Irrigation water utilization coefficient measuring and calculating method based on multi-source remote sensing information

Abstract

The invention discloses an irrigation water utilization coefficient measuring and calculating method based on multi-source remote sensing information. The method comprises the following steps: acquiring total water diversion amount data, first remote sensing data, second remote sensing data and irrigation data for a target area; carrying out space-time fusion on the first remote sensing data and the second remote sensing data, and then inputting the first remote sensing data and the second remote sensing data into a pre-trained irrigation inversion model to obtain irrigation state distribution data; performing consistency matching on the irrigation data and the irrigation state distribution data to determine a first difference degree; judging whether the first difference degree is greater than a preset threshold value; otherwise, judging that the irrigation state distribution data is inaccurate, and finely adjusting the irrigation inversion model; and measuring and calculating the irrigation water utilization coefficient of the target area by combining the data output by the fine-tuned irrigation inversion model with the total water diversion amount data. According to the invention, the accuracy of irrigation water utilization coefficient measurement and calculation is improved, and high-frequency and automatic irrigation monitoring of a large-scale agricultural area is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of irrigation monitoring, in particular to a method for calculating an irrigation water utilization coefficient based on multi-source remote sensing information. Background Art

[0002] Irrigation Water Use Efficiency (IWUE) is a core indicator for evaluating irrigation efficiency. It represents the proportion of effective water actually consumed by farmland to the total water drawn from water sources. It provides a quantitative basis for scientifically evaluating the water use efficiency of irrigation systems, identifying water resource waste, and optimizing irrigation management, improving agricultural sustainability, and supporting water resource policy formulation.

[0003] Using remote sensing information to estimate farmland IWUE is one of the existing measurement methods. However, in actual application, there are temporal and spatial contradictions in remote sensing data. That is, medium- and high-resolution products have good accuracy under clear sky conditions, but are greatly affected by clouds, resulting in data missing and requiring time series interpolation, which affects continuity. Low-resolution products have good temporal continuity but are spatially coarse, making it difficult to accurately capture changes in farmland moisture characteristics at the field scale. This can easily introduce large errors or cause spatial information discontinuity, affecting the assessment of irrigation status.

[0004] In particular, existing remote sensing IWUE assessment methods are generally based on regional scales. This coarse-scale assessment fails to accurately reflect the actual irrigation conditions within the target farmland, leading to biased assessment results. When remote sensing data quality is affected by environmental factors such as cloud cover, these spatial biases and temporal continuity issues are further exacerbated, reducing the accuracy of IWUE estimates. This, in turn, affects the accuracy of irrigation water use coefficient calculations, weakening their support for irrigation system optimization and water resources management. Summary of the Invention

[0005] In view of the problems of data feature fragmentation and coarse-scale evaluation in existing evaluation methods, the present invention provides an irrigation water utilization coefficient calculation method based on multi-source remote sensing information.

[0006] In order to achieve the above object, the technical solution of the present invention is as follows:

[0007] In a first aspect, the present application discloses a method for calculating irrigation water utilization coefficient based on multi-source remote sensing information, comprising the following steps:

[0008] Obtaining total water diversion data for a target area, as well as first remote sensing data collected by a multispectral imaging satellite, second remote sensing data collected by an imaging spectrometer, and irrigation data uploaded by a mobile terminal;

[0009] fusing the first remote sensing data and the second remote sensing data in time and space to generate spatiotemporally continuous moisture characteristic data;

[0010] Inputting the moisture characteristic data into a pre-trained irrigation inversion model to obtain irrigation status distribution data of the target area;

[0011] Performing consistency matching on the irrigation data and the irrigation status distribution data to determine a first difference;

[0012] determining whether the first difference is greater than a preset threshold, and if so, determining that the irrigation status distribution data is accurate;

[0013] Otherwise, the irrigation status distribution data is determined to be inaccurate, the irrigation time when the mobile terminal uploaded the irrigation data is extracted and used as a supervision signal, and the historical accuracy of the data uploaded by the mobile terminal is obtained, and the confidence factor is calculated in a decentralized manner;

[0014] Fine-tuning the irrigation inversion model through supervised machine learning using the supervision signal and the confidence factor as fine-tuning parameters;

[0015] The irrigation state distribution data output by the fine-tuned irrigation inversion model is combined with the total water diversion data to calculate the irrigation water utilization coefficient of the target area.

[0016] In a second aspect, the present application discloses an irrigation water utilization coefficient calculation system based on multi-source remote sensing information, comprising:

[0017] A data acquisition module is used to acquire total water diversion data for a target area, as well as first remote sensing data collected by a multispectral imaging satellite, second remote sensing data collected by an imaging spectrometer, and irrigation data uploaded by a mobile terminal;

[0018] a data fusion module, configured to perform spatiotemporal fusion of the first remote sensing data and the second remote sensing data to generate spatiotemporally continuous moisture characteristic data;

[0019] a data processing module, configured to input the moisture characteristic data into a pre-trained irrigation inversion model to obtain irrigation status distribution data of the target area;

[0020] a data matching module, configured to perform consistency matching between the irrigation data and the irrigation status distribution data to determine a first difference;

[0021] a data judgment module, configured to judge whether the first difference is greater than a preset threshold, and if so, to judge that the irrigation status distribution data is accurate; otherwise, to judge that the irrigation status distribution data is inaccurate, extract the irrigation time when the mobile terminal uploaded the irrigation data and use it as a supervision signal, obtain the historical accuracy of the data uploaded by the mobile terminal, and calculate the confidence factor in a decentralized manner;

[0022] a data fine-tuning module, configured to fine-tune the irrigation inversion model through supervised machine learning using the supervisory signal and the confidence factor as fine-tuning parameters;

[0023] The result output module is used to combine the irrigation state distribution data output by the fine-tuned irrigation inversion model with the total water diversion data to calculate the irrigation water utilization coefficient of the target area.

[0024] In a third aspect, the present application discloses a computer terminal comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of the aforementioned method for calculating the irrigation water utilization coefficient based on multi-source remote sensing information are implemented.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] 1. This application integrates multispectral remote sensing, imaging spectrometer data, and ground-based uploaded data to construct an irrigation status recognition model based on multi-source heterogeneous data, thereby improving the accuracy of irrigation water utilization coefficient calculation and achieving high-frequency, automated irrigation monitoring over large agricultural areas.

[0027] 2. Introducing a supervised learning mechanism and confidence factor control, combined with feedback fine-tuning based on end-user data, to form an inversion model with adaptive evolutionary capabilities; this not only enables spatial identification of whether irrigation has occurred, but also further completes the quantitative assessment of water resource utilization efficiency; it assists in water resource allocation, agricultural precision management, and the implementation of the water rights system; and lays the foundation for building a smart agricultural remote sensing monitoring platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The disclosure of the present invention is described with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of the present invention. In the drawings, the same reference numerals are used to refer to the same components. Among them:

[0029] Figure 1 This is a flow chart of the irrigation water utilization coefficient calculation method based on multi-source remote sensing information introduced in this application;

[0030] Figure 2 Based on Figure 1 A logic flow chart of first remote sensing data acquisition;

[0031] Figure 3 Based on Figure 2 The logic flow chart of assessing the spatiotemporal importance score of an image;

[0032] Figure 4 Based on Figure 2 The logical flow chart of retrieving the minimum cloud cover image;

[0033] Figure 5 Based on Figure 1 A logical flow chart for correcting remote sensing irrigation boundaries;

[0034] Figure 6 Based on Figure 1 The logic flow chart of calculating confidence factor;

[0035] Figure 7 This is a scenario application diagram of the irrigation water utilization coefficient calculation method based on multi-source remote sensing information;

[0036] Figure 8 This is a block diagram of the irrigation water utilization coefficient calculation system based on multi-source remote sensing information;

[0037] Figure 9 Based on Figure 1 An architectural block diagram of a computer terminal. DETAILED DESCRIPTION

[0038] It is easy to understand that according to the technical solution of the present invention, without changing the essential spirit of the present invention, a person skilled in the art can propose a variety of interchangeable structural modes and implementation modes. Therefore, the following specific embodiments and drawings are only exemplary descriptions of the technical solution of the present invention and should not be regarded as the entire invention or as a limitation or restriction of the technical solution of the present invention.

[0039] Application Overview

[0040] In existing technologies, the calculation of irrigation water utilization coefficients mainly relies on a single remote sensing data source. However, medium- and high-resolution satellite images are easily obscured by clouds, resulting in data loss. Low-resolution data, while having strong temporal continuity, cannot accurately reflect changes in field scale. Existing methods often use coarse-scale assessments, which cannot accurately capture irrigation differences within farmland. When remote sensing data is affected by clouds, spatial deviations and temporal discontinuities are superimposed, resulting in increased errors in irrigation status identification and affecting the accuracy of coefficient calculations. For example, when a certain irrigation area was evaluated using a single satellite image, due to continuous rainy weather, data for the key growing period was missing, forcing the use of interpolation estimation. The final calculated result deviated significantly from the actual water consumption, exceeding the allowable error range.

[0041] To address the above issues, actual research has found that multi-source data fusion and dynamic feedback mechanisms can synergistically improve measurement accuracy. First, to address the spatiotemporal contradictions in remote sensing data, the complementary fusion of medium- and high-resolution observations from multispectral satellites and high-frequency observations from imaging spectrometers is considered to generate spatiotemporally continuous moisture characteristic data. Second, the irrigation data uploaded by mobile terminals contains the actual irrigation time and location, which can be used as ground truth to verify the model output, but the problem of terminal data credibility needs to be addressed. Furthermore, when the model output differs significantly from the terminal data, how to dynamically adjust the model parameters becomes the key. By introducing a confidence factor to quantify the reliability of the terminal data and combining it with supervisory signals to fine-tune the model, a closed-loop optimization is formed, ultimately achieving accurate measurement.

[0042] After introducing the basic concept of the present invention, embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0043] Exemplary Methods

[0044] like Figure 1 As shown, this embodiment introduces a method for calculating the irrigation water utilization coefficient based on multi-source remote sensing information, including the following steps:

[0045] 101. Obtaining total water diversion data for the target area, first remote sensing data collected by a multispectral imaging satellite, second remote sensing data collected by an imaging spectrometer, and irrigation data uploaded by a mobile terminal;

[0046] 102. Temporally and spatially fuse the first remote sensing data with the second remote sensing data to generate temporally and spatially continuous moisture characteristic data;

[0047] 103. Input the water characteristic data into the pre-trained irrigation inversion model to obtain the irrigation status distribution data of the target area;

[0048] 104. Perform consistency matching on the irrigation data and the irrigation status distribution data to determine a first difference;

[0049] 105. Determine whether the first difference is greater than a preset threshold, and if so, determine that the irrigation status distribution data is accurate;

[0050] Otherwise, the irrigation status distribution data is determined to be inaccurate. The irrigation time of the irrigation data uploaded by the mobile terminal is extracted and used as a supervision signal. The historical accuracy of the data uploaded by the mobile terminal is also obtained, and the confidence factor is calculated in a decentralized manner.

[0051] 107.Using supervision signal and confidence factor as fine-tuning parameters to fine-tune the irrigation inversion model through supervised machine learning;

[0052] 108. The irrigation water utilization coefficient of the target area is calculated by combining the irrigation status distribution data output by the fine-tuned irrigation inversion model with the total water diversion data.

[0053] Among them, moisture characteristic data include parameters such as soil moisture content and vegetation water stress index, which are used to characterize the temporal and spatial distribution of farmland moisture.

[0054] Specifically, the medium- and high-resolution data provided by multispectral imaging satellites and the high-frequency data from imaging spectrometers are fused in space and time to generate a continuous water feature dataset, breaking through the space-time limitations of a single data source. The irrigation inversion model outputs irrigation status distribution data based on the fused data, but this may lead to local deviations due to cloud interference or model errors. The irrigation data uploaded by the mobile terminal contains the actual irrigation coordinates and time, as well as the irrigation type and intensity. Through spatial overlay analysis with the data output by the model, a first difference is calculated. When the difference falls below the threshold, the model fine-tuning mechanism is triggered: the irrigation time is extracted as a supervisory signal to constrain the model's temporal feature learning. At the same time, the confidence factor is calculated based on the accuracy of the terminal's historical data and the time-attenuation factor, and the supervisory signal weight is dynamically adjusted. The fine-tuned model outputs irrigation distribution data again, which is finally combined with the total water diversion data to calculate the irrigation water utilization coefficient by the ratio of effective water consumption to total water withdrawal.

[0055] Therefore, this embodiment constructs a spatiotemporally continuous moisture signature through multi-source data fusion, integrating it with mobile terminal data for dual verification and dynamic model optimization. Compared to existing technologies that use static models to process cloud-interfered areas, relying solely on interpolation of historical data, this embodiment, upon detecting data anomalies, can fine-tune the model using real-time terminal data, improving localized measurement accuracy. Furthermore, the design of the confidence factor effectively balances the timeliness and reliability of terminal data, preventing model performance degradation caused by individual erroneous data reports.

[0056] This embodiment effectively addresses the difficulty of single-data source assessment methods in balancing spatiotemporal resolution and data integrity. By integrating multiple sources, it obtains high-precision, spatiotemporally continuous water signature data. Combining mobile terminal data with a dynamic model optimization mechanism improves the spatial accuracy and temporal consistency of irrigation status identification, ensuring that irrigation water utilization coefficient calculations are more accurately aligned with actual water use. This approach maintains stable output even in scenarios with fluctuating data quality, providing reliable technical support for precision agricultural irrigation management.

[0057] The overall solution of this embodiment is introduced above. In order to further understand the solution of this embodiment, the following describes in detail the process of 101 obtaining the total water diversion data for the target area, the first remote sensing data collected by the multispectral imaging satellite, the second remote sensing data collected by the imaging spectrometer, and the irrigation data uploaded by the mobile terminal.

[0058] The first remote sensing data is acquired on a daily or weekly basis, and the second remote sensing data is acquired on an hourly basis. The irrigation data uploaded by the mobile terminal includes the location information of the target area, irrigation time information, irrigation type, and irrigation intensity, and is stored in a log database.

[0059] The primary remote sensing data acquired at the daily or weekly level can be obtained using Landsat satellites, which have high spatial resolution but are susceptible to cloud interference. Daily or weekly time intervals are used to balance data quality and observation frequency. Secondary remote sensing data acquired at the hourly level is obtained using imaging spectrometers, such as MODIS or VIIRS sensors. These sensors have high temporal resolution but low spatial resolution, and are used to capture short-term dynamic changes in farmland moisture characteristics. Irrigation data uploaded by mobile terminals refers to structured data uploaded by field equipment (such as drones) or manual terminals. Specifically, GPS positioning modules are used to record location information, timestamps are used to record irrigation time, pre-set classification tags are used to identify irrigation types, and sensors are used to measure irrigation intensity. This data is then uploaded to a log database via an IoT communication protocol, forming standardized spatiotemporal data records. This log database can be implemented using InfluxDB or TimescaleDB, and a joint time-space index is established to enable rapid retrieval and correlation analysis of multi-dimensional data.

[0060] As a result, daily / weekly and hourly remote sensing data form a complementary observation system, which not only retains the advantages of high spatial resolution but also maintains temporal continuity; mobile terminal data provides high-precision ground truth for model training and verification through structured storage, solving the problem of spatiotemporal alignment of multi-source data.

[0061] Daily and weekly data provide a baseline spatial framework, while temporal data fill in temporal gaps, forming a continuous and consistent dataset of farmland moisture characteristics. Structured storage of mobile terminal data and a log database indexing mechanism transform manually reported information into machine-interpretable spatiotemporal event sequences, improving the accuracy of its matching with remote sensing inversion results. The synergy between multi-timescale data and standardized ground data effectively reduces fusion errors caused by inconsistent temporal granularity, providing high-confidence multi-source data support for irrigation status identification.

[0062] In addition, combined Figure 2 , the specific steps of obtaining the first remote sensing data collected by the multispectral imaging satellite are as follows:

[0063] Obtain images of the target area and their corresponding cloud masks collected by multispectral imaging satellites;

[0064] The overall cloud coverage of the image is calculated based on the cloud mask corresponding to the image. The cloud cover level of the image is then determined based on the preset multi-level cloud cover threshold. The following decisions are made based on the cloud cover level:

[0065] (a1) if the cloud cover level of the image is low cloud level, directly using the image as the first remote sensing data;

[0066] (a2) if the cloud cover level of the image is medium cloud level, performing local interpolation processing on the cloud-blocked area in the image, and using the processed image as the first remote sensing data;

[0067] (a3) If the cloud cover level of the image is high cloud level, extracting effective features around the image, performing spatial enhancement processing on the image in combination with the second remote sensing data collected by the imaging spectrometer, and using the processed image as the first remote sensing data;

[0068] (a4) If the cloud cover level of the image is very high, assess its spatiotemporal importance score based on the time period and region in which the image is located;

[0069] Determine whether the spatiotemporal importance score is greater than a preset spatiotemporal threshold; if so, retrieve an image of the lowest cloud cover in the same area at a similar time from a historical image library, perform spatial enhancement processing on the image combined with the second remote sensing data collected by the imaging spectrometer, and use the processed image as the first remote sensing data;

[0070] Otherwise the image is discarded.

[0071] The cloud mask refers to the binary mask data corresponding to the cloud coverage area identified by image processing technology. Specifically, it can be implemented by multispectral threshold segmentation combined with morphological filtering algorithm to accurately quantify the cloud coverage in the image.

[0072] To understand the multi-level cloud cover thresholds and corresponding operations, please refer to the following table:

[0073] Table 1: Comparison table of multi-level cloud cover thresholds

[0074] Cloud cover level Cloud coverage range Treatment Low cloud level ＜10% Directly use the original image Mid-cloud level 10%~30% Local interpolation High cloud level 30%~50% Multi-source data fusion Ultra-high cloud level >50% Select to exclude or trigger replacement of adjacent date data based on temporal and spatial importance

[0075] Regarding the local interpolation process used in (a2), Kriging interpolation or the neighboring pixel mean filling algorithm can be used to repair the cloud-obscured areas, using the information of the surrounding valid pixels to restore the moisture characteristics of the missing areas. This maximizes the original image information and avoids information loss.

[0076] Regarding the spatial enhancement processing used in (a3), specifically, the feature fusion of the effective surrounding features of the multispectral imaging satellite and the high spatiotemporal resolution data of the imaging spectrometer can be performed. Feature fusion can be performed based on CNN and Transformer, or weighted averaging, PCA-based fusion, and Wavelet fusion can be used to improve the data integrity of the cloud coverage area and retain the environmental characteristics of the current phase as much as possible.

[0077] Specifically, after the cloud mask is generated, the cloud cover ratio is calculated to determine the cloud level. For low-cloud-level images, the original data is retained directly to reduce processing overhead; for medium-cloud-level images, cloud-blocked areas are restored through local interpolation to avoid the overall unavailability of data due to small-scale cloud layers; for high-cloud-level images, spatial details are supplemented by fusing imaging spectrometer data to solve the problem of missing information caused by large-scale cloud cover. When the cloud cover reaches an ultra-high level, a spatiotemporal importance scoring mechanism is introduced: if the score exceeds the threshold, it indicates that the data for that period or region has critical value. At this time, the historical lowest cloud cover image for the same area is retrieved and fused with the imaging spectrometer data to ensure the availability of key data; if the score does not reach the threshold, invalid data is directly eliminated to avoid introducing errors. Through a hierarchical processing strategy, the utilization rate of remote sensing data is maximized while ensuring data quality.

[0078] This method addresses the issue of missing or degraded data in multispectral satellite imagery due to cloud cover. A hierarchical processing strategy employs differentiated restoration methods for different cloud cover levels, minimizing data loss while improving restoration accuracy. A spatiotemporal importance scoring mechanism ensures data availability during critical periods and regions, preventing errors introduced by processing invalid data. This improves the accuracy and reliability of remote sensing data required for irrigation status inversion, providing a high-quality data foundation for calculating irrigation water utilization coefficients.

[0079] The above describes the specific steps for obtaining the first remote sensing data. The following describes in detail how to assess the spatiotemporal importance score of the first remote sensing data.

[0080] like Figure 3 As shown, the specific steps for evaluating the spatiotemporal importance score of the image according to the time period and region in (a4) are as follows:

[0081] 201. Obtain the NDVI corresponding to the time point of the image and the period and adjacent periods of the time point;

[0082] 202. Determine whether the time point of the image is at a pre-set key time node, and if so, calculate the NDVI change rate based on the NDVI corresponding to the period at the time point and the adjacent period, and determine its time importance based on the NDVI change rate;

[0083] 203. Obtain the area where the image is located;

[0084] 204. Mapping the region where the image is located into a preset regional priority map, and determining the spatial importance of the overlapping region according to the corresponding level of the location of the overlapping region;

[0085] 205. Dynamically weight the temporal importance and the spatial importance to obtain a temporal and spatial importance score.

[0086] The NDVI rate of change mentioned in 202 refers to the magnitude of change in the Normalized Difference Vegetation Index (NDVI) between adjacent time periods. Specifically, it can be calculated as the ratio of the NDVI difference between adjacent periods to the time interval, and is used to quantify the characteristics of sudden changes in crop growth rates. Common pre-defined key time points include: key growth periods (such as heading and grain filling); disaster-prone periods (such as flood and drought monitoring); and policy-driven time windows (such as the end of a carbon monitoring cycle and remote sensing yield estimation).

[0087] The regional priority map mentioned in 204 refers to a pre-divided spatial distribution map of farmland management levels. Specifically, it can be generated by superimposing soil type, crop type, and irrigation facility density layers using a geographic information system to distinguish the urgency of irrigation monitoring in different regions.

[0088] Therefore, a spatiotemporal importance scoring model can be constructed as follows:

[0089] ;in, represents the total importance of the image at position x and time t, represents the time importance scoring function (0~1), represents the spatial importance scoring function (0~1), 、 Indicates an adjustable weight coefficient, which adjusts the weight distribution ratio based on the current time sensitivity and spatial priority.

[0090] Introducing a threshold, such as If the cloud cover is greater than 0.7 and the cloud cover is greater than 50%, the image must be replaced and cannot be directly removed. 0.7 is the preset spatiotemporal threshold and can be adjusted according to actual needs.

[0091] Specifically, when the image is during a critical period for crop water demand, the NDVI change rate of adjacent periods is calculated to identify sudden changes in vegetation growth and assign a higher temporal weight. Simultaneously, the image area is matched to the priority map, with higher-ranked areas receiving a higher spatial weight. The two are nonlinearly superimposed using a dynamic decentralization model. For example, in high-priority areas during the heading period, the temporal weight can reach 60% to 70%, while in low-priority areas during non-critical periods, the spatial weight is reduced to 30% to 40%. This mechanism ensures that data from sensitive crop growth periods is preserved even when cloud cover is excessively high, while prioritizing data integrity in key monitoring areas.

[0092] Compared to existing methods that rely solely on cloud cover to determine data availability, without considering differences in crop phenological stages and regional management needs, this implementation integrates NDVI temporal characteristics with spatial priority information to establish a two-dimensional assessment system, effectively identifying key spatiotemporal nodes with monitoring value even under high cloud cover conditions. This approach avoids relying solely on cloud cover thresholds, which often leads to the inadvertent deletion of data at key stages such as heading and grain filling. By capturing growth inflection points through the NDVI rate of change and dynamically adjusting regional priorities, data screening accuracy in high cloud cover scenarios is improved.

[0093] This approach addresses the challenge of assessing the spatiotemporal importance of remote sensing imagery under extremely high cloud cover conditions, preventing misjudgments of irrigation status due to mechanical exclusion of high-value images. By dynamically balancing temporal sensitivity and spatial priority, it ensures that effective information on critical periods of crop water demand and key monitoring areas is retained even in the absence of data. This provides a reliable data foundation for subsequent irrigation status inversion and improves the accuracy of irrigation water utilization coefficient calculations.

[0094] The above describes the specific steps for evaluating the spatiotemporal importance score. The following describes in detail how to retrieve the minimum cloud cover image from the historical image library.

[0095] like Figure 4 As shown in (a4), the specific steps for retrieving the minimum cloud cover image of the same area at a similar time from the historical image library are as follows:

[0096] 301. Retrieve historical images from the historical image library according to the preset proximity time length and sort them in ascending order according to cloud cover;

[0097] 302. Determine whether the cloud cover of the top N historical images is the same. If not, output the top 1 historical image as the image with the lowest cloud cover.

[0098] 303. If yes, extract the NDVI of the TopN historical images, calculate the correlation with the NDVI of the image one by one, and output the historical image with the largest correlation value as the lowest cloud cover image.

[0099] The proximity time length mentioned in 301 is a pre-set time range parameter, preferably 7 days or 14 days, which is used to limit the retrieval time span of historical images and ensure that the time difference between the selected image and the current image is within a reasonable range.

[0100] The correlation calculation mentioned in 303 can be done by using the vegetation index correlation calculation. Specifically:

[0101] Taking the preset approach time length of 7 days as an example, obtain the satellite image of the target date to be replaced (cloud coverage rate > 50%) and extract the vegetation index of the cloud-free area from it: ; where n is the number of cloud-free grids.

[0102] Qualified images of the same area within the past 7 days (cloud coverage ≤ 30%) were retrieved from the historical image library. The raster resolution of the candidate images was unified with that of the target date images through geographic coordinate conversion. The NDVI values ​​of the corresponding positions of each candidate image were simultaneously extracted within the cloud-free area of ​​the target date.

[0103] Correlations were calculated using the weighted Pearson correlation coefficient:

[0104]

[0105] in, Indicates the NDVI of the i-th cloud-free grid in the target date image (the cloud-polluted image to be replaced); In the candidate image j, The raster NDVI value at the same geographic coordinate location (j represents the candidate image number); represents the mean NDVI value on the target date; represents the mean NDVI value of candidate image j; The larger the ∈[-1,1] value is, the closer the vegetation growth status is to the target date.

[0106] Weight design: . represents the time decay factor, represents the acquisition time of candidate image j.

[0107] Then, the candidate image corresponding to the maximum correlation coefficient is selected as the minimum cloud cover image output.

[0108] It should be noted that if all is less than the empirical threshold), the backup strategy is activated: the candidate image with the closest time is selected first, or if there are multiple images with equal time intervals, the one with the lowest cloud cover is selected.

[0109] Specifically, when multiple historical images share the same minimum cloud cover, NDVI correlation calculations are used to further select the image that best matches the current farmland moisture characteristics. For example, if the NDVI trend of a historical image is highly consistent with that of the current image, indicating similar vegetation cover, this image is preferred for subsequent processing. This minimizes cloud interference and further reduces the risk of misjudging farmland moisture characteristics due to differences in vegetation growth.

[0110] Due to the problem in existing technologies that when a historical image is close in time but in a different stage of crop growth, its NDVI is quite different from the current image, and direct selection may lead to deviations in the inversion of irrigation status. This embodiment effectively identifies the historical image that best matches the current farmland ecological characteristics by introducing NDVI correlation calculation, avoiding errors introduced by differences in vegetation cover. It solves the problem that historical images cannot be effectively selected under the same cloud cover, ensuring that the selected image meets the requirements of minimizing cloud interference and accurately reflects the current moisture characteristics of the farmland. For example, during the critical period of crop growth, even if the cloud cover is the same, the historical image closest to the current crop growth can be screened out through NDVI correlation, thereby improving the accuracy of irrigation status distribution data.

[0111] like Figure 5 As shown above, the specific steps of retrieving the minimum cloud cover image are introduced. The following is a detailed description of the remote sensing irrigation boundary correction before the spatiotemporal fusion of the first remote sensing data and the second remote sensing data.

[0112] 401. Pre-process the irrigation data uploaded by the mobile terminal, identify irrigation coordinate points using a spatial density-based clustering scheme, construct the irrigation coordinate points into a minimum circumscribed polygon, and obtain a reference irrigation boundary area;

[0113] 402. Performing spatiotemporal association between the first remote sensing data and the second remote sensing data and the preliminary irrigation boundary area to identify remote sensing irrigation boundaries of the first remote sensing data and the second remote sensing data;

[0114] 403. Determine whether the remote sensing irrigation boundary is consistent with the reference irrigation boundary area; otherwise, modify the remote sensing irrigation boundary with reference to the reference irrigation boundary area and historical data.

[0115] The preprocessing in 401 specifically involves cleaning the irrigation data, removing duplicate data, missing fields, and data with GPS deviations greater than a threshold, aggregating or standardizing timestamps, applying historical accuracy as a confidence factor, and filtering low-confidence data points. Spatial density-based clustering schemes can use the DBSCAN or OPTICS clustering algorithms. By setting input parameters such as the minimum number of points and the search radius, multiple clusters (corresponding to multiple irrigation plots) can be output. This algorithm can automatically discover high-density core points and exclude noise points, effectively solving the discrete distribution problem of mobile terminal data. The minimum circumscribed polygon refers to a convex polygon with the minimum area that covers all irrigation coordinate points, constructed using the rotating calcaneal algorithm. This structure eliminates redundant boundary areas and retains the spatial topological characteristics of the actual irrigation range.

[0116] The spatiotemporal correlation mentioned in 402 refers to aligning remote sensing data with the benchmark irrigation boundary area through coordinate transformation, and matching timestamps to ensure that the data are in the same observation period, avoiding spatial deviation caused by time dislocation.

[0117] The historical data mentioned in 403 refers to the stable boundary range in the historical remote sensing data, which is used to constrain the revised remote sensing irrigation boundary.

[0118] Specifically, when a spatial offset between the remote sensing boundary and the baseline boundary is detected, the remote sensing boundary is morphologically corrected by invoking a library of morphological features from historical irrigation boundaries and combining them with the geometric parameters of the current baseline boundary. For example, when the remote sensing boundary exceeds the baseline boundary in a specific direction, the boundary extension threshold for that area in the historical data is used for truncation. This ensures that the corrected boundary is consistent with both real-time observations and long-term spatial distribution patterns.

[0119] This embodiment integrates real-time data from mobile terminals to construct a baseline boundary, creating dynamic spatial constraints and enabling self-correction in the remote sensing boundary identification process. It also incorporates a historical data correction mechanism, incorporating long-term observed spatial patterns into the boundary adjustment process, thus overcoming the potential for random errors in data from a single point in time.

[0120] Through the above technical solution, this embodiment effectively addresses the problem of irrigation boundary identification errors caused by cloud cover or insufficient resolution. The construction of the baseline irrigation boundary provides a reliable spatial reference for remote sensing data. The morphological correction mechanism adaptively adjusts the boundary morphology, ensuring that the final output irrigation boundary is both real-time and spatially continuous. This multi-source data collaborative correction method significantly improves the spatial baseline accuracy required for irrigation status inversion, laying an accurate data foundation for subsequent moisture feature extraction and irrigation coefficient calculation.

[0121] Regarding identifying the remote sensing irrigation boundary of the first remote sensing data and the second remote sensing data in 402, the first remote sensing data and the second remote sensing data are preliminarily divided with reference to the reference irrigation boundary area, and then boundary detection is performed on the preliminarily divided irrigation boundary, and boundary tracking is performed based on the difference in features inside and outside the boundary to generate the remote sensing irrigation boundary.

[0122] Boundary detection involves comparing spectral reflectance, vegetation index, or soil moisture differences within and outside the initially demarcated area to identify areas with significant characteristic changes as potential boundary points. Boundary tracking involves dynamically adjusting boundary orientation based on the spatial continuity of potential boundary points using edge connection or region growing algorithms to generate continuous and accurate irrigation boundary lines.

[0123] Specifically, the irrigation data uploaded by the mobile terminal is preprocessed and clustered to generate a benchmark irrigation boundary area. This area serves as prior knowledge to guide the preliminary division of the remote sensing data. The area after the preliminary division may have blurred or broken boundaries. By calculating the feature differences inside and outside the boundary, for example, using the mutation characteristics of the normalized vegetation index in irrigated and non-irrigated areas, the location of the true boundary is detected. Subsequently, an edge tracking algorithm, such as Canny edge detection combined with Hough transform, is used to connect and optimize the detected boundary points, eliminate noise interference, and form a complete irrigation boundary line. By combining the spatial constraints of the benchmark data with the dynamic feature analysis of the remote sensing data, it is possible to accurately restore the irrigation boundary at the field scale when the data is incomplete or noisy.

[0124] This implementation uses the baseline irrigation boundary as a spatial constraint to ensure the rationality of the initial delineation. Dynamic analysis of feature differences within and outside the boundary allows for adaptive adjustment of the boundary position to compensate for incomplete remote sensing data. This significantly improves the accuracy of irrigation status assessments, laying a reliable foundation for the subsequent accurate calculation of irrigation water utilization coefficients.

[0125] The above describes the specific steps of remote sensing irrigation boundary correction. The following describes in detail the spatiotemporal fusion of the first remote sensing data and the second remote sensing data in step 102 to generate spatiotemporal continuous moisture characteristic data.

[0126] The first remote sensing data is represented as , the second remote sensing data is expressed as .

[0127] Taking the improved STARFM model as an example, the fusion result can be expressed in the following form:

[0128] .

[0129] in, Indicates the location , fused remote sensing data at time t, Indicates the location , the first remote sensing data at time t, represents the weighting factor, Indicates the location ,time The second remote sensing data on Indicates the location ,time The first remote sensing data on.

[0130] The final generated moisture characteristic data It can be expressed as: .

[0131] in, represents the moisture estimation function, Indicates the total water diversion data of the target area.

[0132] The specific expression formula of the irrigation inversion model mentioned in 103 is as follows:

[0133]

[0134] in, Represents the pre-trained irrigation inversion model, which can be trained by forming a training set with historical data. represents the parameter set of the irrigation inverse model, Represents the input mapping function.

[0135] The following describes in detail the consistency matching of the irrigation data and the irrigation status distribution data in step 104 to determine the first difference.

[0136] The differences of all pixels in the target area are weighted or averaged to obtain the overall index.

[0137] Average difference: .in, Represents the pixel set of the target area, that is, all ; Indicates the first degree of difference.

[0138] If some areas are more important, weighted differences are used:

[0139] .in, Represents the pixel importance weight.

[0140] The final output difference matrix can be used to generate difference heat maps, error distribution maps, etc. The value at each grid position represents the degree of inconsistency between the actual irrigation and the predicted irrigation in the area.

[0141] The above describes the specific process of determining the first difference, and the following describes in detail the calculation process of the confidence factor in 106. Figure 6 As shown, obtaining the historical accuracy of data uploaded by mobile terminals and calculating the confidence factor in a decentralized manner specifically includes the following steps:

[0142] 501. Count the proportion of the data reported by the mobile terminal M times in history that matches the authoritative data to obtain the historical accuracy rate of the data uploaded by the mobile terminal; where M is a positive integer;

[0143] 502. Obtain the reporting time of the irrigation data and calculate the difference between the reporting time and the current time to obtain the difference Δt;

[0144] 503. Calculate the time-dependent attenuation coefficient λt by using the preset attenuation rate λ and the difference Δt: λt=exp(-λ×Δt);

[0145] 504. Multiply the time decay coefficient λt and the historical accuracy rate to obtain a confidence factor.

[0146] Specifically, the inherent data quality is first quantified by measuring the proportion of historically reported data from mobile terminals that matches authoritative data, thus preventing accidental errors from interfering with the overall assessment. The difference Δt between the current data reporting time and the processing time is then calculated. Combined with a preset decay rate λ, an exponential decay function is used to generate a time decay coefficient, simulating the decreasing reliability of data over time. Finally, the historical accuracy rate is multiplied by the time decay coefficient to generate a dynamically adjusted confidence factor. This factor preserves the long-term credibility of the data source while reflecting the current data timeliness, providing a multi-dimensional basis for credibility assessment for model fine-tuning.

[0147] This embodiment addresses the reliability issues associated with model fine-tuning due to differences in the timeliness and historical accuracy of mobile terminal data. By dynamically adjusting the confidence factor, the use of outdated or low-quality data for model updates is avoided, improving the accuracy of data screening during fine-tuning. Furthermore, the introduction of an exponential decay function makes timeliness assessment more consistent with the nonlinear nature of data value decay in real-world scenarios, enhancing the rationality of credibility calculations.

[0148] The calculation of the confidence factor in 106 is described in detail above. The following describes in detail 107 how to use the supervisory signal and the confidence factor as fine-tuning parameters to fine-tune the irrigation inversion model through supervised machine learning.

[0149] The weighted supervised loss function for fine-tuning the irrigation inversion model is as follows:

[0150] .

[0151] in, represents the confidence factor, represents the fine-tuning loss function, Represents the model parameters to be adjusted, represents the basic loss function, Indicates the data subset whose difference exceeds the threshold, represents the supervisory signal, Represents the irrigation inverse model.

[0152] Perform gradient descent optimization to update model parameters, specifically: ;

[0153] in, Represents the learning rate, which controls the step size of each parameter update. represents the gradient of the loss function with respect to the model parameters, Represents the loss function of the current model.

[0154] The above introduces the specific steps of fine-tuning the irrigation inversion model. The following is a detailed explanation of the calculation of the irrigation water utilization coefficient.

[0155] The output data of the irrigation inversion model represents the effective irrigation water volume, i.e. ; Indicates the final selected irrigation status distribution data, Represents the area of ​​the pixel (x, y) in the region, Indicates the irrigation water standard per unit area per unit time (unit: m³ / m²).

[0156] The final irrigation water use coefficient is: ;in, Indicates the total water diversion volume (unit: m³).

[0157] The above introduces the specific steps for calculating the irrigation water utilization coefficient. The following further explains the specific steps for fine-tuning the irrigation inversion model.

[0158] The irrigation state distribution data output by the fine-tuned irrigation inversion model is matched with the irrigation data to determine a second difference degree;

[0159] Determine whether the second difference is less than the first difference. Otherwise, repeat the above operation until the second difference is less than the first difference. Use the irrigation state distribution data output by the fine-tuned irrigation inversion model as the output data or the number of iterations reaches the preset value, and reset the fine-tuning parameters.

[0160] The purpose of this embodiment is to ensure that the accuracy of the irrigation status distribution data output by the fine-tuned irrigation inversion model is higher than that before fine-tuning, thereby ensuring the accuracy and precision of the data.

[0161] In order to facilitate understanding of the above embodiments, Figure 7 A specific application scenario of the above embodiment is used as an example for description:

[0162] Taking a typical arid agricultural area as an example, multispectral imaging satellites are used to acquire regional images daily, and ground-based imaging spectrometers (such as those mounted on drones) are used to obtain more frequent moisture changes. Farmers automatically upload irrigation time, method, and water volume information through apps or IoT irrigation terminal devices.

[0163] The target area has approximately 50,000 hectares of irrigated land, primarily cotton, with some corn. Water is diverted from the main river and pumped water from wells and groundwater. Drip irrigation is the primary method, with some flooding. Water resource pressure is high, with annual evaporation far exceeding precipitation, necessitating meticulous management.

[0164] The total water diversion data is obtained from the monitoring system of the local water conservancy bureau. Taking a certain day X in history as an example, the specific data is shown in the table below:

[0165] Table 2: Data collection table for a typical arid agricultural area

[0166] Data Type source Data frequency Collect data Total water diversion data Local Water Conservancy Bureau Monitoring System month / ten-year level The total monthly water diversion on day X is 42 million cubic meters First remote sensing data (multispectral) Gaofen-2, Sentinel-2 Day level (low cloud imagery is prioritized) X-day image, resolution 10m, cloud level: moderate cloud Second remote sensing data (imaging spectrum) UAV-mounted spectrometer Hourly level From 13:00 to 16:00 on X day, scan and shoot key plots Mobile terminal irrigation data Farmer App / IoT Devices Real-time upload [User ID 001, Location: xx°N, xx°E, Irrigation Type: Drip Irrigation, Time: 12:20 PM, Estimated Water Volume: 7.5 mm] Historical Image Library Local remote sensing database Continuous accumulation Low cloud image of the same area in May of the previous year, NDVI=0.68

[0167] Since the image on that day was at the medium cloud level, local interpolation was performed on the cloud-obstructed area and spatially enhanced and fused with the spectral data collected by the drone on that day to generate a continuous surface moisture map (resolution 10m, covering the entire area). The fused data was input into the irrigation inversion model to output an irrigation status map.

[0168] Comparing the irrigation records uploaded by users, it was found that the upload time of user 001 was 12:20, which was consistent with the model prediction and had a high degree of consistency; there was irrigation record uploaded by users for plot 045, but the model showed no signs of irrigation, with a difference of 0.43 (higher than the threshold of 0.3), and the fine-tuning process began.

[0169] According to the user's historical upload accuracy of 85%, Δt is 2 hours, λ=0.15, we calculate λt=exp(-0.15×2≈0.74;

[0170] Confidence factor = 0.85 × 0.74 ≈ 0.629;

[0171] After fine-tuning the model using this factor, plot A045 was identified as “possibly irrigated”.

[0172] The total irrigated area in the entire region was finally identified as 42,000 hectares, and the actual irrigation water consumption was estimated to be approximately 35 million cubic meters based on irrigation intensity data.

[0173] Calculating irrigation water use coefficient =Actual water usage / total water diversion=3500 / 4200=0.833.

[0174] In summary, this embodiment effectively integrates satellite remote sensing, near-Earth remote sensing, and ground-based sensing data to achieve high-resolution and high-accuracy identification of the irrigation status of the target area; introduces supervised learning and a dynamic confidence mechanism to enhance the adaptability and generalization capabilities of the inversion model; and can quantify and measure irrigation water utilization efficiency at the regional scale, facilitating water resource allocation, refined agricultural management, and the implementation of the water rights system. This lays the foundation for building a smart agricultural remote sensing monitoring platform.

[0175] Exemplary Systems

[0176] like Figure 8 As shown, this embodiment introduces an irrigation water utilization coefficient calculation system based on multi-source remote sensing information, including a data acquisition module, a data fusion module, a data processing module, a data matching module, a data judgment module, a data fine-tuning module, and a result output module.

[0177] The data acquisition module is used to obtain the total water diversion data for the target area, as well as the first remote sensing data collected by the multispectral imaging satellite, the second remote sensing data collected by the imaging spectrometer, and the irrigation data uploaded by the mobile terminal;

[0178] The data fusion module is used to perform spatiotemporal fusion of the first remote sensing data and the second remote sensing data to generate spatiotemporal continuous moisture characteristic data;

[0179] The data processing module is used to input the moisture characteristic data into a pre-trained irrigation inversion model to obtain the irrigation status distribution data of the target area;

[0180] The data matching module is used to perform consistency matching between the irrigation data and the irrigation status distribution data to determine a first difference degree;

[0181] The data judgment module is used to determine whether the first difference is greater than a preset threshold value. If so, the irrigation status distribution data is determined to be accurate; otherwise, the irrigation status distribution data is determined to be inaccurate, and the irrigation time when the mobile terminal uploaded the irrigation data is extracted and used as a supervision signal. The historical accuracy rate of the data uploaded by the mobile terminal is also obtained, and a confidence factor is calculated in a decentralized manner.

[0182] The data fine-tuning module is used to fine-tune the irrigation inversion model through supervised machine learning using the supervision signal and the confidence factor as fine-tuning parameters;

[0183] The result output module is used to combine the irrigation state distribution data output by the fine-tuned irrigation inversion model with the total water diversion data to calculate the irrigation water utilization coefficient of the target area.

[0184] In some specific implementations, irrigation data uploaded by mobile terminals may include geographic coordinates, irrigation duration, and water metering information, with data security ensured through an encrypted transmission protocol. The spatiotemporal fusion process can utilize multi-scale feature extraction techniques, such as texture analysis for high-resolution data and time series interpolation for low-resolution data. Transfer learning strategies can be incorporated during model fine-tuning, such as freezing some network layers and fine-tuning only the parameters of the fully connected layer. Confidence factor calculations can incorporate a time decay function, such as an exponential decay model, to dynamically adjust the weight of historical data.

[0185] This implementation effectively improves the accuracy of irrigation water utilization coefficient calculations, overcoming assessment bias caused by spatiotemporal mismatches in multi-source data. Dynamically triggering model fine-tuning and output optimization enhances the system's robustness to complex environmental factors such as cloud interference and equipment errors. Utilizing supervisory signals and confidence factors to optimize parameters improves the relevance and reliability of model adjustments and avoids overfitting. Ultimately, this achieves full process optimization from data collection to output, providing reliable technical support for precision irrigation management.

[0186] Exemplary computer terminal

[0187] like Figure 9 As shown, this embodiment introduces a computer terminal, which includes a memory, a processor, and a computer program stored in the memory and runnable on the processor. When the processor executes the program, the steps of the aforementioned method for calculating the irrigation water utilization coefficient based on multi-source remote sensing information are implemented.

[0188] There can be one or more processors. Figure 8 Taking one as an example, in this embodiment, the processor and the memory may be connected via a bus or other means, wherein: Figure 8 In the figure, the bus connection is taken as an example, and the corresponding input devices and output devices are also shown.

[0189] The dynamic nutrition optimization method based on multi-model fusion can be applied in the form of software, such as a stand-alone program installed on a computer terminal, which can be a computer, a smartphone, etc. It can also be designed as an embedded program and installed on a computer terminal, such as a single-chip microcomputer.

[0190] Specifically, the computer terminal uses memory to store multi-source remote sensing data and irrigation logs reported by mobile terminals. The processor uses a pre-trained irrigation inversion model to analyze the spatiotemporal fusion of water characteristic data to generate a field-scale irrigation status distribution. When the difference between the model output and the mobile terminal data exceeds a threshold, the system automatically extracts the irrigation time as a supervision signal, dynamically adjusts the data weight based on historical accuracy, and fine-tunes the model parameters using a gradient descent algorithm. The fine-tuned model output is compared with the original result for a second degree of difference, and the version with the smallest difference is selected as the final irrigation status data. The total water diversion data is spatially superimposed with the optimized irrigation status distribution, and the regional irrigation water utilization coefficient is calculated using the water balance equation.

[0191] This implementation solves the problem of irrigation status identification bias caused by spatiotemporal inconsistencies in remote sensing data. It improves the spatial consistency of model output through multi-source data fusion and dynamic optimization. The use of dual difference judgment and decentralized confidence factor calculation effectively mitigates the negative impact of abnormal mobile terminal data on model optimization, ensuring that irrigation water utilization coefficient calculation results have both wide-area coverage and local accuracy. By leveraging the hardware computing power of computer terminals and the synergy of software algorithms, efficient processing of multi-source heterogeneous data and multi-scale analysis of irrigation status are achieved.

[0192] The technical scope of the present invention is not limited to the contents of the above description. Those skilled in the art can make various deformations and modifications to the above embodiments without departing from the technical idea of ​​the present invention, and these deformations and modifications should all fall within the protection scope of the present invention.

Claims

1. A method for calculating irrigation water utilization coefficient based on multi-source remote sensing information, characterized in that: It includes the following steps: Obtaining total water diversion data for a target area, as well as first remote sensing data collected by a multispectral imaging satellite, second remote sensing data collected by an imaging spectrometer, and irrigation data uploaded by a mobile terminal; fusing the first remote sensing data and the second remote sensing data in time and space to generate spatiotemporally continuous moisture characteristic data; Inputting the moisture characteristic data into a pre-trained irrigation inversion model to obtain irrigation status distribution data of the target area; Performing consistency matching on the irrigation data and the irrigation status distribution data to determine a first difference; determining whether the first difference is greater than a preset threshold, and if so, determining that the irrigation status distribution data is accurate; Otherwise, the irrigation status distribution data is determined to be inaccurate, the irrigation time when the mobile terminal uploaded the irrigation data is extracted and used as a supervision signal, and the historical accuracy of the data uploaded by the mobile terminal is obtained, and the confidence factor is calculated in a decentralized manner; Fine-tuning the irrigation inversion model through supervised machine learning using the supervision signal and the confidence factor as fine-tuning parameters; The irrigation state distribution data output by the fine-tuned irrigation inversion model is combined with the total water diversion data to calculate the irrigation water utilization coefficient of the target area.

2. The method for calculating irrigation water utilization coefficient based on multi-source remote sensing information according to claim 1, characterized in that: The specific steps of obtaining the first remote sensing data collected by the multispectral imaging satellite are as follows: Obtain images of the target area and their corresponding cloud masks collected by multispectral imaging satellites; The overall cloud coverage of the image is calculated based on the cloud mask corresponding to the image. The cloud cover level of the image is then determined based on the preset multi-level cloud cover threshold. The following decisions are made based on the cloud cover level: (a1) if the cloud cover level of the image is low cloud level, directly using the image as the first remote sensing data; (a2) if the cloud cover level of the image is medium cloud level, performing local interpolation processing on the cloud-obscured area in the image, and using the processed image as the first remote sensing data; (a3) If the cloud cover level of the image is high cloud level, extracting effective features around the image, performing spatial enhancement processing on the image in combination with the second remote sensing data collected by the imaging spectrometer, and using the processed image as the first remote sensing data; (a4) If the cloud cover level of the image is very high, assess its spatiotemporal importance score based on the time period and region in which the image is located; Determine whether the spatiotemporal importance score is greater than a preset spatiotemporal threshold; if so, retrieve an image of the lowest cloud cover in the same area at a similar time from a historical image library, perform spatial enhancement processing on the image combined with the second remote sensing data collected by the imaging spectrometer, and use the processed image as the first remote sensing data; Otherwise the image is discarded.

3. The method for calculating irrigation water utilization coefficient based on multi-source remote sensing information according to claim 2, characterized in that: The specific steps of evaluating the spatiotemporal importance score of the image according to the time period and region in which the image is located are as follows: Obtaining the NDVI corresponding to the time point of the image and the period and adjacent periods of the time point; Determine whether the time point of the image is at a preset key time node, and if so, calculate the NDVI change rate based on the NDVI corresponding to the period at the time point and the adjacent period, and determine its time importance based on the NDVI change rate; Acquire the area where the image is located; Mapping the region where the image is located into a preset regional priority map, and determining the spatial importance of the overlapping region according to the corresponding level of the location of the overlapping region; The temporal importance and the spatial importance are dynamically weighted to obtain a temporal and spatial importance score.

4. The method for calculating irrigation water utilization coefficient based on multi-source remote sensing information according to claim 2, characterized in that: The specific steps for retrieving the minimum cloud cover image of the same area at a nearby time from the historical image library are as follows: Retrieve historical images from the historical image library based on the preset proximity time length and sort them in ascending order of cloud cover; Determine whether the cloud cover of the TopN historical images is the same. Otherwise, the Top1 historical image is output as the image with the lowest cloud cover. If yes, extract the NDVI of the TopN historical images, and perform correlation calculations with the NDVI of the image one by one, and output the historical image with the largest correlation value as the lowest cloud cover image.

5. The method for calculating irrigation water utilization coefficient based on multi-source remote sensing information according to claim 1, characterized in that: Before performing spatiotemporal fusion on the first remote sensing data and the second remote sensing data, the irrigation boundaries of the first remote sensing data and the second remote sensing data are corrected according to the irrigation data uploaded by the mobile terminal, specifically comprising the following steps: The irrigation data uploaded by the mobile terminal is preprocessed, and the processed irrigation data is clustered based on spatial density to identify irrigation coordinate points. The irrigation coordinate points are constructed into a minimum circumscribed polygon to obtain the reference irrigation boundary area. Performing spatiotemporal association of the first remote sensing data and the second remote sensing data with the preliminary irrigation boundary area to identify remote sensing irrigation boundaries of the first remote sensing data and the second remote sensing data; It is determined whether the remote sensing irrigation boundary is consistent with the reference irrigation boundary area; otherwise, the remote sensing irrigation boundary is corrected with reference to the reference irrigation boundary area and historical data.

6. The method for calculating irrigation water utilization coefficient based on multi-source remote sensing information according to claim 1, characterized in that: When the remote sensing irrigation boundaries of the first remote sensing data and the second remote sensing data are identified, the first remote sensing data and the second remote sensing data can be preliminarily divided with reference to the benchmark irrigation boundary area, and then boundary detection is performed on the preliminarily divided irrigation boundaries. Boundary tracking is performed based on the difference in features inside and outside the boundaries to generate a remote sensing irrigation boundary.

7. The method for calculating irrigation water utilization coefficient based on multi-source remote sensing information according to claim 1, characterized in that: The first remote sensing data is acquired on a daily or weekly basis, and the second remote sensing data is acquired on an hourly basis; the irrigation data uploaded by the mobile terminal includes location information of the target area, irrigation time information, irrigation type, and irrigation intensity, and is stored in a log database.

8. The method for calculating irrigation water utilization coefficient based on multi-source remote sensing information according to claim 1, characterized in that: The specific steps for obtaining the historical accuracy of data uploaded by mobile terminals and calculating the confidence factor in a decentralized manner are as follows: Counting the proportion of the data reported by the mobile terminal M times in history that matches the authoritative data to obtain the historical accuracy of the data uploaded by the mobile terminal; where M is a positive integer; Obtain the reporting time of the irrigation data, and calculate the difference Δt with the current time; The aging attenuation coefficient λt is calculated by the preset attenuation rate λ and the difference Δt: λt=exp(-λ×Δt); The confidence factor is calculated by multiplying the time decay coefficient λt and the historical accuracy rate.

9. The irrigation water utilization coefficient calculation system based on multi-source remote sensing information is characterized by: It includes: A data acquisition module is used to acquire total water diversion data for a target area, as well as first remote sensing data collected by a multispectral imaging satellite, second remote sensing data collected by an imaging spectrometer, and irrigation data uploaded by a mobile terminal; a data fusion module, configured to perform spatiotemporal fusion of the first remote sensing data and the second remote sensing data to generate spatiotemporally continuous moisture characteristic data; a data processing module, configured to input the moisture characteristic data into a pre-trained irrigation inversion model to obtain irrigation status distribution data of the target area; a data matching module, configured to perform consistency matching between the irrigation data and the irrigation status distribution data to determine a first difference; a data judgment module, configured to judge whether the first difference is greater than a preset threshold, and if so, to judge that the irrigation status distribution data is accurate; otherwise, to judge that the irrigation status distribution data is inaccurate, extract the irrigation time when the mobile terminal uploaded the irrigation data and use it as a supervision signal, obtain the historical accuracy of the data uploaded by the mobile terminal, and calculate the confidence factor in a decentralized manner; a data fine-tuning module, configured to fine-tune the irrigation inversion model through supervised machine learning using the supervisory signal and the confidence factor as fine-tuning parameters; The result output module is used to combine the irrigation state distribution data output by the fine-tuned irrigation inversion model with the total water diversion data to calculate the irrigation water utilization coefficient of the target area.

10. A computer terminal comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the method for calculating the irrigation water utilization coefficient based on multi-source remote sensing information as described in any one of claims 1 to 8 are implemented.

Images