Sugarcane abnormal growth alarm system based on remote sensing monitoring model

Abstract

The invention relates to the field of alarm systems, in particular to a sugarcane abnormal growth alarm system based on a remote sensing monitoring model, which comprises the following modules: a multi-source remote sensing image preprocessing module, a remote sensing parameter acquisition module, a ground measured data coupling module, an abnormal growth monitoring identification model module and a risk assessment and alarm module. The system can achieve the alarm processing of the abnormal growth of various sugarcanes, solves the problems that the conventional manual monitoring of the abnormal growth of the sugarcane field is low in efficiency and insufficient in coverage, the remote sensing technology is less and single in application, and the abnormal growth of the sugarcanes cannot be comprehensively mastered, early warned and alarmed, and can effectively improve the recognition, prevention and processing efficiency of the abnormal growth condition of the sugarcane field.

Description

Technical Field

[0001] The present invention relates to the field of alarm systems, and more particularly to a sugarcane abnormal growth alarm system based on a remote sensing monitoring model. Background Art

[0002] The abnormal growth of sugarcane is mainly caused by various factors, including natural disasters such as drought, flood, temperature anomaly and typhoon in terms of climate, resulting in abnormal growth of sugarcane, soil and nutrients such as poor soil quality and nutrient imbalance, diseases and pests, and other factors such as too high planting density and poor field management. These factors also affect the normal growth of sugarcane at different levels. At present, data on the planting area, growth, yield, diseases, insects and weeds, and disaster conditions of sugarcane in the whole region of Guangxi are mainly obtained through traditional methods such as on-site investigation by sugar enterprise agricultural staff, reporting level by level, and summarization. There is still a lack of big data related to sugarcane production. Even the collected data is fragmented and it is difficult to guide production in turn, unable to reach the level of digital industrialization, and even more unable to reconstruct the sugarcane industry production, abnormal growth prevention and treatment methods through mathematical models and AI deduction to achieve cost reduction and efficiency increase of sugarcane. The prevention and treatment of abnormal growth can be monitoring and early warning, and the existing technology does not have relatively advanced treatment means for the abnormal growth of sugarcane.

[0003] The application of multi-source remote sensing technology in other fields has been very mature and there are also some applications in agricultural production. Remote sensing technology can accurately obtain the growth status of crops in farmland, including information such as growth stage and growth density. Farmers can carry out precise fertilization, irrigation and pest control on the farmland based on these data. For some abnormal growth situations, there are also applications. For example, through remote sensing monitoring, it is found that the growth of a certain crop in some areas of a farmland is weak, and there has been long-term in-depth research or experience on the weak growth of this crop, such as due to the lack of certain nutrients, such as the lack of nutrient components such as nitrogen, phosphorus and potassium. When remote sensing monitoring discovers this weakness, corresponding nitrogen, phosphorus and potassium fertilizers can be supplemented in these areas, avoiding the blindness of uniform fertilization in the whole field, improving the utilization efficiency of fertilizers and reducing resource waste. However, the current applications are still very single. Because the abnormal growth of different crops varies greatly, it cannot be generally applicable. At the same time, single applications will also cause waste of resources and cannot monitor more comprehensively.

[0004] Although remote sensing technology has been applied in sugarcane planting production, it has not been applied to the monitoring and early warning of abnormal growth conditions in sugarcane fields. At present, the abnormal growth conditions in sugarcane fields still adopt traditional manual monitoring, with low efficiency and insufficient coverage, completely relying on post-disaster loss assessment and experience, and unable to give early warning, which is very inconsistent with the development of remote sensing technology monitoring. Summary of the Invention

[0005] To address the problems of low efficiency and insufficient coverage in the traditional manual monitoring of abnormal growth conditions in sugarcane fields, which can only rely on post-disaster loss assessment and experience, and the limited and single application of remote sensing technology in this aspect of sugarcane fields, making it impossible to comprehensively grasp various types of abnormal growth for early warning and alarm, the present invention provides a sugarcane abnormal growth alarm system based on a remote sensing monitoring model to solve the above problems.

[0006] The technical solution of the present invention is as follows: The present invention provides a sugarcane abnormal growth alarm system based on a remote sensing monitoring model, characterized in that: the abnormal growth refers to five types: diseases and pests, lodging, slow growth, land drought, and flood of sugarcane; the alarm system includes: Multi-source remote sensing image preprocessing module: Collect C-band dual-polarization SAR of sugarcane fields, multi-spectral L2A-level data, UAV multi-spectral data, and airborne hyperspectral data, perform calibration and data fusion, and output a standardized preprocessed image dataset; Remote sensing parameter acquisition module: Perform inversion calculations on the above dataset to obtain parameter values of leaf area index LAI, canopy water content NDWI, chlorophyll content, SAR dielectric constant, and NDRE; Ground measured data coupling module: Couple and calibrate the above parameter values with their corresponding ground measured data and sugarcane planting agricultural situation data, and output the calibrated parameter values; Abnormal growth monitoring and recognition model module: Combine airborne hyperspectral data with the sugarcane diseases and pests sensitive spectral feature library to generate a primary diseases and pests probability distribution map, and combine its data with the calibrated parameter values and sugarcane field geographic information to respectively construct five sugarcane abnormal growth monitoring and recognition models, and respectively output abnormal growth distribution maps with preliminary probabilities; Risk assessment and alarm module: Integrate historical abnormal growth data of sugarcane fields and the sugarcane diseases and pests expert knowledge base to construct a dynamic abnormal growth threshold library, and set alarm levels corresponding to the threshold library; Integrate the abnormal growth distribution map with preliminary probabilities and measured data to obtain the final abnormal growth distribution probability, and compare it with the threshold library. When the comparison result meets the alarm classification conditions, automatically trigger the corresponding level of alarm.

[0007] Preferably, in the multi-source remote sensing image preprocessing module, the C-band dual-polarization SAR data comes from the Sentinel-1 satellite, the multi-spectral L2A-level data comes from the Sentinel-2 satellite, the UAV multi-spectral data comes from the DJI P4 Multispectral, and the airborne hyperspectral data comes from the Headwall Nano-Hyperspec imager; The calibration includes radiometric calibration and geometric calibration; The radiation correction is as follows: Gamma MAP filtering is used to eliminate the speckle noise of C-band dual-polarization SAR data, and terrain radiation correction is performed based on 30m DEM; the airborne hyperspectral data is calibrated by the whiteboard reflectance; the 6S model is used to perform atmospheric correction on the multispectral remote sensing image data, and the solar elevation angle normalization is performed using the SNAP tool. The geometric correction is as follows: all images are registered to the WGS84 UTM coordinate system through GCPs with an accuracy of <0.3 pixels. The SAR images are additionally applied with Range-Doppler terrain correction, and the UAV data achieves centimeter-level positioning through PPK post-processing. The data fusion in the multi-source remote sensing image preprocessing module specifically refers to: fusing the C-band dual-polarization SAR and multispectral L2A-level data through the non-subsampled contourlet transform (NSCT) to enhance the texture features of the sugarcane field lodging area; the data of the standardized preprocessed image dataset is the standardized GeoTIFF image.

[0008] Preferably, in the remote sensing parameter acquisition module, the leaf area index (LAI) parameter value is retrieved based on the PROSAIL model, combined with the optimization of the red-edge band of the multispectral L2A-level data, with an inversion error of <0.5 and R² = 0.88; the canopy water content NDWI parameter value is obtained by using the water-sensitive characteristic band in the UAV hyperspectral data, through the dual-band ratio method, and co-verified with the SMAP soil moisture data; the chlorophyll content is calculated through the REIP red-edge inflection point position index of the sugarcane pest and disease sensitive band; the SAR dielectric constant is retrieved from the C-band dual-polarization SAR data; the NDRE index value is calculated from the reflectance of the red-edge band and the near-infrared band in the multispectral L2A-level data and the airborne hyperspectral data.

[0009] Preferably, among the ground measured data corresponding to the parameter values output by the remote sensing parameter acquisition module, the leaf area index LAI is measured using a leaf area meter, the canopy water content NDWI and the chlorophyll content are measured using a field portable spectrometer, the SAR dielectric constant is measured using a portable soil dielectric sensor; the NDRE index value is calculated using a calculation formula after collecting the canopy reflectance spectrum through a portable hyperspectral spectrometer; the sugarcane planting agricultural situation data includes the RTK measurement data of the sugarcane plant height and stem thickness, the soil temperature and humidity of the sugarcane field, the geospatial information, the meteorological data, and the field soil water holding capacity; the geospatial information includes the DEM terrain data.

[0010] Preferably, in the ground measured data coupling module, the coupling process is as follows: Leaf Area Index LAI: A regression relationship is established between the measured Leaf Area Index LAI and the Leaf Area Index LAI retrieved from multi-spectral L2A data. When R² > 0.85, the model coefficients are updated to improve the inversion accuracy; Canopy Water Content NDWI: Taking the NDWI calculated from the airborne hyperspectral band of 950 - 1700 nm as the reference value, when the deviation from the measured value by the portable spectrometer > 10%, the following are executed in sequence: spatial consistency test of the Sentinel-1 SAR surface water index and the SMAP soil moisture product, spatial correlation analysis of the terrain humidity index TWI and NDWI, and cross-validation of the field soil water holding capacity sensor data. When the passing rate of the three validations ≥ 2 / 3, the final NDWI value is output; Chlorophyll Content: Based on the comparison between the calculated value and the measured value of the REIP red edge inflection point index, a compensation mechanism is triggered when the absolute error > 15%; SAR Dielectric Constant: A regression model is constructed through the VV / VH polarization ratio and the measured value. The Freeman-Durden decomposition is used to correct the scattering mechanism in the area with dielectric constant mutation, and the RTK data of sugarcane plant height is combined to eliminate the multiple scattering interference; NDRE Index: Calibrated stage by stage according to the growth period: the datasets of the seedling stage, tillering stage, jointing stage, and maturity stage are divided; the spatial autocorrelation is tested by the Moran's I index within each growth period; Kriging interpolation is used to supplement the missing data; Temporal data is obtained through the Leaf Area Index LAI, and a historical healthy sugarcane field reference curve is constructed with the measured stem diameter data by RTK in the same period.

[0011] Preferably, in the abnormal growth monitoring model recognition module, the monitoring and recognition models for the five abnormal growths of sugarcane are: pest and disease monitoring and recognition model, lodging monitoring and recognition model, slow growth monitoring and recognition model, land drought monitoring and recognition model, and flood monitoring and recognition model; The primary pest and disease probability distribution map is obtained by comparing with the healthy sugarcane field spectral library based on the Mahalanobis distance and detecting abnormal spectra, and its threshold is set at a 95% confidence interval; The pest and disease monitoring and identification model performs spatial registration on the primary pest and disease probability distribution map, SAR dielectric constant, chlorophyll content, and the geographical information of the field boundary. It resamples satellite data to the field scale through bilinear interpolation to capture multi-dimensional disease characteristics. It uses Spectral Angle Matcher (SAM) to accurately identify the specific spectral fingerprints of sugarcane pests and diseases. During the spectral angle matching process, it introduces the field ID attribute field to calculate the spectral similarity only for the pixels within the same geographical unit, excluding the interference of adjacent fields. It combines Random Forest Classification to integrate NDVI, Normalized Difference Red Edge (NDRE), and SAR texture features, and outputs a pest and disease heat map located in specific fields through a probability space model. Preferably, the lodging monitoring and identification model first extracts high-frequency texture and spectral features from the NSCT fusion result, then corrects the elevation of the digital surface model using the RTK-measured plant height data to eliminate the deviation caused by insufficient vegetation coverage. Subsequently, it couples the corrected plant height data with SAR texture features, dynamically adjusts the local variance threshold through the stem diameter measurement value to enhance the identification of lodging-sensitive areas. It then combines three-dimensional morphological methods to optimize the spatial connectivity of discrete scatter points, where the dilation kernel size is adaptively adjusted according to the median of the stem diameter. Finally, it outputs a lodging probability distribution map that integrates multi-dimensional features.

[0012] Preferably, the slow growth monitoring and identification model matches the current Leaf Area Index (LAI) time series curve with the historical healthy sugarcane field benchmark curve through Dynamic Time Warping (DTW), calculates the pixel-by-pixel deviation rate, combines the stem diameter RTK data and the NDRE index for random forest probability modeling, and outputs a slow growth probability distribution map. The land drought monitoring and identification model fuses the Normalized Difference Water Index (NDWI) of the canopy water content and the SAR dielectric constant to construct a Water Stress Index (WSI), and generates a spatially continuous drought probability distribution map through Kriging interpolation. The flood monitoring and identification model detects water accumulation areas based on the change in SAR backscattering coefficient, superimposes DEM terrain data to simulate the flooding diffusion path, corrects false detections by combining field water holding capacity data, determines the water accumulation range, and finally outputs a water accumulation probability distribution map.

[0013] Preferably, in the risk assessment and alarm module, the dynamic abnormal growth classification threshold library is obtained through sliding time window statistical analysis and machine learning algorithms. The measured data used to obtain the final abnormal growth distribution probability are the data collected in real time by field Internet of Things devices, including: soil moisture, meteorological data, and the NDVI heat map of drone inspections; the meteorological data includes a 72-hour rainfall forecast.

[0014] Preferably, the responses corresponding to the alarm levels are as follows: The alarm levels are divided into three levels, namely: Level 1 alarm, triggered when a single indicator exceeds the threshold; Level 2 alarm, triggered when there are multiple indicator composite risks; Level 3 alarm, triggered by sudden and urgent climate disasters.

[0015] Preferably, in the risk assessment and alarm module, the dynamic abnormal growth classification threshold library is continuously optimized through the sugarcane pest expert knowledge base and real-time feedback.

[0016] The beneficial effects of the present invention are as follows: A sugarcane abnormal growth alarm system based on a remote sensing monitoring model provided by the present invention can achieve alarm processing for various sugarcane abnormal growths through comprehensive processing of multi-source remote sensing images, solving the problems of low efficiency and insufficient coverage of traditional manual monitoring of abnormal growth in sugarcane fields, and the rare and single application of remote sensing technology, which cannot comprehensively grasp sugarcane abnormal growth to give early warning and alarm. Specifically, the C-band dual-polarization SAR data collected by the multi-source remote sensing image preprocessing module can penetrate clouds and rain and fog, capture the texture features of the surface structure, perform soil moisture inversion, and can achieve all-weather monitoring of sugarcane lodging, drought and flood, but its image is not intuitive in terms of color information; the multi-spectral L2A level data has the characteristics of wide coverage and periodicity, can provide rich color and vegetation information, extract the bands (such as green, red and near-infrared) and their indices (such as NDVI) that can be used to identify vegetation lodging, and can be used to monitor some abnormal growth conditions in large areas of sugarcane fields. The red-edge band data contained therein can be used to quantify chlorophyll content, and can identify the phenomenon of chlorophyll reduction in the early stage of pest and disease incidence, and the phenomenon of chlorophyll reduction caused by too slow growth, so it can be used for the monitoring of lodging, pests and diseases, and too slow growth; the unmanned aerial vehicle multi-spectral data has the characteristics of high resolution and real-time, can identify the leaf patches caused by pests and diseases, generate a real-time NVDI heat map, and accurately locate the sugarcane plant areas with abnormal growth due to pests and diseases; the airborne hyperspectral data can combine the spectral characteristics of pests and diseases to find early pest and disease phenomena; the remote sensing parameter acquisition module inversely calculates and obtains relevant physiological parameters such as leaf area index LAI and chlorophyll content and other parameter values, captures stress signals 7-10 days before visible symptoms appear in sugarcane, and can timely obtain parameter values related to lodging, too slow growth, drought and flood, changing the traditional post-disaster assessment mode, achieving early prediction and alarm of abnormal growth, and can be timely intervened manually to reduce economic losses. The pest and disease probability distribution map is generated based on the reflectance of pest and disease characteristic bands, and both the efficiency and accuracy are greatly improved compared with manual random sampling. The ground measured data coupling module introduces ground measured data and geospatial data to correct the remote sensing inversion deviation and solve the problem of regional misjudgment caused by empirical judgment. The abnormal growth monitoring and identification module accurately identifies pests and diseases, lodging, too slow growth, land drought and flood conditions through establishing a multi-model collaborative framework, and forms a closed-loop optimization mechanism of "remote sensing inversion - ground verification - model iteration", replacing manual empirical speculation, and improving the accuracy and efficiency. The risk assessment and alarm module adopts a multi-level dynamic response mechanism, realizes accurate early warning through real-time data fusion and intelligent decision-making. The grading response mechanism sets different alarm levels according to different abnormal situations, enabling farmers to carry out early prevention or timely remedy according to different abnormal situations, making the prevention and remedy measures more targeted and effective.The sugarcane abnormal growth alarm system based on the remote sensing monitoring model provided by the present invention makes full use of the existing data resources, can monitor and alarm various different sugarcane abnormal growths at the same time, and effectively improves the resource utilization rate and the working efficiency of the equipment. Brief Description of the Drawings

[0017] Figure 1 It is a model diagram of the sugarcane abnormal growth alarm system based on the remote sensing monitoring model.

[0018] Figure 2 It is a schematic diagram of the NSCT fusion process. Specific Embodiments

[0019] The technical solutions of the present invention will be described in detail below in conjunction with the embodiments and the drawings. It should be understood that the following embodiments are only for explanation and illustration, and are not used to limit the protection scope of the present invention.

[0020] The present invention provides a sugarcane abnormal growth alarm system based on a remote sensing monitoring model, which is characterized in that: the abnormal growth refers to five types of sugarcane diseases and pests, lodging, slow growth, land drought, and flood; the alarm system includes: Multi-source remote sensing image preprocessing module: Collect C-band dual-polarization SAR, multi-spectral L2A level data, UAV multi-spectral, and airborne hyperspectral data of the sugarcane field, perform calibration and data fusion, and output a standardized preprocessed image data set; Remote sensing parameter acquisition module: Perform inversion calculations on the above data set to obtain parameter values of leaf area index LAI, canopy water content NDWI, chlorophyll content, SAR dielectric constant, and NDRE; Ground measured data coupling module: Couple and model calibrate the above parameter values with their corresponding ground measured data and sugarcane planting agricultural situation data, and output the calibrated parameter values; Abnormal growth monitoring and recognition model module: Combine airborne hyperspectral data with the sugarcane diseases and pests sensitive spectral feature library to generate a primary diseases and pests probability distribution map, and combine its data with the calibrated parameter values and the sugarcane field geographical information to respectively construct five sugarcane abnormal growth monitoring and recognition models, and respectively output abnormal growth distribution maps with preliminary probabilities; Risk assessment and alarm module: Integrate the historical abnormal growth data of sugarcane fields and the expert knowledge base of sugarcane pests and diseases, construct a dynamic abnormal growth threshold library, and set the alarm levels corresponding to the threshold library; Integrate the abnormal growth distribution map with preliminary probability and the measured data to obtain the final abnormal growth distribution probability, and compare it with the threshold library. When the comparison result meets the alarm classification conditions, the corresponding level of alarm is automatically triggered. In some embodiments of the present invention, the spectral feature library of sugarcane pests and diseases integrates the spectral diagnosis indicators of 6 diseases such as red rot disease in the "China Sugarcane Pests and Diseases Control Manual", the reflectance thresholds of 20 characteristic bands of smut disease and mosaic disease provided by the Guangxi Agricultural Cloud Platform, and the pest and disease related bands of Sentinel-2 that can be extracted from the opernicus satellite data.

[0021] In some embodiments of the present invention, the expert knowledge base of sugarcane pests and diseases is a database containing information on 12 types of pests and diseases such as pokkah boeng and stem borer, which is built by integrating 5-year field monitoring data of the Guangxi Agricultural Cloud Platform, and integrates the pest and disease control guidelines issued by the Guangxi Academy of Agricultural Sciences.

[0022] In some embodiments of the present invention, the field Internet of Things is built using the model of the Guangxi Sugarcane Smart Agriculture Demonstration Base to monitor the full-cycle growth of sugarcane fields and control water and fertilizer, and can provide soil moisture, meteorological data and the NDVI thermal map of drone inspections for the risk assessment and alarm module to call.

[0023] Preferably, in the multi-source remote sensing image preprocessing module, the C-band dual-polarization SAR data comes from the Sentinel-1 satellite, the multi-spectral L2A data comes from the Sentinel-2 satellite, the drone multi-spectral data comes from the DJI P4 Multispectral, and the airborne hyperspectral data comes from the Headwall Nano-Hyperspec imager; In some embodiments of the present invention, the C-band dual-polarization SAR data and the multi-spectral L2A data come from the Copernicus Open Access Hub.

[0024] Preferably, the correction includes radiometric correction and geometric correction; The radiometric correction is as follows: Use Gamma MAP filtering to eliminate the speckle noise of the C-band dual-polarization SAR data, and perform terrain radiometric correction based on the 30m DEM; The airborne hyperspectral data is calibrated by the whiteboard reflectance; Use the 6S model to perform atmospheric correction on the multi-spectral remote sensing image data, and use the SNAP tool to normalize the solar altitude angle; The geometric correction is as follows: All images are registered to the WGS84 UTM coordinate system through GCPs with an accuracy of <0.3 pixels. Range-Doppler terrain correction is additionally applied to the C-band dual-polarization SAR data. The centimeter-level positioning of the UAV multi-spectral and airborne hyperspectral data is achieved through PPK post-processing; The data fusion of the multi-source remote sensing image preprocessing module specifically refers to: Fusing the C-band dual-polarization SAR and multi-spectral L2A-level data through the non-subsampled contourlet transform (NSCT) to enhance the texture features of the sugarcane lodging area. The data of the standardized preprocessed image dataset are standardized GeoTIFF images.

[0025] The non-subsampled contourlet transform (NSCT) fusion combines the Sentinel-2 multi-spectral L2A-level optical image data (including vegetation indices such as NDVI) and the Sentinel-1 C-band dual-polarization SAR image data (VV / VH polarization texture) through multi-scale decomposition and feature enhancement to improve the detection accuracy of the sugarcane lodging area. The schematic diagram of the NSCT fusion process is as shown in the appendix Figure 2 as follows: Multi-scale alignment: Resample the two types of data uniformly to a resolution of 10m, and generate matching low-frequency (spectral / scattering intensity) and high-frequency sub-bands (texture / polarization ratio) through 3-layer NSCT decomposition. Figure 2 The left arrow in the figure indicates this step; Feature enhancement fusion: The low-frequency sub-band is fused with NDVI weighting, and the high-frequency sub-band strengthens the lodging texture through local energy matching. Figure 2 The right arrow in the figure indicates this step; Joint feature extraction: Extract the optical-SAR joint texture and temporal change features from the fusion result. After NSCT fusion, the texture of the lodging area, such as the fractures and fragmented structures caused by lodging, will be more obvious and easier to identify. In particular, the high-frequency texture information of SAR strengthens the structures that are difficult to show in the optical due to cloud occlusion or low-light conditions.

[0026] After the two images are transformed by NSCT, they are fused corresponding to the scale, and the fused result is used for the identification of the lodging area features.

[0027] Preferably, in the remote sensing parameter acquisition module, the parameter value of the leaf area index LAI is obtained by inverting based on the PROSAIL model, optimizing with the red-edge band of multi-spectral L2A data, with an inversion error <0.5 and R² = 0.88; the parameter value of the canopy water content NDWI is obtained by using the water-sensitive characteristic bands of 1450 nm / 1940 nm in airborne hyperspectral data, through the dual-band ratio method, and co-verifying with SMAP soil moisture data; the chlorophyll content is obtained by calculating the REIP red-edge inflection point position index of the sugarcane pest and disease sensitive bands; the SAR dielectric constant is obtained by inverting C-band dual-polarization SAR data; the NDRE index value is obtained by calculating the reflectance of the red-edge band and the near-infrared band in multi-spectral L2A data and airborne hyperspectral data.

[0028] In some embodiments of the present invention, the calculation of the red-edge inflection point index REIP is mainly realized through hyperspectral remote sensing data sources, including continuous spectral acquisition of the 680 - 750 nm red-edge band by an airborne hyperspectral imager (such as Headwall Nano-Hyperspec), and the red-edge band combination of B5 (705 nm), B6 (740 nm), and B7 (783 nm) in the multi-spectral L2A data of the Sentinel-2 satellite. Its core algorithm is based on locating the extreme points of the first derivative spectrum in the red-edge region, realizing the calculation of the wavelength-reflectance derivative through Python code, and locating the maximum point of the reflectance change rate within the range of 680 - 750 nm. When the calculation error exceeds 15%, cross-verification is required by combining the ground measured data of a portable hyperspectral spectrometer, the laboratory test results of chlorophyll, and the NDRE index.

[0029] Preferably, in the ground measured data corresponding to the parameter values output by the remote sensing parameter acquisition module, for the leaf area index LAI, it is measured by a leaf area meter; for the canopy water content NDWI and chlorophyll content, they are measured by a field portable spectrometer; for the SAR dielectric constant, it is measured by a portable soil dielectric sensor; the NDRE index value is obtained by measuring the canopy reflectance spectrum with a portable hyperspectral spectrometer and then calculated using the formula: NDRE = (NIR - Red Edge) / (NIR + Red Edge); the sugarcane planting agricultural situation data includes sugarcane plant height and stem diameter RTK measurement data, soil temperature and humidity in the sugarcane field, geospatial information, meteorological data, and field soil water holding capacity; the geospatial information includes DEM terrain data.

[0030] In some embodiments of the present invention, the sugarcane planting agricultural situation data comes from the Guangxi Agricultural Cloud Platform.

[0031] RTK, namely Real-Time Kinematic differential positioning technology, is a GNSS (Global Navigation Satellite System) positioning technology that can provide centimeter-level positioning accuracy in real time. RTK measurement data is high-precision geographical location information obtained through this technology. These data not only include basic information such as longitude, latitude, and altitude, but may also contain additional information such as speed, timestamp, and satellite signal quality. The Sugarcane Research Institute of Guangxi Academy of Agricultural Sciences established a field RTK reference station network, and combined fixed altimeters with mobile RTK devices to achieve dynamic monitoring of the stem diameter and plant height at the single-plant scale. In some embodiments of the present invention, the RTK measurement data of the sugarcane plant height and stem diameter is obtained from the field RTK reference station network established by the Sugarcane Research Institute of Guangxi Academy of Agricultural Sciences.

[0032] Preferably, in the ground measured data coupling module, the coupling process is as follows: Leaf Area Index LAI: Establish a regression relationship between the measured Leaf Area Index LAI and the Leaf Area Index LAI inverted from multi-spectral L2A level data. When R²>0.85, update the model coefficients to improve the inversion accuracy; Canopy water content NDWI: Take the NDWI calculated from the airborne hyperspectral band of 950 - 1700nm as the reference value. When the deviation from the measured value of the portable spectrometer >10%, perform the following operations in sequence: spatial consistency test of the Sentinel-1 SAR surface water index and the SMAP soil moisture product, spatial correlation analysis of the terrain humidity index TWI and NDWI, and cross-validation of the field soil water holding capacity sensor data. When the passing rate of the three validations ≥2 / 3, output the final NDWI value; In some embodiments of the present invention, the difference threshold of the spatial consistency test between the Sentinel-1 SAR surface water index and the SMAP soil moisture product ≤0.05), and the R² of the spatial correlation analysis between the terrain humidity index TWI and NDWI ≥0.7.

[0033] Chlorophyll content: Compare the calculated value of the REIP red edge inflection point index with the measured value. When the absolute error >15%, trigger the compensation mechanism; In some embodiments of the present invention, the compensation mechanism is to perform first derivative resampling on the airborne hyperspectral data, apply Savitzky-Golay filtering to smooth the reflectance curve, and the accuracy requirement of RMSE<0.8mg / cm² should be met after compensation.

[0034] SAR dielectric constant: Construct a regression model through the VV / VH polarization ratio and the measured value, correct the scattering mechanism for the dielectric constant mutation region using Freeman-Durden decomposition, and eliminate the multiple scattering interference by combining the sugarcane plant height RTK data; In some embodiments of the present invention, the regression model constructed by the VV / VH polarization ratio and the measured value has R²≥0.75, and the difference between adjacent pixels in the dielectric constant mutation region > 30%.

[0035] NDRE index: Calibrated stage by growth stage: Divide the datasets into seedling stage, tillering stage, jointing stage, and maturity stage; Test the spatial autocorrelation through Moran's I index within each growth stage; Use Kriging interpolation to supplement missing data; In some embodiments of the present invention, the Moran's I index I≥0.6, and the relative error of the Kriging interpolation ≤12%.

[0036] Obtain time-series data through the leaf area index LAI, and construct a reference curve for healthy sugarcane fields with the measured stem diameter data of RTK in the same period.

[0037] In some embodiments of the present invention, the process of constructing the reference curve for healthy sugarcane fields is as follows: Integrate the LAI time-series data (5-day resolution) of Sentinel-2 multispectral L2A level data and the RTK stem diameter measurement data (accuracy of ±0.5 cm), use K-means clustering to divide the growth stages and establish differential growth curve models such as polynomial regression and Gaussian process to achieve dynamic reference construction.

[0038] Preferably, in the abnormal growth monitoring model recognition module: In the abnormal growth monitoring model recognition module, the monitoring and recognition models for five types of abnormal growth of sugarcane are: pest and disease monitoring and recognition model, lodging monitoring and recognition model, slow growth monitoring and recognition model, land drought monitoring and recognition model, and flood monitoring and recognition model; The primary pest and disease probability distribution map is obtained by comparing with the healthy sugarcane field spectral library based on the Mahalanobis distance and detecting abnormal spectra, and its threshold is set at a 95% confidence interval; In some embodiments of the present invention, the healthy sugarcane field spectral library is obtained by sorting out the sugarcane field remote sensing monitoring record data in the whole-region agricultural machinery operation data of the Guangxi Agricultural Machinery Informatization Management Platform.

[0039] The pest and disease monitoring and recognition model spatially registers the primary pest and disease probability distribution map with the SAR dielectric constant, chlorophyll content, and the geographical information of the field boundary, resamples the satellite data to the field scale through bilinear interpolation to achieve multi-dimensional disease feature capture; uses spectral angle matching (SAM) to accurately identify the specific spectral fingerprints of sugarcane pests and diseases, and in the process of spectral angle matching, introduces the field ID attribute field to calculate the spectral similarity only for the pixels within the same geographical unit, excluding the interference of adjacent fields, combines random forest classification to integrate NDVI, red-edge index NDRE, and SAR texture features, and outputs a pest and disease heat map located in specific fields through a probability space model; In the lodging determination process of the lodging monitoring and recognition model, high-frequency texture and spectral features are first extracted from the NSCT fusion result. Then, the digital surface model is corrected in elevation using the RTK-measured plant height data to eliminate the deviation caused by insufficient vegetation coverage. Subsequently, the corrected plant height data is coupled with the SAR texture features, and the local variance threshold is dynamically adjusted through the stem diameter measurement value to enhance the recognition of the lodging-sensitive area. Then, the three-dimensional morphology method is combined to optimize the spatial connectivity of discrete scattering points, where the dilation kernel size is adaptively adjusted according to the median of the stem diameter. Finally, a lodging probability distribution map integrating multi-dimensional features is output. In some embodiments of the present invention, the three-dimensional morphology adaptive kernel algorithm dynamically adjusts the kernel size based on the RTK stem diameter median and the DSM plant height coefficient of variation, and the kernel duration constraint is ≤ 2 cm / day growth amount.

[0040] Preferably, in the slow growth monitoring and recognition model, the current leaf area index (LAI) time series curve is matched with the historical healthy sugarcane field benchmark curve through dynamic time warping (DTW), the per-pixel deviation rate is calculated, and random forest probability modeling is performed by combining the stem diameter RTK data and the NDRE index to output a slow growth probability distribution map. The land drought monitoring and recognition model fuses the canopy water content (NDWI) and the SAR dielectric constant to construct a water stress index (WSI), and generates a spatially continuous drought probability distribution map through Kriging interpolation. The flood monitoring and recognition model detects the water accumulation area based on the change of the SAR backscattering coefficient, superimposes the DEM terrain data to simulate the flood diffusion path, corrects the false detection by combining the field water holding capacity data, determines the water accumulation range, and finally outputs a water accumulation probability distribution map.

[0041] Preferably, in the risk assessment and alarm module, the dynamic abnormal growth grading threshold library is obtained by using sliding time window statistical analysis and machine learning algorithms, and the dynamic abnormal growth grading threshold library contains information on different growth periods and plots. The measured data used to obtain the final abnormal growth distribution probability is the data collected in real time by field Internet of Things devices, including: soil moisture, meteorological data, and the NDVI heat map of drone inspections; the meteorological data includes a 72-hour rainfall forecast.

[0042] In some embodiments of the present invention, the measured data comes from the Guangxi Agricultural Cloud Platform.

[0043] In some embodiments of the present invention, the process of obtaining the threshold library using sliding time window statistical analysis and machine learning algorithms is as follows: By defining a three-dimensional spatio-temporal data structure (time × field ID × monitoring index) with a 30-day benchmark window and a 5-day sliding step, perform moving average, standard deviation, and trend slope calculations for each field. Use the 25-75 percentile range as the health baseline, and dynamically adjust the threshold through a weighted formula (α = 0.7). Combine the LSTM network to predict threshold drift (input 30 days × 5 indicators, output threshold). When an anomaly is triggered, extend to a 60-day window for Granger causality test and SHAP value analysis. Finally, achieve real-time monitoring with double spatio-temporal sliding, 5-day online update, and growth stage perception (±15% at the seedling stage / ±8% at the maturity stage), and deploy it on a Jetson AGX Xavier edge device (delay < 3 seconds / km²).

[0044] Preferably, the alarm level is three levels, namely: Level 1 alarm, triggered when a single indicator exceeds the threshold; Level 2 alarm, triggered when there is a composite risk of multiple indicators; Level 3 alarm, triggered by sudden and urgent climate disasters.

[0045] Preferably, in the risk assessment and alarm module, the dynamic abnormal growth classification threshold library is continuously optimized through the sugarcane pest and disease expert knowledge base and real-time feedback.

[0046] In some embodiments of the present invention, the update frequency of the abnormal growth classification threshold library is the same as that of the sugarcane pest and disease expert knowledge base, which is quarterly update; the update frequency of the final abnormal growth distribution probability is the same as that of the remote sensing data, which is minute-level. In some embodiments of the present invention, the trigger condition of the abnormal growth classification threshold library is the growth stage switch of sugarcane.

[0047] In some embodiments of the present invention, the alarm levels are set as shown in the following table: In some embodiments of the present invention, when any two P-values exceed the action-level threshold and the correlation coefficient R of the two abnormal growths > 0.7, the alarm level is automatically upgraded.

[0048] In some embodiments of the present invention, the calculation formula of the correlation coefficient R is: Where: k is the environmental correction coefficient (take 1.2 on high-temperature days, take 0.8 in the rainy season), and the data window n = 15 (obtained from the best practice results in the Guangxi sugarcane area).

Claims

1. A sugarcane abnormal growth alarm system based on a remote sensing monitoring model, characterized in that: The abnormal growth refers to five types: diseases and pests of sugarcane, lodging, slow growth, drought in the land, and flood; the alarm system includes: Multi-source remote sensing image preprocessing module: Collect C-band dual-polarization SAR, multi-spectral L2A-level data, UAV multi-spectral, and airborne hyperspectral data of the sugarcane field, perform calibration and data fusion, and output a standardized preprocessed image data set; Remote sensing parameter acquisition module: Invert and calculate the above data set to obtain parameter values of leaf area index LAI, canopy water content NDWI, chlorophyll content, SAR dielectric constant, and NDRE; Ground measured data coupling module: Couple and calibrate the above parameter values with their corresponding ground measured data and sugarcane planting agricultural situation data, and output the calibrated parameter values; Abnormal growth monitoring and recognition model module: Combine the airborne hyperspectral data with the sugarcane diseases and pests sensitive spectral feature library to generate a primary diseases and pests probability distribution map, and combine its data with the calibrated parameter values and the sugarcane field geographical information to respectively construct monitoring and recognition models for the five abnormal growths of sugarcane, and respectively output abnormal growth distribution maps with preliminary probabilities; Risk assessment and alarm module: Integrate the historical abnormal growth data of the sugarcane field and the sugarcane diseases and pests expert knowledge base to construct a dynamic abnormal growth threshold library, and set alarm levels corresponding to the threshold library; Integrate the abnormal growth distribution map with preliminary probabilities and the measured data to obtain the final abnormal growth distribution probability, and compare it with the threshold library. When the alarm classification conditions are met, automatically trigger the corresponding level of alarm.

2. The sugarcane abnormal growth alarm system based on the remote sensing monitoring model according to claim 1, wherein In the multi-source remote sensing image preprocessing module, the C-band dual-polarization SAR data comes from the Sentinel-1 satellite, the multi-spectral L2A-level data comes from the Sentinel-2 satellite, the UAV multi-spectral data comes from the DJI P4 Multispectral, and the airborne hyperspectral data comes from the Headwall Nano-Hyperspec imager; The calibration includes radiometric calibration and geometric calibration; The radiometric calibration is as follows: Use Gamma MAP filtering to eliminate the speckle noise of the C-band dual-polarization SAR data, and perform terrain radiometric calibration based on the 30m DEM; The airborne hyperspectral data is calibrated by the whiteboard reflectance; Use the 6S model to perform atmospheric calibration on the multi-spectral remote sensing image data, and use the SNAP tool to normalize the solar altitude angle; The geometric calibration is as follows: All images are registered to the WGS84 UTM coordinate system through GCPs with an accuracy of <0.3 pixels. The C-band dual-polarization SAR data additionally applies Range-Doppler terrain correction. The UAV multi-spectral and airborne hyperspectral data achieve centimeter-level positioning through PPK post-processing; The data fusion of the multi-source remote sensing image preprocessing module is as follows: Fuse the C-band dual-polarization SAR and multi-spectral L2A-level data through the non-subsampled contourlet transform NSCT to enhance the texture features of the lodging area in the sugarcane field; The data of the standardized preprocessed image data set is a standardized GeoTIFF image.

3. The sugarcane abnormal growth alarm system based on the remote sensing monitoring model according to claim 2, characterized in that, In the remote sensing parameter acquisition module, the parameter value of the leaf area index LAI is obtained by inverting based on the PROSAIL model, combined with the optimization of the red edge band of the multi-spectral L2A level data, with an inversion error <0.5 and R² = 0.88; the parameter value of the canopy water content NDWI is obtained by using the water-sensitive characteristic band in the airborne hyperspectral data, through the dual-band ratio method, and co-verified with the SMAP soil moisture data; the chlorophyll content is calculated through the REIP red edge inflection point position index in the sugarcane pest and disease sensitive band; the SAR dielectric constant is obtained by inverting the C-band dual-polarization SAR data; the NDRE index value is calculated through the reflectance of the red edge band and the near-infrared band in the multi-spectral L2A level data and the airborne hyperspectral data.

4. The sugarcane abnormal growth alarm system based on the remote sensing monitoring model according to claim 3, characterized in that, In the ground measured data coupling module, among the ground measured data corresponding to the parameter values output by the remote sensing parameter acquisition module, the leaf area index LAI is measured by a leaf area meter, the canopy water content NDWI and the chlorophyll content are measured by a field portable spectrometer, and the SAR dielectric constant is measured by a portable soil dielectric sensor; the NDRE index value is calculated using a calculation formula after collecting the canopy reflectance spectrum through a portable hyperspectral spectrometer; the sugarcane planting agricultural situation data includes the RTK measurement data of the sugarcane plant height and stem thickness, the soil temperature and humidity of the sugarcane field, the geospatial information, the meteorological data, and the field soil water holding capacity; the geospatial information includes the DEM terrain data.

5. The sugarcane abnormal growth alarm system based on the remote sensing monitoring model according to claim 4, characterized in that, In the ground measured data coupling module, the coupling process is as follows: Leaf area index LAI: Establish a regression relationship between the measured leaf area index LAI and the leaf area index LAI inverted from the multi-spectral L2A level data. When R² > 0.85, update the model coefficients to improve the inversion accuracy; Canopy water content NDWI: Take the NDWI calculated from the 950 - 1700 nm band of the airborne hyperspectral data as the reference value. When the deviation from the measured value of the portable spectrometer > 10%, the following are executed in sequence: the spatial consistency test of the Sentinel-1 SAR surface water index and the SMAP soil moisture product, the spatial correlation analysis of the terrain humidity index TWI and NDWI, and the cross-validation of the field soil water holding capacity sensor data. When the passing rate of the three validations ≥ 2 / 3, output the final NDWI value; Chlorophyll content: Based on the comparison between the calculated value and the measured value of the REIP red edge inflection point index, trigger the compensation mechanism when the absolute error > 15%; SAR dielectric constant: Construct a regression model through the VV / VH polarization ratio and the measured value, correct the scattering mechanism for the dielectric constant mutation area using the Freeman-Durden decomposition, and eliminate the multiple scattering interference by combining the sugarcane plant height RTK data; NDRE index: Calibrate in stages according to the growth period, and divide the data sets into seedling stage, tillering stage, jointing stage, and maturity stage; test the spatial autocorrelation through the Moran's I index within each growth period; use Kriging interpolation to supplement the missing data; Obtain the time series data through the leaf area index LAI, and construct the historical healthy sugarcane field reference curve with the simultaneously measured stem thickness data of RTK.

6. The sugarcane abnormal growth alarm system based on the remote sensing monitoring model according to claim 5, characterized in that, In the abnormal growth monitoring model recognition module, the monitoring and recognition models for five types of abnormal growth of sugarcane are: pest and disease monitoring and recognition model, lodging monitoring and recognition model, slow growth monitoring and recognition model, land drought monitoring and recognition model, and flood monitoring and recognition model; The primary pest and disease probability distribution map is obtained by comparing the spectral library of healthy sugarcane fields based on the Mahalanobis distance and detecting abnormal spectra, and its threshold is set at a 95% confidence interval; The pest and disease monitoring and recognition model registers the primary pest and disease probability distribution map with the SAR dielectric constant, chlorophyll content, and the geographical information of the field boundary in space, resamples the satellite data to the field scale through bilinear interpolation, and realizes the capture of multi-dimensional disease characteristics; The Spectral Angle Matcher (SAM) is used to accurately identify the specific spectral fingerprints of sugarcane pests and diseases. During the spectral angle matching process, the field ID attribute field is introduced to calculate the spectral similarity only for the pixels within the same geographical unit, excluding the interference of adjacent fields. Combining the Random Forest Classification to integrate the NDVI, Normalized Difference Red Edge (NDRE) index, and SAR texture features, the pest and disease heat map located in specific fields is output through the probability space model.

7. The sugarcane abnormal growth alarm system based on the remote sensing monitoring model according to claim 6, characterized in that, In the abnormal growth monitoring model recognition module, the lodging monitoring and recognition model first extracts high-frequency texture and spectral features from the NSCT fusion result, then uses the RTK measured plant height data to correct the elevation of the digital surface model to eliminate the deviation caused by insufficient vegetation coverage; subsequently, the corrected plant height data is coupled with the SAR texture features, and the local variance threshold is dynamically adjusted through the stem diameter measurement value to enhance the recognition of the lodging-sensitive area; then, the three-dimensional morphological method is combined to optimize the spatial connectivity of discrete scatter points, where the dilation kernel size is adaptively adjusted according to the median of the stem diameter; finally, the lodging probability distribution map integrating multi-dimensional features is output; The slow growth monitoring and recognition model matches the current Leaf Area Index (LAI) time series curve with the historical healthy sugarcane field benchmark curve through Dynamic Time Warping (DTW), calculates the per-pixel deviation rate, and combines the stem diameter RTK data and NDRE index for random forest probability modeling to output the slow growth probability distribution map; The land drought monitoring and recognition model fuses the Normalized Difference Water Index (NDWI) of the canopy water content and the SAR dielectric constant to construct a Water Stress Index (WSI), and generates a spatially continuous drought probability distribution map through Kriging interpolation; The flood monitoring and recognition model detects the water accumulation area based on the change in the SAR backscattering coefficient, superimposes the Digital Elevation Model (DEM) terrain data to simulate the flood diffusion path, corrects the false detection by combining the field water holding capacity data, determines the water accumulation range, and finally outputs the water accumulation probability distribution map.

8. The sugarcane abnormal growth alarm system based on the remote sensing monitoring model according to claim 7, characterized in that, The dynamic abnormal growth grading threshold library in the risk assessment and alarm module is obtained by using the sliding time window statistical analysis and machine learning algorithms; The measured data used to obtain the final abnormal growth distribution probability are the data collected in real time by the field Internet of Things devices, including: soil moisture, meteorological data, and the NDVI heat map of the drone inspection; The meteorological data includes a 72-hour rainfall forecast.

9. The sugarcane abnormal growth alarm system based on the remote sensing monitoring model according to claim 8, characterized in that, The alarm level is level three, which are: level one alarm, triggered when a single indicator exceeds the threshold; level two alarm, triggered when there is a multi-indicator composite risk; level three alarm, triggered by a sudden and urgent climate disaster.

10. The sugarcane abnormal growth alarm system based on the remote sensing monitoring model according to claim 9, characterized in that, In the risk assessment and alarm module, the dynamic abnormal growth grading threshold library is continuously optimized through the sugarcane pest expert knowledge base and real-time feedback.

Images