A new reclamation intelligent monitoring and evaluation method and system

Abstract

This invention discloses a method and system for intelligent monitoring and evaluation of newly reclaimed farmland, relating to the field of smart agriculture. The method includes: planning UAV flight paths and ground sampling grids based on the boundaries and topographic data of the newly reclaimed farmland; acquiring multispectral and thermal infrared images via UAV remote sensing, and simultaneously collecting spectral data and soil samples on the ground; conducting laboratory analysis on the soil samples to obtain soil physicochemical indicators, and constructing a training sample set by correlating these with remote sensing spectral features; based on the training sample set, fusing remote sensing features, and using a machine learning model to invert and obtain a spatial distribution map of soil physicochemical indicators; comprehensively classifying soil physicochemical indicators, crop growth sequence, and planting history information to perform quality grade classification and obstacle type diagnosis; based on the classification and diagnosis results, combining an agricultural knowledge rule base to form agronomic management suggestions and generate a variable operation prescription map to guide precision operations. Through air-ground collaborative monitoring and soil quality assessment, the accuracy of farmland quality monitoring is improved.

Description

Technical Field

[0001] This invention relates to the field of smart agriculture, specifically to a method and system for intelligent monitoring and evaluation of newly reclaimed farmland. Background Technology

[0002] With the development of agricultural modernization, precision agriculture technology has gradually become an important means to improve agricultural production efficiency and sustainable development. Newly reclaimed farmland, as a crucial component of agricultural development, requires particularly stringent soil quality assessment and monitoring. Traditional farmland monitoring methods largely rely on manual sampling and laboratory analysis, which are inefficient, costly, and unable to reflect real-time dynamic changes in the soil. With the rapid development of remote sensing, drone technology, and artificial intelligence, combining high-resolution remote sensing imagery with ground-based measured data for precise monitoring has become a trend. However, existing farmland monitoring methods often lack a comprehensive air-ground collaborative monitoring framework, failing to fully cover the soil's physicochemical properties and crop growth. Furthermore, traditional monitoring technologies and analysis methods are not precise enough in identifying the spatial distribution characteristics of large areas of farmland, failing to provide dynamic farmland quality distribution maps and obstacle type diagnostic maps, making targeted agricultural management and operations difficult. To address these issues, a new technological solution is urgently needed that can provide an efficient and accurate intelligent monitoring and evaluation method for newly reclaimed farmland by comprehensively utilizing technologies such as drone remote sensing, ground sampling, machine learning, and agricultural knowledge bases. Summary of the Invention

[0003] Based on the shortcomings of the prior art described above, the purpose of this invention is to provide a method and system for intelligent monitoring and evaluation of newly reclaimed farmland to solve the aforementioned technical problems.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a method for intelligent monitoring and evaluation of newly reclaimed arable land, comprising:

[0005] S1: Based on the boundary vector and digital elevation model of newly reclaimed farmland, generate optimized flight paths and ground sampling grids for UAVs to establish an air-ground collaborative monitoring spatial framework;

[0006] S2: Acquire multispectral and thermal infrared images through UAV remote sensing, collect surface spectral data and soil samples on the ground based on sampling grids, and perform registration and preprocessing on UAV remote sensing images and ground-collected data.

[0007] S3: Conduct laboratory analysis on soil samples to obtain soil physicochemical indicators, and correlate the soil physicochemical indicators with the remote sensing spectral characteristics of the corresponding sampling points to construct a training sample set;

[0008] S4: Based on the training sample set, spectral features, thermal infrared features and historical time-series remote sensing data are integrated, and the spatial distribution map of soil physicochemical indicators of cultivated land is obtained by inverting through machine learning model.

[0009] S5: Integrate soil physicochemical indicators, crop growth dynamics and time-series characteristics and historical planting context information to classify arable land quality grades and identify soil obstacle types, and generate arable land quality grade distribution maps and obstacle type diagnosis maps.

[0010] S6: Based on the quality grade distribution map and obstacle type diagnosis map, combined with the agricultural knowledge rule base, agronomic management suggestions are generated, and then variable operation prescription maps are generated to drive precise variable operations.

[0011] The present invention is further configured such that S1 includes:

[0012] Read the boundary vector data and digital elevation model data of the newly reclaimed farmland area, and perform spatial reference consistency verification on the boundary vector data and digital elevation model data;

[0013] Based on the preset ground resolution and sensor field of view parameters, the flight altitude of the UAV is calculated, and a flight path planning file is generated according to the preset forward overlap and lateral overlap. The flight path planning file includes the spatial coordinates, flight parameters and sensor imaging angles of each waypoint.

[0014] Within the area defined by the boundary vector data, a sampling point array with a regular grid distribution is generated based on a preset sampling interval. Each grid point is assigned a unique identifier and given three-dimensional coordinates based on the digital elevation model. The output is a sampling grid point file containing point identifiers and spatial coordinates.

[0015] The present invention is further configured such that S2 includes:

[0016] Using a multispectral imager and a thermal infrared imager configured on a UAV, aerial remote sensing data is collected along a preset flight path, and positioning and attitude data are recorded simultaneously. The aerial remote sensing data includes multispectral images and thermal infrared images.

[0017] Simultaneously employing a handheld ground object spectrometer and a real-time dynamic differential global navigation satellite system receiver, surface spectral measurements and soil sample collection were conducted at pre-set sampling grid points on the ground. The surface spectral measurements included vertical observations based on standard whiteboard correction and averaging of multiple measurements.

[0018] Aerial triangulation processing was performed on the collected aerial remote sensing data and positioning and attitude data. Among them, radiometric calibration and orthorectification were performed on the multispectral images, and temperature calibration and orthorectification were performed on the thermal infrared images.

[0019] The corrected multispectral image is fused with the thermal infrared image to generate an orthophoto containing multispectral band information and surface temperature information.

[0020] The spectral data and positioning coordinates collected on the ground are spatially registered with the orthophoto to establish the spatial correspondence between ground sampling points and remote sensing image pixels.

[0021] The present invention is further configured such that S3 includes:

[0022] The collected soil samples were analyzed in the laboratory to obtain soil physicochemical indicators, including soil organic matter content, pH value, total nitrogen content, available phosphorus content, and available potassium content.

[0023] Based on the registered orthophoto, multi-band reflectance data within a preset-size neighborhood window of the corresponding image location is extracted for each ground sampling point, and the arithmetic mean and standard deviation of the reflectance of each band are calculated.

[0024] The measured soil physicochemical indicators were spatially correlated and matched with the corresponding remote sensing spectral features extracted from orthophotos.

[0025] A training sample set is constructed based on the correlation matching results. The training sample set contains soil physicochemical index data and its corresponding spectral feature data for each sampling point.

[0026] The present invention is further configured such that S4 includes:

[0027] Acquire time-series remote sensing data of newly reclaimed farmland for a preset historical period, filter and smooth the time-series remote sensing data, and generate a vegetation index growth time series.

[0028] The spectral features and vegetation index growth time series in the training sample set are fused to generate a fused feature vector.

[0029] Based on the soil physicochemical indicators and their corresponding fused feature vectors in the training sample set, a soil physicochemical indicator inversion model is trained, and the model is used to invert the whole-domain orthophoto to generate a spatial distribution map of soil physicochemical indicators of cultivated land.

[0030] The present invention is further configured such that S5 includes:

[0031] The newly reclaimed farmland is divided into regular grid evaluation units of preset size, and each unit is assigned a unique identifier;

[0032] The arithmetic mean of soil physicochemical indicators for each evaluation unit is extracted from the spatial distribution map of soil physicochemical indicators, and a static indicator feature vector is constructed.

[0033] The current average biomass and historical biomass change trends of each evaluation unit are extracted from the multi-period crop biomass distribution map to construct a dynamic time series index feature vector.

[0034] Crop rotation information and yield levels of each evaluation unit are extracted from the historical planting database, and historical context feature vectors are constructed after feature encoding.

[0035] The static indicator feature vector, dynamic time-series indicator feature vector, and historical context feature vector are concatenated and merged to generate a comprehensive feature vector for each evaluation unit.

[0036] Based on historical farmland quality assessment results, each assessment unit is assigned farmland quality grade labels and obstacle type labels, and a deep learning assessment model is constructed with comprehensive feature vectors as input and quality grade classification and obstacle type identification as parallel output tasks.

[0037] The deep learning evaluation model is trained and validated using a labeled evaluation unit dataset. The trained deep learning evaluation model is then applied to predict the quality level and obstacle type of the comprehensive feature vector of all evaluation units in the entire domain, generating a quality level distribution map and obstacle type diagnostic map of arable land.

[0038] The present invention is further configured such that S6 includes:

[0039] A structured agricultural knowledge rule base is constructed, which includes judgment rules based on the combination of arable land quality grade, obstacle type and soil physicochemical index threshold, as well as agronomic management measures corresponding to each combination of conditions.

[0040] Based on the classification results of cultivated land quality grades, the diagnosis results of obstacle types, and the quantitative values ​​of soil physicochemical indicators of each evaluation unit, differentiated agronomic management suggestions for each evaluation unit are generated by matching the corresponding rules in the agricultural knowledge rule base.

[0041] Using agronomic management recommendations and spatial distribution maps of soil physicochemical indicators as input, variable operation prescription maps are generated based on preset decision logic.

[0042] The present invention is further configured to transmit the variable operation prescription map to the intelligent agricultural equipment control system through a standard data interface, thereby driving the variable execution mechanism to achieve precise operation.

[0043] The present invention is further configured such that the method also includes visually displaying the quality grade distribution map of cultivated land, the obstacle type diagnosis map, and the variable operation prescription map.

[0044] This invention also provides an intelligent monitoring and evaluation system for newly reclaimed farmland, the system comprising:

[0045] Collaborative Framework Module: Based on the boundary vector and digital elevation model of newly reclaimed farmland, it generates optimized flight paths for UAVs and ground sampling grids to establish an air-ground collaborative monitoring spatial framework;

[0046] Collaborative acquisition module: acquires multispectral and thermal infrared images through UAV remote sensing, collects surface spectral data and soil samples on the ground based on sampling grid, and performs registration and preprocessing on UAV remote sensing images and ground-collected data;

[0047] Truth value construction module: Conduct laboratory analysis on soil samples to obtain soil physicochemical indicators, and associate the soil physicochemical indicators with the remote sensing spectral characteristics of the corresponding sampling points to construct a training sample set;

[0048] Collaborative Inversion Module: Based on the training sample set, it integrates spectral features, thermal infrared features and historical time-series remote sensing data, and uses a machine learning model to invert the spatial distribution map of soil physicochemical indicators of cultivated land.

[0049] Evaluation module: Integrates soil physicochemical indicators, crop growth dynamic time series characteristics and historical planting context information to classify arable land quality grades and identify soil obstacle types, and generate arable land quality grade distribution maps and obstacle type diagnosis maps;

[0050] Decision Management Module: Based on the quality level distribution map and obstacle type diagnosis map, combined with the agricultural knowledge rule base, agronomic management suggestions are generated, and then variable operation prescription maps are generated to drive precise variable operations.

[0051] This invention provides a method and system for intelligent monitoring and evaluation of newly reclaimed farmland. The method comprises: S1: Generating an optimized UAV flight path and ground sampling grid based on the boundary vector and digital elevation model of the newly reclaimed farmland to establish an air-ground collaborative monitoring spatial framework; S2: Acquiring multispectral and thermal infrared images through UAV remote sensing, collecting surface spectral data and soil samples on the ground based on the sampling grid, and registering and preprocessing the UAV remote sensing images and ground-collected data; S3: Performing laboratory analysis on the soil samples to obtain soil physicochemical indicators, and correlating these indicators with the remote sensing spectral characteristics of the corresponding sampling points to construct... Training sample set; S4: Based on the training sample set, spectral features, thermal infrared features, and historical time-series remote sensing data are fused to obtain the spatial distribution map of soil physicochemical indicators of cultivated land through machine learning model inversion; S5: Soil physicochemical indicators, crop growth dynamic time-series characteristics, and historical planting context information are fused to classify cultivated land quality grades and identify soil obstacle types, generating cultivated land quality grade distribution maps and obstacle type diagnostic maps; S6: Based on the quality grade distribution map and obstacle type diagnostic map, combined with the agricultural knowledge rule base, agronomic management suggestions are generated, and then a variable operation prescription map is generated to drive precise variable operations. The beneficial effects include:

[0052] 1. This invention constructs an air-ground collaborative monitoring spatial framework by combining UAV remote sensing and ground sampling grids, which can efficiently collect remote sensing data and soil samples of newly reclaimed farmland, ensuring the comprehensiveness and high accuracy of data collection, and providing accurate and rich basic data for farmland quality assessment;

[0053] 2. This invention analyzes soil samples in the laboratory and combines them with the spectral characteristics of remote sensing images. It then uses machine learning inversion technology to generate a spatial distribution map of soil physicochemical indicators for arable land. This method can accurately assess soil quality, provide a scientific basis for agricultural decision-making, and improve the accuracy of arable land quality monitoring.

[0054] 3. Based on the distribution map of arable land quality and the diagnostic map of obstacle type, this invention can generate differentiated agronomic management suggestions and achieve precise variable operations through intelligent agricultural equipment. This precise operation and intelligent management not only improves the productivity of arable land but also reduces resource waste and helps to achieve sustainable development of agricultural production.

[0055] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0056] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:

[0057] Figure 1 A flowchart illustrating an intelligent monitoring and evaluation method for newly reclaimed farmland, as shown in an exemplary embodiment of the present invention;

[0058] Figure 2 This is a schematic diagram of the structure of an intelligent monitoring and evaluation system for newly reclaimed farmland, which is an exemplary embodiment of the present invention. Detailed Implementation

[0059] The embodiments of the present invention will be described below with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention and not for limiting the scope of protection of the present invention.

[0060] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0061] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.

[0062] Example 1:

[0063] A method for intelligent monitoring and evaluation of newly reclaimed farmland, such as Figure 1 As shown, it includes:

[0064] S1: Based on the boundary vector and digital elevation model of newly reclaimed farmland, generate optimized flight paths and ground sampling grids for UAVs to establish an air-ground collaborative monitoring spatial framework;

[0065] S2: Acquire multispectral and thermal infrared images through UAV remote sensing, collect surface spectral data and soil samples on the ground based on sampling grids, and perform registration and preprocessing on UAV remote sensing images and ground-collected data.

[0066] S3: Conduct laboratory analysis on soil samples to obtain soil physicochemical indicators, and correlate the soil physicochemical indicators with the remote sensing spectral characteristics of the corresponding sampling points to construct a training sample set;

[0067] S4: Based on the training sample set, spectral features, thermal infrared features and historical time-series remote sensing data are integrated, and the spatial distribution map of soil physicochemical indicators of cultivated land is obtained by inverting through machine learning model.

[0068] S5: Integrate soil physicochemical indicators, crop growth dynamics and time-series characteristics and historical planting context information to classify arable land quality grades and identify soil obstacle types, and generate arable land quality grade distribution maps and obstacle type diagnosis maps.

[0069] S6: Based on the quality grade distribution map and obstacle type diagnosis map, combined with the agricultural knowledge rule base, agronomic management suggestions are generated, and then variable operation prescription maps are generated to drive precise variable operations.

[0070] The present invention is further configured such that S1 includes:

[0071] Read the boundary vector data and digital elevation model data of the newly reclaimed farmland area, and perform spatial reference consistency verification on the boundary vector data and digital elevation model data;

[0072] Based on the preset ground resolution and sensor field of view parameters, the flight altitude of the UAV is calculated, and a flight path planning file is generated according to the preset forward overlap and lateral overlap. The flight path planning file includes the spatial coordinates, flight parameters and sensor imaging angles of each waypoint.

[0073] Within the area defined by the boundary vector data, a sampling point array with a regular grid distribution is generated based on a preset sampling interval. Each grid point is assigned a unique identifier and given three-dimensional coordinates based on the digital elevation model (DEM). The output is a sampling grid point file containing point identifiers and spatial coordinates. Specifically, the boundary vector data and DEM data of the newly reclaimed farmland area are imported into professional geographic information system software. The boundary vector data is stored in a standard format, and the spatial reference uses a specific projection zone of the National Geodetic Coordinate System. Through field surveying or high-definition image interpretation, the boundaries of the newly reclaimed farmland to be monitored are accurately delineated. The DEM data is provided in a standard format and generated using advanced surveying technology, possessing high grid spacing accuracy. The elevation accuracy accurately reflects the micro-topographic undulations of the land parcel. Subsequently, a spatial reference consistency check is performed by simultaneously loading two files and automatically comparing key parameters such as coordinate system name, central meridian, projection method, and elevation datum using software tools. If inconsistencies are found, the system automatically or prompts the operator to perform a coordinate system unification conversion, ensuring all subsequent calculations are based on a unified spatial datum. Flight path planning aims to achieve a preset centimeter-level ground resolution. The specific implementation process includes: the flight control unit has a built-in sensor parameter database; based on the selected multispectral camera model, it automatically reads the sensor size and focal length parameters; and based on optical imaging principles, it automatically calculates the relative flight path required to achieve the target resolution. The flight altitude is determined by referencing digital elevation model data to ensure it adapts to terrain undulations. To guarantee image stitching quality, the ground coverage of a single image is calculated based on preset forward and lateral overlap requirements, combined with sensor field of view and flight altitude parameters, thus determining the forward and lateral spacing. Finally, using the verified boundaries of newly reclaimed farmland as the range, an optimal flight path is generated, with the path direction optimized to be perpendicular to the main terrain slope. The final generated flight path planning file contains complete information such as the spatial coordinates of all waypoints, flight parameters, and sensor attitude angles, and can be directly used for automated UAV flight missions. The flight parameters include flight altitude and flight speed. Ground sampling grid generation... By precisely matching aerial remote sensing data, the appropriate grid spacing is first determined based on the monitoring accuracy requirements, taking into account factors such as sampling efficiency and spatial variation representation capabilities. Subsequently, a regular grid sampling point array is generated in GIS software using a grid creation tool, with the boundary of newly reclaimed farmland as the range. Each grid point automatically acquires its planar coordinates and extracts its elevation value from the digital elevation model using interpolation methods, thereby forming a complete three-dimensional coordinate system. A unique identifier is assigned to each point. Finally, a sampling grid point file containing point identification and spatial coordinates is output to guide on-site sampling personnel in accurate positioning. This design process ensures the spatial benchmark uniformity and data acquisition coordination of the integrated air-space-ground monitoring framework.

[0074] The present invention is further configured such that S2 includes:

[0075] Using a multispectral imager and a thermal infrared imager configured on a UAV, aerial remote sensing data is collected along a preset flight path, and positioning and attitude data are recorded simultaneously. The aerial remote sensing data includes multispectral images and thermal infrared images.

[0076] Simultaneously employing a handheld ground object spectrometer and a real-time dynamic differential global navigation satellite system receiver, surface spectral measurements and soil sample collection were conducted at pre-set sampling grid points on the ground. The surface spectral measurements included vertical observations based on standard whiteboard correction and averaging of multiple measurements.

[0077] Aerial triangulation processing was performed on the collected aerial remote sensing data and positioning and attitude data. Among them, radiometric calibration and orthorectification were performed on the multispectral images, and temperature calibration and orthorectification were performed on the thermal infrared images.

[0078] The corrected multispectral image is fused with the thermal infrared image to generate an orthophoto containing multispectral band information and surface temperature information.

[0079] The spectral data and positioning coordinates collected from the ground are spatially registered with the orthophoto image to establish a spatial correspondence between ground sampling points and remote sensing image pixels. Specifically, this embodiment details the entire process of air-space-ground collaborative data acquisition and preprocessing. A UAV equipped with a high-precision real-time dynamic positioning system is used as the flight platform. This platform is equipped with both a multispectral imager and a thermal infrared imager. The multispectral imager is configured with multiple spectral bands to collect reflectance spectral data in the visible to near-infrared range. The thermal infrared imager is dedicated to capturing surface thermal radiation information. Flight operations are strictly performed according to a pre-planned flight path. The airborne positioning and attitude determination system continuously records the aircraft's spatial coordinates and attitude parameters at each exposure time at a preset sampling frequency. Spatial coordinates include longitude, latitude, and elevation. Attitude parameters include roll angle, pitch angle, and yaw angle. All positioning and attitude data are strictly synchronized with the acquisition time of each original image and stored in a dedicated data storage device. During the simultaneous ground data acquisition operation, investigators are equipped with a handheld ground feature spectrometer and a high-precision real-time dynamic differential global navigation satellite system receiver. Following the preset sampling grid coordinates, the satellite navigation receiver performs centimeter-level precision point layout. At each sampling point, the ground feature spectrometer lens is first vertically pointed downwards at the ground, maintaining a measurement height of approximately one meter. Before each measurement, the instrument is calibrated using a standard white board. Subsequently, multiple spectral curves are continuously acquired, and the instrument's built-in processor automatically calculates the average reflectance. The system acquires continuous spectral data from the visible to the shortwave infrared range. Simultaneously, soil samples are collected at specified depths using a stainless steel soil auger at each sampling point. The sample weight must meet laboratory analysis requirements. Collected samples are immediately placed in sealed bags marked with location numbers and stored in a portable insulated box at low temperatures to ensure stable physicochemical properties. After data acquisition, the system enters the data preprocessing stage. The raw images acquired by the UAV and the corresponding positioning and attitude data are imported into professional remote sensing processing software. Aerial triangulation is performed first, and a connection point network for the entire newly reclaimed farmland area is constructed by automatically matching common feature points of adjacent images. Combined adjustment calculations are then performed on the exterior orientation elements of all images to generate a high-density 3D point cloud and digital surface model. Based on this... The process involves radiometric calibration of the multispectral image, converting the raw digital quantization values ​​into surface reflectance, and performing orthorectification to eliminate topographic and angular distortions. Similarly, the thermal infrared image is calibrated by converting the radiance values ​​into surface temperature and undergoing orthorectification. The calibrated multispectral and thermal infrared images are then fused to generate an orthorectified image containing both multispectral and surface temperature information. To achieve precise matching of aerial and ground data, the accurate coordinates of ground sampling points are registered with the generated orthorectified image. Using the orthorectified image as a reference, the coordinates of each ground sampling point are used as control points. A coordinate transformation algorithm is employed to establish the correspondence between image-side coordinates and object-side coordinates, ensuring that each ground sampling point accurately corresponds to a specific pixel location on the image.Ultimately, orthophotos with accurate georeferenced coordinates, along with spectral data from ground sampling points and soil samples aligned with the image coordinate system, were obtained, establishing a complete data foundation for subsequent parameter inversion.

[0080] The present invention is further configured such that S3 includes:

[0081] The collected soil samples were analyzed in the laboratory to obtain soil physicochemical indicators, including soil organic matter content, pH value, total nitrogen content, available phosphorus content, and available potassium content.

[0082] Based on the registered orthophoto, multi-band reflectance data within a preset-size neighborhood window of the corresponding image location is extracted for each ground sampling point, and the arithmetic mean and standard deviation of the reflectance of each band are calculated.

[0083] The measured soil physicochemical indicators were spatially correlated and matched with the corresponding remote sensing spectral features extracted from orthophotos.

[0084] A training sample set is constructed based on the correlation matching results. This training sample set includes soil physicochemical index data and corresponding spectral characteristic data for each sampling point. Specifically, the collected soil samples are sent to a qualified testing laboratory for analysis. Soil organic matter content is determined using the potassium dichromate external heating method. This method oxidizes soil organic matter in a potassium dichromate solution under heating conditions, and the organic matter content is quantitatively calculated by titrating the remaining potassium dichromate. Soil pH is determined using the potentiometric method. After mixing soil with deionized water in a specific ratio, the solution potential value is directly measured using a precision pH meter. Total nitrogen content is determined using the Kjeldahl method. Persulfate digestion converts nitrogen in the soil into ammonium salts, and the nitrogen content is calculated after distillation titration. Available phosphorus content is determined using sodium bicarbonate extraction-molybdenum antimony colorimetric method; available phosphorus is extracted with a specific extractant and then quantitatively analyzed using a spectrophotometer. Available potassium content is determined using ammonium acetate extraction-flame photometry; potassium ion concentration is determined using a neutral ammonium acetate solution after extraction and then measured using a flame photometer. Based on the registered orthophoto image, remote sensing features are extracted for each ground sampling point. A neighborhood window of a preset size is selected, centered on the image coordinates corresponding to each sampling point, and the reflectance data of all pixels within this window in each band are extracted. Statistical characteristics of reflectance for each band within a window are calculated. These characteristics include the arithmetic mean and standard deviation. The arithmetic mean reflects the central tendency of the reflectance in that band, and the standard deviation characterizes the spatial variability of the reflectance. Finally, a set of spectral feature vectors containing the arithmetic mean and standard deviation of reflectance for each sampling point is obtained. These spectral feature vectors characterize the spectral response characteristics and spatial variability of that sampling point. Soil physicochemical indicators obtained from laboratory analysis are spatially correlated and matched with the corresponding spectral feature vectors extracted from remote sensing images. Each sample point contains two sets of data: one set measured in the laboratory. The dataset consists of two sets of data: soil physicochemical indices and spectral feature vectors extracted from remote sensing images. All sample data points are organized into a structured data table, with each sample point corresponding to a row and the soil physicochemical indices and spectral features serving as columns. To ensure the reliability and generalization ability of the model training, the complete dataset is randomly rearranged and divided into two subsets according to a predetermined ratio: the majority of samples are used as the training set for model training, and the remaining samples are used as the validation set for model validation. The final training sample set file contains sample point numbers, spectral feature data, and soil physicochemical indices, providing sufficient data support for the subsequent remote sensing inversion model establishment.

[0085] The present invention is further configured such that S4 includes:

[0086] Acquire time-series remote sensing data of newly reclaimed farmland for a preset historical period, filter and smooth the time-series remote sensing data, and generate a vegetation index growth time series.

[0087] The spectral features and vegetation index growth time series in the training sample set are fused to generate a fused feature vector.

[0088] Based on the soil physicochemical indicators and corresponding fused feature vectors in the training sample set, a soil physicochemical indicator inversion model is trained, and this model is used to invert the whole-domain orthophoto to generate a spatial distribution map of soil physicochemical indicators for cultivated land. Specifically, firstly, medium-resolution satellite image data of newly reclaimed cultivated land for multiple consecutive growing seasons are acquired from the remote sensing data cloud platform. The acquired raw images are preprocessed with radiometric calibration, atmospheric correction, and cloud detection. Based on the preprocessed image data, the normalized vegetation growth index of each pixel is calculated, and the original vegetation growth index sequence is smoothed and denoised using a time series reconstruction algorithm to eliminate the noise effects caused by cloud cover and atmospheric disturbance. Through preset time intervals, a continuous cloudless vegetation growth index time series is generated for the entire growing season. This vegetation growth index time series reflects the dynamic change pattern of vegetation growth at each pixel location in historical periods. For each pixel to be inverted, two types of features are extracted and fused. Firstly, preset local features are extracted. The spectral features of the window, including the arithmetic mean and standard deviation of reflectance in each band, reflect the spectral response characteristics of the pixel and its surrounding area. Secondly, the historical vegetation growth index time series features of the pixel location are analyzed. Principal component analysis is used to reduce the dimensionality of the original vegetation index growth time series, extracting the most representative temporal variation features. These two types of features are then combined to form a fused feature vector containing both spectral and temporal data. Based on the soil physicochemical indicators and corresponding fused feature vectors from the training sample set, a gradient boosting decision tree algorithm is used to train the soil physicochemical indicator inversion model. Through iterative optimization, the model learns the complex nonlinear mapping relationship from multi-source fused features to soil parameter values. After training, the model is applied to the entire newly reclaimed farmland area, generating a spatial distribution map of soil physicochemical indicators for each pixel. All inversion results are output in geographic raster data format, with each pixel value representing the predicted value of the soil physicochemical indicator at that location.

[0089] The present invention is further configured such that S5 includes:

[0090] The newly reclaimed farmland is divided into regular grid evaluation units of preset size, and each unit is assigned a unique identifier;

[0091] The arithmetic mean of soil physicochemical indicators for each evaluation unit is extracted from the spatial distribution map of soil physicochemical indicators, and a static indicator feature vector is constructed.

[0092] The current average biomass and historical biomass change trends of each evaluation unit are extracted from the multi-period crop biomass distribution map to construct a dynamic time series index feature vector.

[0093] Crop rotation information and yield levels of each evaluation unit are extracted from the historical planting database, and historical context feature vectors are constructed after feature encoding.

[0094] The static indicator feature vector, dynamic time-series indicator feature vector, and historical context feature vector are concatenated and merged to generate a comprehensive feature vector for each evaluation unit.

[0095] Based on historical farmland quality assessment results, each assessment unit is assigned farmland quality grade labels and obstacle type labels, and a deep learning assessment model is constructed with comprehensive feature vectors as input and quality grade classification and obstacle type identification as parallel output tasks.

[0096] A deep learning evaluation model is trained and validated using a labeled evaluation unit dataset. The trained model is then applied to predict the quality level and obstacle type of the comprehensive feature vectors of all evaluation units across the entire region, generating a quality level distribution map and an obstacle type diagnostic map for cultivated land. Specifically, a regular grid method is used in the geographic information system platform to divide newly reclaimed cultivated land into evaluation units. Square grids covering the entire region are generated as basic evaluation units with a preset size, and each unit is assigned a unique identifier. The design of the evaluation unit size must comprehensively consider the spatial detail representation capability and data stability requirements, ensuring that the unit size is smaller than the size of medium- and low-resolution satellite pixels to fully reflect the differences within the land parcels, while avoiding data noise caused by excessively small evaluation units. Sound amplification; In the feature extraction stage, a hierarchical processing strategy is adopted to construct a multi-dimensional feature system; The static feature layer uses the generated spatial distribution map of soil physicochemical indicators to calculate the arithmetic mean of all pixel values ​​in each evaluation unit using regional statistical methods, extracts the arithmetic mean of the evaluation unit of soil physicochemical indicators, and forms a static indicator feature vector representing the background characteristics of the soil; The dynamic temporal feature layer focuses on analyzing the crop growth change pattern, obtains crop biomass monitoring data for multiple consecutive growing seasons, and calculates two temporal features for each evaluation unit. The two temporal features include the average biomass value of the latest phase reflecting the current growth status and the linear regression slope based on data from the same period over many years representing the trend of biomass change. This temporal feature constitutes the dynamic temporal indicator feature vector; Historical The contextual feature layer extracts crop rotation information from the planting database, uses one-hot encoding to transform crop types into numerical features, normalizes historical yield data, and combines a quantification index of crop rotation pattern complexity to construct a historical contextual feature vector reflecting the planting system background. After feature extraction, the three types of feature vectors are concatenated dimensionally to form a comprehensive feature vector. The model training phase then begins by acquiring historical farmland quality evaluation results. Spatial overlay analysis is used to assign quality level and obstacle type labels to each evaluation unit. Based on this, a deep learning evaluation model is constructed. The number of nodes in the input layer matches the feature dimensions, the hidden layer uses a fully connected structure, and the output layer is designed with a multi-task learning architecture to simultaneously classify quality levels and obstacles. Type identification; Model training employs a standardized data processing workflow, dividing the sample set proportionally into training, validation, and test sets; the loss function is calculated through forward propagation, and parameters are optimized through backpropagation using an adaptive moment estimation algorithm, with an early stopping strategy applied to prevent overfitting; Model performance is comprehensively evaluated on the test set to ensure good generalization ability; The trained model is applied to the global evaluation unit to generate a farmland quality grade distribution map and an obstacle type diagnostic map; The quality grade distribution map uses discrete scores to represent the quality status, while the obstacle type diagnostic map uses multi-band coding or attribute tables to identify the spatial distribution of various obstacles. All resulting maps maintain a spatial reference system consistent with the original data, providing a scientific basis for subsequent precision agriculture management.

[0097] The present invention is further configured such that S6 includes:

[0098] A structured agricultural knowledge rule base is constructed, which includes judgment rules based on the combination of arable land quality grade, obstacle type and soil physicochemical index threshold, as well as agronomic management measures corresponding to each combination of conditions.

[0099] Based on the classification results of cultivated land quality grades, the diagnosis results of obstacle types, and the quantitative values ​​of soil physicochemical indicators of each evaluation unit, differentiated agronomic management suggestions for each evaluation unit are generated by matching the corresponding rules in the agricultural knowledge rule base.

[0100] Using agronomic management recommendations and spatial distribution maps of soil physicochemical indicators as input, this invention generates variable operation prescription maps based on preset decision logic. Further, the invention transmits these variable operation prescription maps to an intelligent agricultural equipment control system via a standard data interface, driving variable execution mechanisms to achieve precise operations. Specifically, a structured agricultural knowledge rule base is established, stored using Extensible Markup Language (EXPLAIN) or a Lightweight Data Exchange (LDLE) format to ensure readability and maintainability. Each rule consists of a condition judgment section and a measure recommendation section. The condition judgment section uses multi-dimensional combinations of conditions, including quality level, obstacle type, and threshold ranges for soil physicochemical indicators. The measure recommendation section provides specific agronomic operation plans, including detailed information such as fertilizer type, dosage, and irrigation. The rule matching process is implemented using an automated algorithm, traversing each evaluation unit, reading the corresponding quality level, obstacle type, and soil physicochemical indicators, and matching them one by one with the conditions in the rule base. When the characteristics of an evaluation unit satisfy all the conditions of a specific rule, the agronomic measures corresponding to that rule are automatically assigned to that evaluation unit. For complex cases where multiple rules are satisfied simultaneously, all applicable measures are integrated. This process forms a comprehensive management plan. The variable operation prescription map is generated based on agronomic management recommendations and spatial distribution maps of soil physicochemical indicators. Taking variable fertilization as an example, the agronomic management recommendations in text form are first converted into quantitative operational parameters, such as the specific fertilizer application rate. Then, the spatial distribution map of soil physicochemical indicators is imported as the base map data into a professional agricultural management software platform. Through the built-in decision logic engine, the optimal fertilizer application rate is calculated based on the quality level, obstacle type, and current soil nutrient status of each spatial location, combined with the target yield requirements of the crop. The calculation process comprehensively considers the principle of soil nutrient balance and the crop's nutrient requirement law to ensure the scientificity and accuracy of the recommended application rate. Finally, a variable operation prescription map conforming to the international agricultural machinery bus standard is generated. This prescription map is in raster data format, and each pixel value represents the precise fertilizer application rate at that spatial location. The generated variable operation prescription map is transmitted to the intelligent agricultural equipment control system through a universal serial bus storage device or wireless communication network. During the fertilization operation, the controller of the variable fertilizer applicator obtains the machine's position information in real time and adjusts the fertilizer application rate according to the instructions in the prescription map. The fertilizer discharge device is precisely controlled through an electric or hydraulic drive system to ensure on-demand fertilization.

[0101] The present invention is further configured such that the method also includes visually displaying the quality grade distribution map, obstacle type diagnosis map, and variable operation prescription map of cultivated land; specifically, the generated quality grade distribution map, obstacle type diagnosis map, and variable operation prescription map are integrated and displayed through a visualization interface; the quality grade distribution map uses a gradient color scheme to intuitively present the spatial distribution of cultivated land of different grades, the obstacle type diagnosis map uses differentiated symbols to identify the distribution location of various obstacle factors, and the variable operation prescription map displays the agronomic operation requirements of different areas in a layered color scheme.

[0102] Example 2:

[0103] Please see Figure 2 This exemplary intelligent monitoring and evaluation system for newly reclaimed farmland includes:

[0104] Collaborative Framework Module: Based on the boundary vector and digital elevation model of newly reclaimed farmland, it generates optimized flight paths for UAVs and ground sampling grids to establish an air-ground collaborative monitoring spatial framework;

[0105] Collaborative acquisition module: acquires multispectral and thermal infrared images through UAV remote sensing, collects surface spectral data and soil samples on the ground based on sampling grid, and performs registration and preprocessing on UAV remote sensing images and ground-collected data;

[0106] Truth value construction module: Conduct laboratory analysis on soil samples to obtain soil physicochemical indicators, and associate the soil physicochemical indicators with the remote sensing spectral characteristics of the corresponding sampling points to construct a training sample set;

[0107] Collaborative Inversion Module: Based on the training sample set, it integrates spectral features, thermal infrared features and historical time-series remote sensing data, and uses a machine learning model to invert the spatial distribution map of soil physicochemical indicators of cultivated land.

[0108] Evaluation module: Integrates soil physicochemical indicators, crop growth dynamic time series characteristics and historical planting context information to classify arable land quality grades and identify soil obstacle types, and generate arable land quality grade distribution maps and obstacle type diagnosis maps;

[0109] Decision Management Module: Based on the quality level distribution map and obstacle type diagnosis map, combined with the agricultural knowledge rule base, agronomic management suggestions are generated, and then variable operation prescription maps are generated to drive precise variable operations.

[0110] It should be noted that the intelligent monitoring and evaluation system for newly reclaimed arable land provided in the above embodiments and the intelligent monitoring and evaluation method for newly reclaimed arable land provided in the above embodiments belong to the same concept. The specific operation methods of each module and unit have been described in detail in the method embodiments and will not be repeated here. In practical applications, the intelligent monitoring and evaluation system for newly reclaimed arable land provided in the above embodiments can be assigned to different functional modules as needed, that is, the internal structure of the system can be divided into different functional modules to complete all or part of the functions described above. This is not a limitation here.

[0111] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for intelligent monitoring and evaluation of newly cultivated land, characterized in that, include: S1: Based on the boundary vector and digital elevation model of newly reclaimed farmland, generate optimized flight paths and ground sampling grids for UAVs to establish an air-ground collaborative monitoring spatial framework; S2: Acquire multispectral and thermal infrared images through UAV remote sensing, collect surface spectral data and soil samples on the ground based on sampling grids, and perform registration and preprocessing on UAV remote sensing images and ground-collected data. S3: Conduct laboratory analysis on soil samples to obtain soil physicochemical indicators, and correlate the soil physicochemical indicators with the remote sensing spectral characteristics of the corresponding sampling points to construct a training sample set; S4: Acquire time-series remote sensing data of newly reclaimed farmland for a preset historical period, filter and smooth the time-series remote sensing data to generate a vegetation index growth time series; fuse the spectral features of the training sample set with the vegetation index growth time series to generate a fused feature vector; train a soil physicochemical index inversion model based on the soil physicochemical index in the training sample set and the corresponding fused feature vector, and use the model to invert the whole-domain orthophoto to generate a spatial distribution map of soil physicochemical index of farmland. S5: Divide newly reclaimed farmland into regular grid evaluation units of preset size, and assign a unique identifier to each unit; extract the arithmetic mean of soil physicochemical indicators for each evaluation unit from the spatial distribution map of soil physicochemical indicators, and construct a static indicator feature vector; extract the current average biomass and historical biomass change trend of each evaluation unit from the multi-period crop biomass distribution map, and construct a dynamic time-series indicator feature vector; extract crop rotation information and yield level of each evaluation unit based on the historical planting database, perform feature encoding, and construct a historical context feature vector; combine the static indicator feature vector, dynamic time-series indicator feature vector, and historical context feature vector. Vectors are concatenated and fused to generate a comprehensive feature vector for each evaluation unit. Based on historical farmland quality evaluation results, each evaluation unit is assigned a farmland quality level label and an obstacle type label. A deep learning evaluation model is constructed with the comprehensive feature vector as input and quality level classification and obstacle type identification as parallel output tasks. The deep learning evaluation model is trained and validated using a dataset of labeled evaluation units. The trained deep learning evaluation model is then applied to predict the quality level and obstacle type of the comprehensive feature vectors of all evaluation units in the entire domain, generating a farmland quality level distribution map and an obstacle type diagnostic map. S6: Based on the quality grade distribution map and obstacle type diagnosis map, combined with the agricultural knowledge rule base, agronomic management suggestions are generated, and then variable operation prescription maps are generated to drive precise variable operations.

2. The intelligent monitoring and evaluation method for newly reclaimed arable land according to claim 1, characterized in that, S1 includes: Read the boundary vector data and digital elevation model data of the newly reclaimed farmland area, and perform spatial reference consistency verification on the boundary vector data and digital elevation model data; Based on the preset ground resolution and sensor field of view parameters, the flight altitude of the UAV is calculated, and a flight path planning file is generated according to the preset forward overlap and lateral overlap. The flight path planning file includes the spatial coordinates, flight parameters and sensor imaging angles of each waypoint. Within the area defined by the boundary vector data, a sampling point array with a regular grid distribution is generated based on a preset sampling interval. Each grid point is assigned a unique identifier and given three-dimensional coordinates based on the digital elevation model. The output is a sampling grid point file containing point identifiers and spatial coordinates.

3. The intelligent monitoring and evaluation method for newly reclaimed arable land according to claim 1, characterized in that, S2 includes: Using a multispectral imager and a thermal infrared imager configured on a UAV, aerial remote sensing data is collected along a preset flight path, and positioning and attitude data are recorded simultaneously. The aerial remote sensing data includes multispectral images and thermal infrared images. Simultaneously employing a handheld ground object spectrometer and a real-time dynamic differential global navigation satellite system receiver, surface spectral measurements and soil sample collection were conducted at pre-set sampling grid points on the ground. The surface spectral measurements included vertical observations based on standard whiteboard correction and averaging of multiple measurements. Aerial triangulation processing was performed on the collected aerial remote sensing data and positioning and attitude data. Among them, radiometric calibration and orthorectification were performed on the multispectral images, and temperature calibration and orthorectification were performed on the thermal infrared images. The corrected multispectral image is fused with the thermal infrared image to generate an orthophoto containing multispectral band information and surface temperature information. The spectral data and positioning coordinates collected on the ground are spatially registered with the orthophoto to establish the spatial correspondence between ground sampling points and remote sensing image pixels.

4. The intelligent monitoring and evaluation method for newly reclaimed farmland according to claim 1, characterized in that, S3 includes: The collected soil samples were analyzed in the laboratory to obtain soil physicochemical indicators, including soil organic matter content, pH value, total nitrogen content, available phosphorus content, and available potassium content. Based on the registered orthophoto, multi-band reflectance data within a preset-size neighborhood window of the corresponding image location is extracted for each ground sampling point, and the arithmetic mean and standard deviation of the reflectance of each band are calculated. The measured soil physicochemical indicators were spatially correlated and matched with the corresponding remote sensing spectral features extracted from orthophotos. A training sample set is constructed based on the correlation matching results. The training sample set contains soil physicochemical index data and its corresponding spectral feature data for each sampling point.

5. The intelligent monitoring and evaluation method for newly reclaimed arable land according to claim 1, characterized in that, S6 includes: A structured agricultural knowledge rule base is constructed, which includes judgment rules based on the combination of arable land quality grade, obstacle type and soil physicochemical index threshold, as well as agronomic management measures corresponding to each combination of conditions. Based on the classification results of cultivated land quality grades, the diagnosis results of obstacle types, and the quantitative values ​​of soil physicochemical indicators of each evaluation unit, differentiated agronomic management suggestions for each evaluation unit are generated by matching the corresponding rules in the agricultural knowledge rule base. Using agronomic management recommendations and spatial distribution maps of soil physicochemical indicators as input, variable operation prescription maps are generated based on preset decision logic.

6. The intelligent monitoring and evaluation method for newly reclaimed arable land according to claim 5, characterized in that, The variable operation prescription map is transmitted to the intelligent agricultural equipment control system through a standard data interface, driving the variable execution mechanism to achieve precise operation.

7. The intelligent monitoring and evaluation method for newly reclaimed arable land according to claim 1, characterized in that, The method also includes visualizing the distribution map of farmland quality grades, the diagnostic map of obstacle types, and the variable operation prescription map.

8. A smart monitoring and evaluation system for newly reclaimed arable land, used to implement the smart monitoring and evaluation method for newly reclaimed arable land as described in any one of claims 1-7, characterized in that, include: Collaborative Framework Module: Based on the boundary vector and digital elevation model of newly reclaimed farmland, it generates optimized flight paths for UAVs and ground sampling grids to establish an air-ground collaborative monitoring spatial framework; Collaborative acquisition module: acquires multispectral and thermal infrared images through UAV remote sensing, collects surface spectral data and soil samples on the ground based on sampling grid, and performs registration and preprocessing on UAV remote sensing images and ground-collected data; Truth value construction module: Conduct laboratory analysis on soil samples to obtain soil physicochemical indicators, and associate the soil physicochemical indicators with the remote sensing spectral characteristics of the corresponding sampling points to construct a training sample set; Collaborative Inversion Module: Based on the training sample set, it integrates spectral features, thermal infrared features and historical time-series remote sensing data, and uses a machine learning model to invert the spatial distribution map of soil physicochemical indicators of cultivated land. Evaluation module: Integrates soil physicochemical indicators, crop growth dynamic time series characteristics and historical planting context information to classify arable land quality grades and identify soil obstacle types, and generate arable land quality grade distribution maps and obstacle type diagnosis maps; Decision Management Module: Based on the quality level distribution map and obstacle type diagnosis map, combined with the agricultural knowledge rule base, agronomic management suggestions are generated, and then variable operation prescription maps are generated to drive precise variable operations.

Images