered preliminary irrigation areas.

[0110] Based on the initial irrigation area, further prioritize areas that urgently need irrigation and will have better irrigation effects, avoiding irrigating less urgent areas first and causing water shortages in critical areas. During this process, focus on the urgency of water need (whether water shortage will immediately affect yield) and irrigation feasibility (whether irrigation is possible and whether it will be effective). Both types of data need to be quantified into scoreable indicators. The following indicators serve as examples:

[0111] Duration of water shortage:

[0112] Water shortage ≤1 day: 0.3 (Minor impact, can be restored by short-term water replenishment)

[0113] Water shortage for 2-3 days: 0.7 (Slight root damage, watering needed as soon as possible)

[0114] Water shortage for ≥4 days: 1.0 (Roots may die; failure to replenish water in time will reduce yield)

[0115] Soil water retention capacity:

[0116] Sandy soil / sandy loam (poor water retention): 1.0 (humidity decreases by 2%-3% daily, requiring emergency water replenishment)

[0117] Loam (medium water retention): 0.8 (humidity decreases by 1%-2% daily, requiring rapid water replenishment)

[0118] Clay / clay loam (good water retention): 0.6 (If the humidity decreases by less than 1% per day, water can be added gradually)

[0119] Water loss rate:

[0120] Loss rate ≤ 5% (new pipes, with insulation and sun protection): 1.0

[0121] Loss rate 6%-15% (slight pipe aging, no insulation): 0.8

[0122] Loss rate > 15% (severe pipe aging, exposed to sunlight): 0.6

[0123] Then, a weighted score is calculated, with weights allocated according to "water urgency 60% and irrigation feasibility 40%". Water urgency corresponds to the duration of water shortage and soil water retention capacity. The weight of the duration of water shortage can be set to 40%, and the weight of soil water retention capacity can be set to 20%. Irrigation feasibility corresponds to the water loss rate. The comprehensive score of each preliminary irrigation area is calculated according to the weights.

[0124] Based on the overall score, the areas are sorted from highest to lowest to form a priority irrigation area ranking list. For example, area A has an overall score of 9.5, area B has 9.0, and area C has 8.2, so the ranking is A > B > C.

[0125] Geographic location data for each region is obtained from a Geographic Information System (GIS) database, including its latitude and longitude coordinates, plot boundaries, plot area, and irrigation zone. This geographic location data is then correlated with region labels in a high-priority sorting list to create a priority-location mapping table. Using GIS visualization tools, the correlated priority-location data is overlaid onto an electronic farmland map, with labeling rules highlighting priority differences and core information.

[0126] Priority is distinguished by using a dual coding system of "color + symbol size". High priority (score ≥ 9.0) is marked with a red circle, and medium priority (8.0 ≤ score < 9.0) is marked with a yellow circle. The size of the circle is positively correlated with the area of ​​the plot (e.g., plot A has an area of ​​1000㎡ and a circle diameter of 8mm; plot B has an area of ​​800㎡ and a circle diameter of 6mm), which intuitively reflects the scale of the plot.

[0127] Information pop-up: When the mouse hovers over or clicks on a marker, the core information of that area pops up (e.g., "Plot A: Priority 9.5, Soil Moisture 20%, Water Requirement 10m³"). 3 (e.g., "equipment type: sprinkler irrigation"), which allows managers to quickly obtain details.

[0128] Auxiliary layers: Overlay irrigation equipment distribution layers (e.g., sprinkler head locations marked with blue triangles) and water source distribution layers (e.g., reservoirs marked with blue polygons) to provide a reference for "regional correlation" during subsequent clustering.

[0129] The final spatial distribution map clearly shows that the red areas are concentrated on the northwest side of the map (A, B), and the yellow areas are concentrated on the southeast side (C, D), providing an intuitive basis for spatial grouping during clustering.

[0130] Next, K-means clustering is performed. The core of K-means clustering is to group spatially close areas with similar soil moisture (irrigation demand) into several large partitions, solving the problem of "how to efficiently schedule irrigation in scattered high-priority areas". First, the K value is set, which is the final number of priority partitions to be divided. The K value needs to be determined based on the coverage of irrigation equipment and the water supply capacity of the water source, and is set to 2. The latitude and longitude coordinates of each area and the difference between the soil moisture of each area and the preset threshold are used as input feature variables to initialize the cluster centers. The Euclidean distance from each area to the cluster center is calculated, and the cluster centers are updated according to the Euclidean distance. The affiliation of each area is recalculated using the new centers. If the grouping results are completely consistent with the previous round, the clustering converges and the iteration stops; if they are inconsistent, the centers continue to be updated until the grouping is stable.

[0131] After clustering to obtain two partitions, it is necessary to further determine "which partition is the higher priority irrigation area". This can be done by averaging the comprehensive scores of the irrigation areas within the partition, sorting them, and finally outputting the higher priority irrigation area.

[0132] In some embodiments, based on the above embodiments, the step of using the K-means clustering algorithm to partition the high-priority irrigation area, and obtaining the sub-partition attribute description by calculating the average soil moisture value and average crop growth stage value of the sub-partition, includes:

[0133] The soil moisture and crop growth stage of the high-priority irrigation area are obtained and integrated to obtain a standardized data set.

[0134] The standardized dataset was subjected to cluster analysis using the K-means clustering algorithm to determine sub-partitions;

[0135] The sub-region attribute description is obtained by calculating the average soil moisture value and average crop growth stage value of the sub-region.

[0136] First, collect two core data points from farmland (e.g., 1000 plots) in the high-priority irrigation area: soil moisture (in percentage) and crop growth stage (numerical representation, range 0-1). Soil moisture and crop growth stage can be obtained from previous steps.

[0137] Soil moisture and crop growth stage data were integrated into a single raw dataset. Due to the significant difference in the units of measurement between the two types of data, preprocessing was required to eliminate unit interference. Z-score normalization was employed to normalize all data to the [0,1] interval, using the formula x'=(x-μ) / σ.

[0138] Where x' is the normalized value, x is the original value, μ is the mean of the original data, and σ is the standard deviation of the original data.

[0139] Next, the K-means clustering algorithm is used to analyze the standardized dataset: First, based on the actual scale and management needs of the high-priority irrigation area, the number of clusters K is preset (e.g., K=3), and three centroids are randomly initialized (e.g., centroid 1 (0.5, 0.3, 0.2), centroid 2 (0.7, 0.6, 0.5), centroid 3 (0.3, 0.8, 0.4)). Then, the algorithm enters an iterative process: the Euclidean distance between the standardized data of each farmland and the three centroids is calculated, and the farmland is assigned to the category of the nearest centroid. After all farmland is assigned, the new centroid for each category is recalculated (taking the average value of all farmland data within that category), and the "assignment-update centroid" steps are repeated until the centroid positions stabilize and no longer change. Finally, three sub-partitions are determined, for example, sub-partition 1 contains 400 farmland, sub-partition 2 contains 350 farmland, and sub-partition 3 contains 250 farmland.

[0140] Finally, for each stabilized sub-region, the standardized data of its cluster centroids are first denormalized (restored to actual physical quantities), and then the average soil moisture, average crop growth stage, and average nutrient content of all farmland within that sub-region are calculated. For example, after denormalization, the centroid of sub-region 1 has an average soil moisture of 55%, an average crop growth stage of 0.35, and an average nutrient content of 250 mg / kg. Based on this, the attributes of this sub-region can be described as "moderate soil moisture, early crop growth stage". Similarly, the attribute descriptions for sub-region 2 ("high soil moisture, mid-crop growth stage") and sub-region 3 ("low soil moisture, near crop maturity") are obtained. These descriptions clearly reflect the core characteristics of each sub-region, providing a basis for subsequent irrigation scheduling.

[0141] It is worth noting that using the K-means to group spatially close areas with similar soil moisture into large zones and output high-priority irrigation areas is essentially about solving the problem of equipment or water waste caused by regional dispersion, starting from the efficiency of irrigation resource allocation. On the other hand, using the K-means to divide high-priority areas into sub-zones is essentially about solving the problem of improper irrigation caused by differences in demand within large zones, starting from the "precision of irrigation effect".

[0142] In some embodiments, based on the above embodiments, the step of matching a pre-established irrigation model database according to the sub-partition attribute description to obtain irrigation quantity parameters includes:

[0143] Obtain soil type data and terrain slope data for sub-regions;

[0144] Key features that directly affect irrigation volume are extracted from the sub-region attribute descriptions, and the soil type data and terrain slope data are integrated to obtain a key feature set;

[0145] Based on the key feature set, irrigation parameters are obtained by matching a pre-established irrigation model database using a cosine similarity algorithm.

[0146] Extract key features that directly affect irrigation volume from the sub-partition attribute descriptions. Features such as "soil type" and "topographic slope" that are not in the attribute descriptions but are present in the pre-established irrigation model database need to be supplemented from the sub-partition's basic database.

[0147] A basic database is a core data storage system for a specific domain (such as agriculture or water conservancy). It centrally manages the most fundamental and stable core data within that domain in a structured manner, providing data support for upper-level business systems. In agricultural irrigation scenarios, the basic database is essentially a digital twin of farmland's spatial attributes. By systematically collecting and storing the inherent characteristics of farmland, it achieves accurate mapping and efficient management of physical farmland. It typically contains the following five categories of data:

[0148] Spatial basic information, soil attribute data, topographic data, farmland infrastructure data, and management attribute data.

[0149] After obtaining the key features, the first step is to process non-numerical key features such as soil type and crop type. A text-to-number mapping table needs to be established in advance, and the mapping rules must be completely consistent with the coding in the irrigation model database (ensuring "language consistency" during subsequent matching). For example, for soil type, the database specifies "sandy loam = 1, clay loam = 2, loam = 3"; for crop type, it specifies "corn = 2, wheat = 3, rice = 4"; for crop growth stage (textual description corresponding to numerical range), it specifies "early growth stage = 0.2, middle growth stage = 0.5, late growth stage = 0.8". When the sub-region attribute description mentions "soil is sandy loam, crop is corn, in early growth stage", these textual descriptions are converted into numbers according to the mapping table: sandy loam → 1, corn → 2, early growth stage → 0.2.

[0150] Next, process the numerical features (quantified indicators such as slope and humidity). These features are numbers themselves and do not require encoding, but the units or precision must be standardized according to the database format. For example, if the sub-partition attributes "topography slope 2%" and "soil moisture 15%" are stored as percentage values ​​in the database, then the original quantified values ​​(2% and 15%) are directly retained. If the database stores them in decimal form (such as 0.02), then the format is converted synchronously to ensure consistency with the database dimensions.

[0151] Finally, the processed classification feature numbers and numerical feature quantification values ​​are arranged in a fixed order (consistent with the feature order recorded in the database) according to "soil type, crop type, growth stage, slope, humidity" to form a feature vector.

[0152] Subsequently, the cosine similarity algorithm was used to match a pre-established irrigation model database. This database integrates multiple types of real and valid data, including records of actual irrigation cases over many years in the region (such as irrigation amounts and effect feedback under different soils and crops), crop water requirement experimental data from agricultural research (such as the requirements of specific crop growth stages for soil moisture and irrigation amount), and observational data on regional topography and soil characteristics (such as water retention patterns under different slopes). After screening and verification, these data are formed into standardized record entries. Each record contains dimensions corresponding to the extracted features (soil type, crop type, slope, moisture, etc.) and corresponding irrigation amount parameters.

[0153] Calculate the cosine similarity between the feature vector of the sub-partition and the vector of each record in the database. The cosine similarity is defined as the cosine of the angle between two vectors. Its essence is to calculate the size of the angle between two vectors in space: the smaller the angle, the closer the similarity value is to 1, indicating that the features of the two vectors are more similar.

[0154] The formula is

[0155] cosθ=(A·B) / (||A||·||B||)

[0156] Where A is the input vector, B is the database vector, and cosθ is the cosine of the angle between A and B in space.

[0157] The formula finds the record with the highest matching degree (e.g., similarity 0.98), and its corresponding irrigation quantity parameter (e.g., 30 mm) is the preliminary irrigation quantity parameter for the current sub-region, achieving accurate matching based on historical experience and the characteristics of the current sub-region.

[0158] In some embodiments, based on the above embodiments, the step of updating the climate condition distribution of the sub-region through real-time weather data, and adjusting the irrigation quantity parameter based on the climate condition distribution using a linear regression model to obtain the adjusted irrigation quantity parameter includes:

[0159] After obtaining the preliminary irrigation parameters for the sub-regions, it is necessary to dynamically adjust them in conjunction with real-time weather data to ensure that the irrigation amount is adapted to the impact of the current climate on soil moisture and crop water requirements. The core is to achieve accurate adaptation by "updating the distribution of climate conditions + quantitative adjustment of linear regression models".

[0160] First, the system will collect key weather indicators that affect soil moisture consumption in real time, including real-time temperature, air humidity, rainfall, evaporation, etc. of sub-regions (e.g., the current temperature of a sub-region is 28℃, air humidity is 60%, rainfall in the past 24 hours is 5 mm, and evaporation is 8 mm). The system will then associate these data with the spatial identifier of the sub-regions and update the "climate condition distribution" of each sub-region. This will clarify the specific climate environment currently faced by the sub-region and avoid using uniform climate data for vague adjustments.

[0161] Next, using a pre-built linear regression model, the various indicators in the climate condition distribution are transformed into quantitative impacts on irrigation. The model sets the influence coefficients for each climate factor, such as temperature coefficient k1 = 0.01, meaning that irrigation needs to be increased by 0.01 times for every 1°C increase in temperature; humidity coefficient k2 = -0.005, meaning that irrigation needs to be reduced by 0.005 times for every 1% increase in humidity; rainfall coefficient k3 = -0.02, meaning that irrigation needs to be reduced by 0.02 times for every 1 mm increase in rainfall; and evaporation coefficient k4 = 0.015, meaning that irrigation needs to be increased by 0.015 times for every 1 mm increase in evaporation. These values ​​are then substituted into the formula for calculation.

[0162] Q a =Q b ×[1+k1×T+k2×H-k3×R+k4×E]

[0163] Among them, Q a The adjusted irrigation amount is expressed in millimeters; Q b The values ​​are: basal irrigation amount (in millimeters); T: real-time temperature of the sub-region (in °C); H: real-time air humidity of the sub-region (in %); R: real-time rainfall of the sub-region over the past 24 hours (in millimeters); E: real-time evaporation of the sub-region (in millimeters); k1: temperature coefficient (in 1 / °C); k2: humidity coefficient (in 1 / %); k3: rainfall coefficient (in 1 / mm); and k4: evaporation coefficient (in 1 / mm).

[0164] The adjusted irrigation parameters are finally obtained through the formula. This process essentially uses a linear regression model to quantify the "consumption / replenishment" effect of real-time climate on water, allowing the initial irrigation amount to dynamically adapt to the current environment. This avoids insufficient irrigation due to rising temperatures and increased evaporation, or excessive irrigation due to rainfall, further improving the accuracy of irrigation.

[0165] In some embodiments, based on the above embodiments, after updating the climate condition distribution of the sub-region through real-time weather data, adjusting the irrigation quantity parameter based on the climate condition distribution using a linear regression model, and obtaining the adjusted irrigation quantity parameter, the method further includes:

[0166] Recalculate the water demand priority score of the sub-partition. If the water demand priority score of the sub-partition meets the dynamic perception requirements, then use the rasterization method to generate the partition decision map.

[0167] Based on the zoning decision map, a linear regression algorithm is used to predict the water demand of sub-zoning areas, and an irrigation water allocation scheme is obtained.

[0168] When the adjusted irrigation parameters (e.g., 1200 liters / hour) exceed the preset upper limit threshold (e.g., 1000 liters / hour), the system triggers the information processing parameter reduction step. The core is to avoid excessive reduction that could lead to crop water shortage while meeting the threshold limit. The entire process is not a fixed percentage reduction, but rather a dynamic decision based on the actual water conditions of the farmland.

[0169] The water demand priority score for each sub-region is recalculated using a decision tree algorithm: The first-level nodes calculate the humidity difference and divide the data into different branches based on the magnitude of the difference; the second-level nodes introduce crop water demand weights and adjust the score according to the weight value during peak water demand periods; the third-level nodes process the area factor, and the decision tree adjusts the score based on the ratio of farmland area to the baseline area (1000 square meters): if the area is greater than 1000 square meters, the score is multiplied by 2; if it is less than 1000 square meters (e.g., 500 square meters), it is multiplied by 0.5; the fourth-level nodes introduce historical irrigation weights by checking for irrigation records in the last 7 days: if there is no irrigation, the score is multiplied by a 1.2 bonus factor; if there is irrigation (e.g., irrigated 3 days ago), the score is multiplied by a 0.8 decay factor. The terminal node of the decision tree outputs the final priority score. If the score is higher than the threshold required for dynamic sensing (e.g., 7.0), the requirement is considered met.

[0170] Once the requirements are met, a rasterization method is used to generate a zoning decision map: the sub-zones are mapped to a two-dimensional grid, and the humidity status is represented by color coding (e.g., red for arid areas, green for moderate areas, and blue for humid areas), and includes the coordinates, humidity classification, and irrigation recommendations for each sub-zone.

[0171] Finally, based on the decision map, the water demand of each sub-region is predicted by using a linear regression algorithm combined with features such as humidity classification and crop type.

[0172] Zoning decision maps provide structured input features for linear regression. The map clearly defines the humidity classification of each sub-zone, and soil type, area size, and crop type characteristics of each sub-zone can be obtained from the sub-zone's basic database. These features will serve as independent variables in the linear regression model, while the irrigation water requirement of each sub-zone (e.g., liters per square meter) will be the dependent variable.

[0173] The parameters of the linear regression model are not set arbitrarily, but are obtained through training with historical irrigation data. The system calls upon feature-water matching data for the same region and crop type in the past (e.g., in the past 3 years, 5 liters of water per square meter was needed to optimally restore soil moisture in drought-stricken cornfields; 2 liters per square meter was sufficient for moderately moist cornfields). It uses this historical data to fit a linear formula, for example:

[0174] Q = a × C h +b×C c+c×C s +d

[0175] Where Q is the irrigation water demand per unit area of ​​the sub-zone, that is, the irrigation water required per square meter of the sub-zone;

[0176] C h The humidity classification code for sub-regions is assigned based on the humidity classification of the zoning decision map, such as: arid region = 3, moderate region = 2, humid region = 1;

[0177] C c Encode the crop type for each sub-region, assigning values ​​based on the actual crops planted in the sub-region, such as: corn = 3, wheat = 2, rice = 1;

[0178] C s The soil type coefficient for sub-regions reflects the soil's water retention capacity. For example: sandy loam (poor water retention, high water demand) = 1.5, loam (medium water retention) = 1.0, clay loam (good water retention, low water demand) = 0.5.

[0179] a is the regression coefficient of the humidity classification code, b is the regression coefficient of the crop type code, c is the regression coefficient of the soil type coefficient, and d is the intercept term.

[0180] After obtaining the linear regression equation, the feature code of the specific sub-region can be substituted to calculate the theoretical water demand of that sub-region, which is consistent with the water demand standard of arid areas in actual cases.

[0181] Finally, an integrated irrigation water allocation scheme is obtained. The "water demand per unit area" of each sub-region is converted into "total water demand," and adjusted in conjunction with global resource constraints: for example, a drought-stricken sub-region with an area of ​​200 square meters, calculated at 5 liters / square meter, has a total demand of 1000 liters. After summing the total demand of all sub-regions, if it does not exceed the total regional water resources and the hourly water supply capacity of irrigation equipment, the allocation amount for each sub-region is directly determined; if there is a resource gap, the coefficients of the linear regression model are adjusted to ensure that the water needs of key sub-regions are met first. The final allocation scheme not only corresponds to the actual situation of each sub-region in the decision map, but also avoids the error of "estimating water based on experience" through quantitative calculation of linear regression, providing an actionable water quantity basis for precision irrigation.

[0182] like Figure 2 As shown in the figure, this embodiment of the invention provides an intelligent water-saving irrigation system assisted by unmanned aerial vehicle (UAV) remote sensing. Through integrated modular design, the system realizes dynamic intelligent zoning of farmland irrigation areas, thereby improving water resource utilization efficiency and crop yield.

[0183] Data Acquisition Module 10: As the data foundation for the entire process, a drone equipped with a 5-band multispectral camera scans 100 hectares of farmland daily from 8:00 to 10:00 at a flight altitude of 100 meters and a resolution of 0.1 meters per pixel. Simultaneously, it acquires soil moisture images (based on the reflectance of the near-infrared band 850nm and the green band, and calculates and inverts soil moisture through the normalized difference water index) and crop growth stage images (based on the reflectance of the red edge band 730nm, the red band, and the near-infrared band, and quantifies growth status through the normalized vegetation index). At the same time, it uses airborne sensors to collect climate data such as temperature (-10~40℃, accuracy 0.1℃), relative humidity (0~100%, accuracy 1%), and wind speed (0~20m / s, accuracy 0.1m / s). This raw data is directly transmitted to the next stage for regional classification.

[0184] Classification Module 20: First, the collected raw data is preprocessed. Sensors are deployed in the corresponding areas of the soil moisture images to record moisture data hourly (range 10%-50%, mean 30%, standard deviation 5%). After Z-score standardization to [-1,1], K-means clustering (K=3) is used to classify soil moisture categories. From the crop growth stage images, the color of the plots is matched according to the threshold (germination stage 0.2-0.4, growth stage 0.4-0.7, maturity stage >0.7) to determine the growth stage. Then, the entropy value of the growth stage is calculated using the entropy method. After standardizing the climate data, the core climate principal components are screened by dimensionality reduction through principal component analysis (PCA). Then, the soil moisture category, growth stage entropy value, and climate principal components are used as inputs to construct a random forest classification model to generate farmland area classification labels. These labels provide feature basis for subsequent high-priority irrigation area screening.

[0185] Priority Confirmation Module 30: Subsequently, high-priority irrigation areas are screened. When an area meets the criteria of "soil moisture below a preset threshold (e.g., 30%) and crops are in their peak water demand period," priority assessment is triggered. Scores are calculated using a decision tree algorithm. The first layer branches according to the moisture difference, the second layer introduces crop water demand weights, the third layer processes area factors, and the fourth layer incorporates the weights of irrigation history over the past 7 days. The terminal node outputs a score that is compared with the threshold of 8 to preliminarily determine the initial irrigation areas. The GPS module is connected to obtain the latitude and longitude and plot boundary data of each area and associate them with the high-priority sorting list. Through an industrial touch screen, the priority-location data is overlaid onto the farmland electronic base map, with priorities marked by "red (≥9.0 points) / yellow (8.0-9.0 points) circles." Clicking the pop-up window displays "plot number, moisture, water demand, and equipment type," while simultaneously overlaying sprinkler head (blue triangle icon) and reservoir (blue polygon icon) layers. Using an AI acceleration module, with latitude and longitude and soil moisture difference as feature variables, the K-means clustering algorithm (K=2) is run. After iterative calculation until the grouping is stable, the mean comprehensive score of each partition is automatically calculated, higher priority partitions are determined, and the final high priority irrigation area results are output.

[0186] Partitioning Module 40: Performs sub-partitioning of high-priority irrigation areas, collects soil moisture (percentage) and crop growth stage (0-1 numerical value) of each farmland in the area, and standardizes the data to [0,1] to form a dataset; sets K=3 to randomly initialize 3 centroids in the feature space, assigns farmland to the nearest centroid through Euclidean distance, iteratively updates the centroids until stable, and obtains 3 sub-partitions; denormalizes the centroid data of each sub-partition, calculates the average soil moisture and average growth stage, and generates sub-partition attribute descriptions (such as "medium soil moisture, early crop growth"). These attribute descriptions provide sub-partition level features for subsequent irrigation quantity parameter matching.

[0187] Parameter matching module 50: Enters the preliminary irrigation quantity parameter matching of sub-regions, extracts dynamic features from the sub-region attribute descriptions, and supplements static features such as soil type and terrain slope from the basic database; for classification features such as soil type and crop type, it encodes them according to the "text-number" mapping table consistent with the irrigation model database; for numerical features such as slope and humidity, it unifies the units / precision and forms feature vectors in a fixed order of "soil type, crop type, growth stage, slope, humidity"; it uses the cosine similarity algorithm to calculate the similarity between this vector and the record vector in the database, selects the record with the highest matching degree (e.g., similarity 0.98), and uses its corresponding irrigation quantity parameter (e.g., 30 mm) as the preliminary irrigation quantity parameter to provide a basis for subsequent dynamic adjustment.

[0188] Adjustment Module 60: Collects real-time climate indicators such as temperature, air humidity, rainfall, and evaporation in sub-regions, and updates the "climate condition distribution" by associating them with the corresponding sub-regions through spatial identifiers; calls a pre-built linear regression model, substitutes the climate indicators into the formula, and calculates the adjusted irrigation parameters. These parameters serve as the core input of the irrigation control system, driving actuators such as solenoid valves and water pumps to precisely control the irrigation duration and flow rate according to the allocation plan. At the same time, the execution status is fed back through flow meters and pressure sensors, forming a closed loop of "parameter adjustment-execution-feedback", completing the final implementation from "parameter optimization" to "actual irrigation".

[0189] The UAV remote sensing-assisted intelligent water-saving irrigation system provided in this embodiment can improve water resource utilization efficiency and crop yield.

[0190] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A method for intelligent water-saving irrigation assisted by unmanned aerial vehicle (UAV) remote sensing, characterized in that, Executed by a computer, including: The farmland area was scanned using a multispectral sensor to obtain soil moisture images, crop growth stage images, and climate condition data; Based on the soil moisture image, crop growth stage image and climate condition data, the random forest algorithm is used for classification to obtain the classified farmland area labels. When the farmland area label shows that the soil moisture is lower than the preset moisture threshold and the crop growth stage is in the peak water demand period, the decision tree algorithm is used to calculate the water demand priority score, and based on the water demand priority score, the high priority irrigation area is determined. The high-priority irrigation area was divided into sub-regions using the K-means clustering algorithm. The sub-region attribute descriptions were obtained by calculating the average soil moisture value and average crop growth stage value of each sub-region. Based on the sub-partition attribute description, match the pre-established irrigation model database to obtain irrigation quantity parameters; The climate condition distribution of the sub-region is updated by real-time weather data. Based on the climate condition distribution, the irrigation parameters are adjusted by a linear regression model to obtain the adjusted irrigation parameters.

2. The intelligent water-saving irrigation method assisted by UAV remote sensing according to claim 1, characterized in that, The process involves classifying farmland areas using a random forest algorithm based on the soil moisture image, crop growth stage image, and climate condition data to obtain classified farmland area labels, including: Based on the soil moisture images, K-means clustering is used to classify the soil moisture categories. Based on the crop growth stage images, the information entropy is calculated using the entropy method to obtain the growth stage entropy value; Based on the climate condition data, principal component analysis was used to extract principal components to obtain climate principal components. Based on the soil moisture category, the growth stage entropy value, and the climate principal component, a random forest algorithm is used for classification to obtain the classified farmland area labels.

3. The intelligent water-saving irrigation method assisted by UAV remote sensing according to claim 1, characterized in that, When the farmland area label shows soil moisture below a preset moisture threshold and the crop is in its peak water demand stage, a decision tree algorithm is used to calculate a water demand priority score. Based on the water demand priority score, high-priority irrigation areas are determined, including: When the farmland area label shows that the soil moisture is lower than the preset moisture threshold and the crop growth stage is in the peak water demand period, the decision tree algorithm is used to calculate the water demand priority score, and the preliminary irrigation area is determined based on the water demand priority score. Based on the crop growth data and irrigation demand data of the preliminary irrigation area, a weighted score is used to obtain a ranking list of high-priority irrigation areas; Based on the sorted list of high-priority irrigation areas, obtain the corresponding geographical location data and generate a spatial distribution map of the irrigation areas; Based on the spatial distribution map and soil moisture, the K-means clustering algorithm is used to divide the irrigation area into priority zones to obtain high-priority irrigation zones.

4. The intelligent water-saving irrigation method assisted by UAV remote sensing according to claim 1, characterized in that, The high-priority irrigation area is divided into sub-regions using the K-means clustering algorithm. By calculating the average soil moisture value and average crop growth stage value of each sub-region, a description of the sub-region's attributes is obtained, including: The soil moisture and crop growth stage of the high-priority irrigation area are obtained and integrated to obtain a standardized data set. The standardized dataset was subjected to cluster analysis using the K-means clustering algorithm to determine sub-partitions; The sub-region attribute description is obtained by calculating the average soil moisture value and average crop growth stage value of the sub-region.

5. The intelligent water-saving irrigation method assisted by UAV remote sensing according to claim 1, characterized in that, The step of matching the pre-established irrigation model database with the sub-partition attribute description to obtain irrigation quantity parameters includes: Obtain soil type data and terrain slope data for sub-regions; Key features that directly affect irrigation volume are extracted from the sub-region attribute descriptions, and the soil type data and terrain slope data are integrated to obtain a key feature set; Based on the key feature set, irrigation parameters are obtained by matching a pre-established irrigation model database using a cosine similarity algorithm.

6. The intelligent water-saving irrigation method assisted by UAV remote sensing according to claim 1, characterized in that, The process of updating the climate condition distribution of the sub-region using real-time weather data, adjusting the irrigation quantity parameters based on the climate condition distribution using a linear regression model, and obtaining the adjusted irrigation quantity parameters further includes: Recalculate the water demand priority score of the sub-partition. If the water demand priority score of the sub-partition meets the dynamic perception requirements, then use the rasterization method to generate the partition decision map. Based on the zoning decision map, a linear regression algorithm is used to predict the water demand of sub-zoning areas, and an irrigation water allocation scheme is obtained.

7. A drone-based remote sensing-assisted intelligent water-saving irrigation system, characterized in that, include: The data acquisition module is used to scan farmland areas using a multispectral sensor to acquire soil moisture images, crop growth stage images, and climate condition data; The classification module is used to classify farmland areas based on the soil moisture image, crop growth stage image and climate condition data using a random forest algorithm to obtain classified farmland area labels. The priority confirmation module is used to calculate the water demand priority score using a decision tree algorithm when the farmland area label shows that the soil moisture is lower than the preset moisture threshold and the crop growth stage is in the peak water demand period. Based on the water demand priority score, the high priority irrigation area is determined. The partitioning module is used to partition the high-priority irrigation area using the K-means clustering algorithm. By calculating the average soil moisture value and average crop growth stage value of the sub-partitions, the attribute description of the sub-partitions is obtained. The parameter matching module is used to match the pre-established irrigation model database according to the sub-partition attribute description to obtain irrigation quantity parameters; The adjustment module is used to update the climate condition distribution of the sub-region through real-time weather data, and adjust the irrigation quantity parameters based on the climate condition distribution through a linear regression model to obtain the adjusted irrigation quantity parameters.

Images