A decision-making method and system for multi-source information fusion

Abstract

The present invention relates to the field of data processing technology, and in particular to a decision-making method and system for multi-source information fusion. The method comprises the following steps: obtaining hyperspectral image data, visible light image data and infrared image data collected by drones, and collecting ground environment monitoring data collected by ground sensors; performing image preprocessing on the hyperspectral image data, visible light image data and infrared image data to obtain a standardized multispectral data set, wherein the image preprocessing includes geometric correction, radiation correction and atmospheric correction; extracting color and texture features from the visible light image in the multispectral data set, and identifying the appearance features and growth density of crops to obtain morphological feature data. The present invention can more accurately identify abnormal crop growth and potential risks of pests and diseases through the integrated analysis of multi-source data, thereby assisting engineers in making irrigation and fertilization decisions.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a decision-making method and system for multi-source information fusion. Background Art

[0002] With the development of agricultural science and technology, modern technologies such as drones and sensors are increasingly used in agricultural production. These technologies can provide multi-source and multi-dimensional data, such as hyperspectral images, visible light images, infrared images, and ground environmental monitoring data. These data can be used to monitor crop growth, environmental changes, and potential pest and disease risks. However, due to the diversity and complexity of data sources, how to effectively integrate these data and generate meaningful decision-making recommendations has become an important challenge in agricultural informatization.

[0003] Traditional agriculture relies on farmers' experience and limited field data, which is inadequate in the face of large-scale, complex and ever-changing agricultural production environments. With the development of sensing technology, more and more data and information can be obtained in agricultural production, including high-resolution image data from drones and environmental monitoring data from ground sensors. However, a single data source cannot fully reflect the actual situation of crop growth, and the accuracy and scientificity of decision-making must be improved through the fusion of multi-source data. When making decisions based on multi-source information fusion, the following problems often exist: there are often clouds and shadows in remote sensing images, which will affect the extraction of image information features; the agricultural production process is affected by a variety of complex factors such as the natural environment and climatic conditions. There is a high degree of uncertainty and nonlinearity in many scenarios, which brings great challenges to mathematical modeling. Summary of the invention

[0004] Based on this, it is necessary for the present invention to provide a decision-making method and system for multi-source information fusion to solve at least one of the above technical problems.

[0005] To achieve the above purpose, a decision-making method for multi-source information fusion includes the following steps:

[0006] Step S1: Acquire hyperspectral image data, visible light image data and infrared image data collected by the UAV, and collect ground environment monitoring data collected by ground sensors; perform image preprocessing on the hyperspectral image data, visible light image data and infrared image data to obtain a standardized multispectral data set, wherein the image preprocessing includes geometric correction, radiation correction and atmospheric correction;

[0007] Step S2: extracting color and texture features from the visible light images in the multispectral data set, and identifying the appearance features and growth density of the crops, thereby obtaining morphological feature data; analyzing the thermal features of the crops from the infrared images in the multispectral data set, and identifying water stress and temperature anomalies, thereby obtaining thermal feature data;

[0008] Step S3: performing image integration on the morphological feature data and the thermal feature data according to the hyperspectral image in the multispectral data set, thereby generating an enhanced multispectral data set; performing calculation of the normalized vegetation index and the enhanced vegetation index according to the enhanced multispectral data set, thereby obtaining vegetation index data;

[0009] Step S4: Estimating crop physiological parameters based on spectral reflectance characteristics according to vegetation index data, thereby obtaining crop physiological parameter data; performing microclimate analysis using ground environment monitoring data, thereby generating crop growth environment characteristic data;

[0010] Step S5: performing spatiotemporal registration and correlation analysis on the enhanced multispectral data set, crop physiological parameter data, and growth environment characteristic data to generate a comprehensive characteristic data set; using a pre-trained pest and disease recognition model to identify abnormal growth and potential pest and disease risks of crops in the comprehensive characteristic data set, and generating decision recommendations based on preset expert knowledge, thereby obtaining decision recommendation data;

[0011] Step S6: Evaluate the irrigation amount based on humidity and crop water demand according to the decision recommendation data, and evaluate the nutrients required for crop growth, so as to generate an irrigation prescription map to assist engineers in making irrigation and fertilization decisions.

[0012] The present invention ensures the diversity and comprehensiveness of information sources through multi-type image data collected by drones and ground sensor monitoring data; the image preprocessing steps (geometric correction, radiation correction, atmospheric correction) ensure the accuracy and reliability of the data set, and provide high-quality basic data for subsequent analysis. The color and texture feature extraction of visible light images helps to identify the appearance characteristics and growth density of crops, and provides visual information for evaluating crop growth conditions; the thermal feature analysis of infrared images can reveal the moisture status and temperature anomalies of crops, and provide important information for detecting water stress and pests and diseases. The integration of hyperspectral images with morphological feature data and thermal feature data generates a richer enhanced multispectral data set; by calculating the vegetation index, the vegetation growth status can be quantitatively evaluated, providing key indicators for monitoring vegetation health. Using vegetation index data to estimate the physiological parameters of crops can provide a deep understanding of the physiological state and growth potential of crops; microclimate analysis provides a scientific basis for evaluating the crop growth environment and helps to identify the impact of environmental factors on crop growth. Spatiotemporal registration and association analysis ensure the consistency and comparability between different data sources and improve the comprehensive analysis capabilities of data sets; the generation of comprehensive feature data sets provides comprehensive data support for pest and disease identification and risk assessment; the application of pre-trained pest and disease identification models, combined with expert knowledge, enables rapid identification of abnormal crop growth and potential pest and disease risks, and provides scientific decision-making recommendations for agricultural production managers. Based on the assessment of irrigation volume based on humidity and crop water demand, as well as nutrient assessment, accurate irrigation prescription maps can be generated to guide engineers to perform precise irrigation and fertilization, thereby improving the utilization efficiency of water resources and fertilizers, reducing waste, and promoting healthy crop growth. Overall, this process uses high-tech means to achieve comprehensive monitoring and analysis of crop growth conditions, which helps to improve the level of intelligence in agricultural production, optimize resource utilization, reduce costs, increase yields, and ultimately achieve sustainable agricultural development.

[0013] The present invention also provides a decision system for multi-source information fusion, which is used to execute the above-mentioned decision method for multi-source information fusion. The decision system for multi-source information fusion comprises:

[0014] The data acquisition module is used to obtain the hyperspectral image data, visible light image data and infrared image data collected by the UAV, and collect the ground environment monitoring data collected by the ground sensor; perform image preprocessing on the hyperspectral image data, visible light image data and infrared image data to obtain a standardized multispectral data set, wherein the image preprocessing includes geometric correction, radiation correction and atmospheric correction;

[0015] The feature extraction module is used to extract color and texture features from visible light images in the multispectral data set, and identify the appearance features and growth density of crops to obtain morphological feature data; analyze the thermal features of crops from infrared images in the multispectral data set, and identify water stress and temperature anomalies to obtain thermal feature data;

[0016] The multi-source data fusion module is used to integrate the morphological feature data and thermal feature data according to the hyperspectral images in the multispectral data set, so as to generate an enhanced multispectral data set; the normalized vegetation index and the enhanced vegetation index are calculated according to the enhanced multispectral data set, so as to obtain vegetation index data;

[0017] The physiological parameter estimation module is used to estimate the crop physiological parameters based on the spectral reflectance characteristics according to the vegetation index data, so as to obtain the crop physiological parameter data; and to use the ground environment monitoring data to perform microclimate analysis, so as to generate the crop growth environment characteristic data;

[0018] The intelligent decision-making module is used to perform spatiotemporal registration and correlation analysis on the enhanced multispectral data set, crop physiological parameter data, and growth environment characteristic data to generate a comprehensive characteristic data set; the pre-trained pest and disease identification model is used to identify the growth anomalies and potential pest and disease risks of crops in the comprehensive characteristic data set, and decision recommendations are generated based on preset expert knowledge to obtain decision recommendation data;

[0019] The irrigation and fertilization decision-making assistance module is used to evaluate the irrigation amount based on humidity and crop water demand according to the decision recommendation data, and to evaluate the nutrients required for crop growth, so as to generate an irrigation prescription map to assist engineers in making irrigation and fertilization decisions.

[0020] The present invention collects multi-source data through unmanned aerial vehicles and ground sensors, and after image preprocessing steps such as geometric correction, radiation correction, and atmospheric correction, the standardization and quality of the data are ensured, providing accurate basic data for subsequent analysis. Feature extraction of visible light images and infrared images, including color, texture, thermal features, etc., helps to identify the appearance characteristics, growth density, water stress and temperature anomalies of crops, and provides important information for evaluating crop growth conditions. Integrate hyperspectral image data, morphological feature data and thermal feature data to generate enhanced multispectral data sets, and calculate vegetation indexes, providing a scientific basis for estimating crop physiological parameters. Combined with ground environmental monitoring data and meteorological data, microclimate analysis and pest and disease identification are carried out, which helps to timely discover crop growth abnormalities and potential pest and disease risks, and provides scientific decision support for agricultural production. Through spatiotemporal registration and association analysis, a comprehensive feature data set is generated, and a pest and disease identification model is used for risk identification, and decision-making recommendations are generated in combination with expert knowledge, which improves the intelligence and precision level of agricultural management. Based on the irrigation volume assessment and nutrient assessment of decision-making recommendation data, accurate irrigation prescription maps can be generated to guide engineers to carry out precise irrigation and fertilization, which helps to improve the utilization efficiency of water resources and fertilizers, reduce waste, and promote healthy growth of crops. Automated data collection and analysis reduce manpower requirements and improve the efficiency of monitoring and decision-making; through accurate data analysis, more accurate irrigation and fertilization can be achieved, reducing resource waste; pest and disease identification models can detect problems early and reduce crop losses; decision support systems based on comprehensive data provide scientific decision-making basis and improve the accuracy of decision-making; optimizing resource utilization and reducing environmental impact help achieve sustainable agricultural development. The entire system realizes comprehensive monitoring and analysis of crop growth conditions through high-tech means, which helps to improve the level of intelligence in agricultural production, optimize resource utilization, reduce costs, increase output, and ultimately achieve sustainable agricultural development. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments thereof made with reference to the following drawings:

[0022] Figure 1 It is a schematic diagram of the steps of the decision-making method of multi-source information fusion of the present invention;

[0023] Figure 2 for Figure 1 Detailed step flow diagram of step S1;

[0024] Figure 3 for Figure 1 Detailed step flow chart of step S2 in FIG. DETAILED DESCRIPTION

[0025] The technical method of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0026] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0027] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0028] To achieve this, please refer to Figures 1 to 3 The present invention provides a decision-making method for multi-source information fusion, the method comprising the following steps:

[0029] Step S1: Acquire hyperspectral image data, visible light image data and infrared image data collected by the UAV, and collect ground environment monitoring data collected by ground sensors; perform image preprocessing on the hyperspectral image data, visible light image data and infrared image data to obtain a standardized multispectral data set, wherein the image preprocessing includes geometric correction, radiation correction and atmospheric correction;

[0030] Step S2: extracting color and texture features from the visible light images in the multispectral data set, and identifying the appearance features and growth density of the crops, thereby obtaining morphological feature data; analyzing the thermal features of the crops from the infrared images in the multispectral data set, and identifying water stress and temperature anomalies, thereby obtaining thermal feature data;

[0031] Step S3: performing image integration on the morphological feature data and the thermal feature data according to the hyperspectral image in the multispectral data set, thereby generating an enhanced multispectral data set; performing calculation of the normalized vegetation index and the enhanced vegetation index according to the enhanced multispectral data set, thereby obtaining vegetation index data;

[0032] Step S4: Estimating crop physiological parameters based on spectral reflectance characteristics according to vegetation index data, thereby obtaining crop physiological parameter data; performing microclimate analysis using ground environment monitoring data, thereby generating crop growth environment characteristic data;

[0033] Step S5: performing spatiotemporal registration and correlation analysis on the enhanced multispectral data set, crop physiological parameter data, and growth environment characteristic data to generate a comprehensive characteristic data set; using a pre-trained pest and disease recognition model to identify abnormal growth and potential pest and disease risks of crops in the comprehensive characteristic data set, and generating decision recommendations based on preset expert knowledge, thereby obtaining decision recommendation data;

[0034] Step S6: Evaluate the irrigation amount based on humidity and crop water demand according to the decision recommendation data, and evaluate the nutrients required for crop growth, so as to generate an irrigation prescription map to assist engineers in making irrigation and fertilization decisions.

[0035] In the embodiment of the present invention, reference Figure 1 The above is a schematic flow chart of the steps of a decision-making method for multi-source information fusion of the present invention. In this example, the decision-making method for multi-source information fusion includes the following steps:

[0036] Step S1: Acquire hyperspectral image data, visible light image data and infrared image data collected by the UAV, and collect ground environment monitoring data collected by ground sensors; perform image preprocessing on the hyperspectral image data, visible light image data and infrared image data to obtain a standardized multispectral data set, wherein the image preprocessing includes geometric correction, radiation correction and atmospheric correction;

[0037] The embodiment of the present invention uses a drone equipped with a multispectral camera, a hyperspectral camera, and an infrared camera to obtain hyperspectral image data, visible light image data, and infrared image data of the target area respectively; deploys ground sensors in the farmland area to collect environmental monitoring data such as temperature, humidity, soil moisture, and wind speed. The collected image data is geometrically corrected to eliminate geometric distortion caused by changes in sensor position and viewing angle, so that the image data is spatially aligned; the radiation response of the sensor is corrected to ensure that the radiation intensity between different image data is comparable; the influence of atmospheric conditions on the image data, such as clouds, aerosols, etc., is eliminated, and correction is performed through an atmospheric correction algorithm (such as 6S or FLAASH) to obtain a standardized multispectral data set.

[0038] Step S2: extracting color and texture features from the visible light images in the multispectral data set, and identifying the appearance features and growth density of the crops, thereby obtaining morphological feature data; analyzing the thermal features of the crops from the infrared images in the multispectral data set, and identifying water stress and temperature anomalies, thereby obtaining thermal feature data;

[0039] The embodiment of the present invention extracts color features (such as RGB values, color histograms) and texture features (such as GLCM, LBP) from visible light images, and uses computer vision algorithms to identify the appearance features of crops and their growth density; the extracted color and texture features are analyzed to generate data describing the morphological characteristics of crops, including leaf size, plant height, and canopy density. Thermal feature analysis is performed on infrared images to identify thermal anomalies of crops, such as water stress and temperature anomalies; the thermal feature data obtained by analysis includes temperature distribution of crop leaves, water stress areas, etc.

[0040] Step S3: performing image integration on the morphological feature data and the thermal feature data according to the hyperspectral image in the multispectral data set, thereby generating an enhanced multispectral data set; performing calculation of the normalized vegetation index and the enhanced vegetation index according to the enhanced multispectral data set, thereby obtaining vegetation index data;

[0041] The embodiment of the present invention utilizes the spectral information of the hyperspectral image to integrate the morphological feature data and the thermal feature data obtained in step S2 to generate an enhanced multispectral data set, and generates a normalized vegetation index by calculating the reflectivity difference between the visible light and near-infrared bands; based on NDVI, a more accurate vegetation index is calculated by considering factors such as atmospheric influence and background noise to obtain vegetation index data.

[0042] Step S4: Estimating crop physiological parameters based on spectral reflectance characteristics according to vegetation index data, thereby obtaining crop physiological parameter data; performing microclimate analysis using ground environment monitoring data, thereby generating crop growth environment characteristic data;

[0043] The embodiment of the present invention analyzes the spectral reflectance characteristics of crops based on vegetation index data, estimates their physiological parameters, such as chlorophyll content, nitrogen content, photosynthetic efficiency, etc., and generates crop physiological parameter data; uses the collected environmental monitoring data to perform microclimate analysis on the farmland area to generate characteristic data describing the growth environment of crops, including soil moisture, temperature changes, light intensity, etc.

[0044] Step S5: performing spatiotemporal registration and correlation analysis on the enhanced multispectral data set, crop physiological parameter data, and growth environment characteristic data to generate a comprehensive characteristic data set; using a pre-trained pest and disease recognition model to identify abnormal growth and potential pest and disease risks of crops in the comprehensive characteristic data set, and generating decision recommendations based on preset expert knowledge, thereby obtaining decision recommendation data;

[0045] The embodiment of the present invention performs spatiotemporal registration of enhanced multispectral data sets, crop physiological parameter data, and growth environment characteristic data to ensure the consistency of different data sources in space and time; the registered data is comprehensively analyzed to generate a comprehensive characteristic data set, which contains information such as the growth status of crops and environmental influencing factors. The comprehensive characteristic data set is analyzed using a pre-trained pest and disease recognition model to identify abnormal growth of crops and potential pest and disease risks; based on the analysis results and preset expert knowledge, decision-making recommendation data for crop management is generated, including pest and disease control measures, irrigation adjustments, fertilization strategies, etc.

[0046] Step S6: Evaluate the irrigation amount based on humidity and crop water demand according to the decision recommendation data, and evaluate the nutrients required for crop growth, so as to generate an irrigation prescription map to assist engineers in making irrigation and fertilization decisions.

[0047] The embodiment of the present invention evaluates the irrigation amount based on the decision recommendation data, combined with information such as humidity and crop water demand, and determines the amount of irrigation water required for each area. ) and reference evapotranspiration ( ), calculate the actual water requirement of each crop. The formula is: ; For example, for a certain type of wheat, is 1.1, is 5mm / day, then the water requirement of the wheat is 5.5mm / day. According to the soil moisture data, the soil water content (SWC) monitored in real time by the sensor is compared with the field water holding capacity (FieldCapacity) required by the crop to determine the frequency and amount of irrigation. For example, if the SWC is 30% and the field water holding capacity should be 50%, the irrigation system needs to add enough water to make the soil moisture reach 50%. The decision-making recommendation data includes crop physiological parameter data (such as chlorophyll content, soil nutrient analysis results, etc.), as well as environmental monitoring data; based on the current concentration of key nutrients such as nitrogen, phosphorus, and potassium in the soil and the nutrient requirements of the crop, the amount of fertilizer is calculated; the nutrient demand difference (NutrientDemandDeficit) is used to determine the type and amount of nutrients that need to be supplemented. Example: If the soil analysis shows that the nitrogen content is 30mg / kg, and the crop growth requires 50mg / kg of nitrogen, nitrogen-containing fertilizers need to be applied to make up for the 20mg / kg gap. The chlorophyll index (such as NDVI) generated by hyperspectral images is used to assess the nitrogen demand of crops, and the amount of fertilizer is determined through correlation analysis. For example, the NDVI value of a certain farmland is low, indicating that the chlorophyll content is insufficient, and the system recommends applying nitrogen fertilizer to promote crop growth. Based on GIS tools, the results of irrigation assessment and fertilization demand assessment are interpolated in geographic space to generate refined irrigation and fertilization prescription maps (PrescriptionMap). These prescription maps show the specific irrigation amount and fertilizer types and quantities required for different regions; for example, for a certain field, it is recommended to reduce the irrigation amount in the eastern region due to poor soil drainage, while it is recommended to increase the irrigation amount in the western region due to the drier soil, and different fertilizer formulas are used in different regions. The irrigation prescription map is output in a standardized format (such as GeoTIFF or Shapefile format) and automatically executed by agricultural equipment (such as precision irrigation system or fertilization system); for example, the irrigation prescription map shows the irrigation amount for each area, such as the northern area needs to be irrigated with 10mm water, the southern area needs to be irrigated with 15mm water, and 20kg / ha of nitrogen fertilizer and 10kg / ha of phosphate fertilizer are applied at the same time.

[0048] The present invention ensures the diversity and comprehensiveness of information sources through multi-type image data collected by drones and ground sensor monitoring data; the image preprocessing steps (geometric correction, radiation correction, atmospheric correction) ensure the accuracy and reliability of the data set, and provide high-quality basic data for subsequent analysis. The color and texture feature extraction of visible light images helps to identify the appearance characteristics and growth density of crops, and provides visual information for evaluating crop growth conditions; the thermal feature analysis of infrared images can reveal the moisture status and temperature anomalies of crops, and provide important information for detecting water stress and pests and diseases. The integration of hyperspectral images with morphological feature data and thermal feature data generates a richer enhanced multispectral data set; by calculating the vegetation index, the vegetation growth status can be quantitatively evaluated, providing key indicators for monitoring vegetation health. Using vegetation index data to estimate the physiological parameters of crops can provide a deep understanding of the physiological state and growth potential of crops; microclimate analysis provides a scientific basis for evaluating the crop growth environment and helps to identify the impact of environmental factors on crop growth. Spatiotemporal registration and association analysis ensure the consistency and comparability between different data sources and improve the comprehensive analysis capabilities of data sets; the generation of comprehensive feature data sets provides comprehensive data support for pest and disease identification and risk assessment; the application of pre-trained pest and disease identification models, combined with expert knowledge, enables rapid identification of abnormal crop growth and potential pest and disease risks, and provides scientific decision-making recommendations for agricultural production managers. Based on the assessment of irrigation volume based on humidity and crop water demand, as well as nutrient assessment, accurate irrigation prescription maps can be generated to guide engineers to perform precise irrigation and fertilization, thereby improving the utilization efficiency of water resources and fertilizers, reducing waste, and promoting healthy crop growth. Overall, this process uses high-tech means to achieve comprehensive monitoring and analysis of crop growth conditions, which helps to improve the level of intelligence in agricultural production, optimize resource utilization, reduce costs, increase yields, and ultimately achieve sustainable agricultural development.

[0049] Preferably, step S1 comprises the following steps:

[0050] Step S11: using a drone equipped with a multispectral sensor to collect hyperspectral image data, visible light image data, and infrared image data;

[0051] Step S12: collecting temperature, humidity, soil moisture and light intensity information through a ground sensor network deployed on the target farmland, thereby obtaining ground environment monitoring data;

[0052] Step S13: performing integrity detection on the hyperspectral image data, the visible light image data and the infrared image data, and removing or marking areas where the degree of influence of clouds and shadows exceeds a predetermined threshold, thereby obtaining hyperspectral image marking data, visible light image marking data and infrared image marking data;

[0053] Step S14: performing image preprocessing on the hyperspectral image labeling data, the visible light image labeling data and the infrared image labeling data, so as to obtain a standardized multispectral data set, wherein the image preprocessing includes geometric correction, radiation correction and atmospheric correction.

[0054] As an embodiment of the present invention, refer to Figure 2 As shown, Figure 1 Detailed step flow diagram of step S1 in the embodiment of the present invention, step S1 includes the following steps:

[0055] Step S11: using a drone equipped with a multispectral sensor to collect hyperspectral image data, visible light image data, and infrared image data;

[0056] The drone of the embodiment of the present invention is equipped with multispectral, hyperspectral and infrared sensors. Before the flight mission, the drone and the sensor are calibrated to ensure the accuracy of the image data; the flight path of the drone is planned according to the terrain and crop distribution of the target farmland to ensure that the entire area is covered to avoid repeated scanning or missed scanning; the drone flies along the planned path to obtain hyperspectral image data, visible light image data and infrared image data respectively. In order to ensure the high quality of the data, the data needs to be collected under good weather conditions to avoid the influence of strong wind, rain and fog.

[0057] Step S12: collecting temperature, humidity, soil moisture and light intensity information through a ground sensor network deployed on the target farmland, thereby obtaining ground environment monitoring data;

[0058] The embodiment of the present invention deploys various types of sensors (such as temperature sensors, humidity sensors, soil moisture sensors, and light intensity sensors) on the target farmland, and arranges them in a dense grid form or according to the characteristics of the farmland to ensure comprehensive data coverage. The sensor network collects environmental data such as temperature, humidity, soil moisture, and light intensity in real time, and transmits the data to the central control system or data storage server through the wireless communication module; the data collected by the sensor is preliminarily verified to eliminate abnormal values ​​or noise interference to ensure the accuracy of the data.

[0059] Step S13: performing integrity detection on the hyperspectral image data, the visible light image data and the infrared image data, and removing or marking areas where the degree of influence of clouds and shadows exceeds a predetermined threshold, thereby obtaining hyperspectral image marking data, visible light image marking data and infrared image marking data;

[0060] The embodiment of the present invention uses an image analysis algorithm to detect cloud-covered areas and shadow parts in hyperspectral images, visible light images, and infrared images, and identifies areas affected by clouds or shadows by analyzing the spectral characteristics of the images; according to a predetermined standard, a threshold for cloud and shadow influence is set, and if the blocked area exceeds 20% of the total image area, it is marked as unqualified data. For the detected areas affected by clouds and shadows exceeding the threshold, they are marked or eliminated to generate hyperspectral image marking data, visible light image marking data, and infrared image marking data; the marking information is recorded in the image data file so that the affected areas can be identified in subsequent analysis.

[0061] Step S14: performing image preprocessing on the hyperspectral image labeling data, the visible light image labeling data and the infrared image labeling data, so as to obtain a standardized multispectral data set, wherein the image preprocessing includes geometric correction, radiation correction and atmospheric correction.

[0062] The embodiment of the present invention geometrically corrects the image through geographic reference points (such as landmarks, road intersections, etc.) to eliminate geometric distortion caused by changes in sensor angles and flight altitudes, ensuring that the image is consistent with the geographic coordinate system; spatially aligns the data of different sensors to ensure accurate alignment between multispectral images, infrared images, and hyperspectral images. For sensors of different bands, correct their radiation response curves to eliminate differences between sensors and ensure the consistency of image data in the spectral dimension; standardize the image so that its radiation value conforms to a standardized scale for subsequent analysis. Apply an atmospheric correction model (such as 6S or FLAASH) to correct the effects of atmospheric scattering, absorption, and other factors on image data, especially hyperspectral images, to ensure the accuracy of reflectance data; after completing geometric correction, radiation correction, and atmospheric correction, generate a standardized multispectral data set, including hyperspectral, visible light, and infrared data, and provide a reliable data basis for subsequent analysis.

[0063] The present invention uses a drone equipped with a multispectral sensor to collect hyperspectral image data, visible light image data and infrared image data. This multi-source data collection method can provide more comprehensive surface information, including vegetation index, land use, etc., which helps to improve the accuracy and reliability of analysis. The temperature, humidity, soil moisture and light intensity information are collected in real time through the ground sensor network, which can timely reflect the microclimate changes in the crop growth environment and provide real-time data support for the precise management of crops. The image data is checked for integrity, and the areas affected by clouds and shadows are removed or marked to ensure the quality of the data set, avoid the interference of these factors on the subsequent analysis results, and improve the accuracy of data processing. Image preprocessing includes geometric correction, radiation correction and atmospheric correction. These steps can eliminate sensor errors and atmospheric effects, standardize multispectral data sets, and provide accurate basic data for subsequent feature extraction and analysis. Through geometric correction, the spatial accuracy of image data is ensured, the image is aligned with the actual geographic coordinate system, and the availability of data is improved. Radiation correction and atmospheric correction help to eliminate the effects caused by sensor differences and changes in atmospheric conditions, so that the biophysical parameters extracted from the image are more accurate, such as vegetation coverage, biomass, etc. The preprocessed standardized multispectral data set provides high-quality input for the subsequent application of image classification, feature extraction, quantitative evaluation, and machine learning methods, thereby improving the effectiveness of the entire remote sensing monitoring system. Through these steps, the efficiency and accuracy of remote sensing monitoring can be significantly improved, providing strong data support and decision-making basis for multiple fields such as agriculture and environmental monitoring.

[0064] Preferably, step S13 comprises the following steps:

[0065] Step S131: performing pixel-level integrity scanning on the hyperspectral image data, visible light image data, and infrared image data to identify pixels with missing or damaged data, thereby obtaining incomplete pixel data;

[0066] The embodiment of the present invention scans each pixel of the hyperspectral image data, visible light image data, and infrared image data to detect whether there are pixels with missing or damaged data. For example, incomplete pixels can be identified by detecting abnormal values, noise, or invalid pixel flags. The identified pixels with missing or damaged data are marked to generate incomplete pixel data for subsequent processing.

[0067] Step S132: using the shadow detection model and the cloud detection model to identify the cloud layer and the shadow area of ​​the hyperspectral image data, the visible light image data and the infrared image data, thereby obtaining the cloud layer and the shadow area data;

[0068] The embodiment of the present invention uses a pre-trained shadow detection model to analyze image data and identify areas covered by shadows in the image. The shadow detection model usually relies on light intensity, color change and texture features to determine the shadow area. The identified shadow area is recorded as shadow area data and marked in the corresponding image file. The cloud detection model is used to analyze the image and identify areas obscured by clouds; cloud detection usually relies on spectral feature analysis, such as specific band combinations in hyperspectral data, to accurately locate areas covered by clouds; the data of the cloud area is recorded as part of the cloud and shadow area data.

[0069] Step S133: Calculate the ratio of cloud and shadow coverage areas in each image based on the cloud and shadow area data, and mark the images whose coverage ratio exceeds a predetermined threshold as low-quality data;

[0070] The embodiment of the present invention calculates the ratio of the cloud and shadow coverage area in each image based on the cloud and shadow area data generated in step S132; for example, the coverage ratio is obtained by calculating the number of pixels in the cloud and shadow areas and comparing them with the total number of pixels in the image. The calculated coverage ratio is compared with a preset threshold, and images with a coverage ratio exceeding the preset threshold (such as 10%) are marked as low-quality data and recorded in the metadata.

[0071] Step S134: For low-quality data, if the coverage ratio exceeds a preset critical threshold, the image is removed from the data set; if the coverage ratio is between the preset threshold and the critical threshold, a quality mark is added to the metadata to obtain marked image data;

[0072] For low-quality data, if the cloud and shadow coverage ratio exceeds a preset critical threshold (such as 50%), the image is directly removed from the data set to avoid adverse effects on subsequent analysis. For images with a coverage ratio between the preset threshold and the critical threshold (such as 10% to 50%), a quality tag is added to the metadata of the image to record the cloud or shadow impact of the image, reminding the subsequent analysis to consider its potential impact.

[0073] Step S135: using an image restoration algorithm to fill in missing or damaged pixels, small cloud areas and shadow areas according to the incomplete pixel data and the marked image data, thereby obtaining hyperspectral image marked data, visible light image marked data and infrared image marked data.

[0074] The embodiment of the present invention uses an image restoration algorithm (such as an interpolation-based method, a deep learning-based image restoration model, etc.) to fill in missing or damaged pixels based on incomplete pixel data. The restoration algorithm reconstructs the data of the missing area based on the characteristics and patterns of the surrounding pixels. For areas with a small cloud and shadow coverage ratio (such as less than a preset secondary threshold, such as 10%), the image restoration algorithm is used to fill these small cloud and shadow areas to preserve the integrity of the image as much as possible. After restoration, updated hyperspectral image marker data, visible light image marker data, and infrared image marker data are obtained, which will be used for subsequent analysis and processing.

[0075] The present invention can identify and record pixel points with missing or damaged data by performing pixel-level integrity scanning on hyperspectral image data, visible light image data, and infrared image data, and provide necessary information for subsequent data repair, thereby ensuring the integrity of the data set and the accuracy of the analysis. Analyzing image data using shadow detection models and cloud detection models helps to distinguish and identify cloud and shadow areas, which may interfere with data analysis. Accurately identifying these areas is crucial to improving the accuracy of subsequent data processing. By calculating the ratio of cloud and shadow coverage areas, high-quality and low-quality data can be effectively distinguished. Low-quality data that exceeds a predetermined threshold will be marked, which helps to improve the reliability of the final analysis results. For images with a coverage ratio exceeding a critical threshold, they are removed from the data set to avoid the impact of low-quality data on the analysis results; for data with a moderate coverage ratio, by adding quality tags in the metadata, a reference basis can be provided for further processing and analysis of the data. Using image restoration algorithms to fill in missing or damaged pixels and small areas of cloud and shadow areas not only restores the integrity of the data, but also improves the availability of the data, so that the processed hyperspectral image labeling data, visible light image labeling data, and infrared image labeling data can better reflect the actual surface characteristics. Through these steps, the quality of remote sensing image data can be significantly improved, providing more accurate and reliable data support for a variety of applications such as crop growth monitoring and environmental assessment.

[0076] Preferably, step S14 comprises the following steps:

[0077] Step S141: Correcting the geometric distortion of the hyperspectral image marker data due to the terrain and the shooting angle, converting the digital values ​​of the hyperspectral image marker data into reflectance and radiance values, thereby obtaining the reflectance or radiance data of the hyperspectral image after radiation correction;

[0078] The embodiment of the present invention uses geographic reference data (such as digital elevation model DEM) and drone flight records to correct geometric distortion caused by terrain undulations and photography angles; uses image registration technology (such as landmark-based registration) to perform geometric correction on images to ensure accurate spatial positioning of images; uses linear transformation, affine transformation or more complex nonlinear transformation methods to adjust the geometric distortion of images based on terrain information. Convert the digital value (DN value) in the image to the actual radiation value or reflectivity value. This process includes:

[0079] Radiation calibration: Use the sensor's calibration coefficient to convert the digital value into a radiance value (Radiance). The conversion formula is usually:

[0080]

[0081] Among them, Slope and Intercept are sensor-specific calibration coefficients.

[0082] Reflectivity calculation: The radiance value is further converted into a reflectivity value. The formula is usually:

[0083]

[0084] in, is the solar radiation value, is the solar altitude angle.

[0085] Obtain radiation-corrected data: After completing the above conversion, the reflectance or radiance data of the radiation-corrected hyperspectral image is obtained.

[0086] Step S142: Eliminate the influence of atmospheric scattering and absorption according to the reflectivity or radiance data, so as to obtain surface reflectivity data;

[0087] The embodiment of the present invention uses an atmospheric correction model (such as 6S or FLAASH) to correct the atmospheric scattering and absorption effects in the hyperspectral data. This usually includes: the model calculates the influence of atmospheric scattering on the image and the influence of atmospheric absorption on the image according to the band characteristics, atmospheric conditions and observation angle of the image; after correction, the real reflectivity data of the surface is obtained, and the influence of atmospheric interference is removed.

[0088] Step S143: performing color balance adjustment based on the actual object color on the visible light image mark data, and performing contrast enhancement processing, so as to obtain visible light contrast enhancement data;

[0089] The embodiment of the present invention uses a color balance algorithm (such as the gray world hypothesis or white balance algorithm) to adjust the color of the visible light image so that the color of the image is closer to the real situation; according to the color distribution of the image, the color deviation of the RGB channel is corrected to ensure color consistency. Contrast enhancement technology, such as histogram equalization or CLAHE (contrast limited adaptive histogram equalization), is applied to improve the contrast of the image and make the details clearer; the brightness and contrast of the image are adjusted to improve the visual effect and information readability of the image.

[0090] Step S144: converting the infrared signal in the infrared image tag data into an actual temperature value, and performing emissivity correction based on the emissivity difference of different ground objects, so as to obtain accurate temperature distribution data;

[0091] The embodiment of the present invention converts the infrared signal in the infrared image into an actual temperature value, usually using the radiation transfer equation for conversion: ;

[0092] where is the infrared radiation brightness, and are sensor-specific calibration coefficients.

[0093] Correction is performed based on the difference in emissivity of different objects to ensure the accuracy of temperature data, and the image data is adjusted using the actual emissivity value:

[0094]

[0095] in, It is the emissivity of the ground object. After completing the emissivity correction, accurate temperature distribution data can be obtained.

[0096] Step S145: aligning the surface reflectance data, visible light ratio enhancement data and precise temperature distribution data based on ground space and time series, thereby obtaining a standardized multispectral data set.

[0097] The embodiment of the present invention spatially registers the surface reflectance data, visible light ratio enhancement data, and precise temperature distribution data to ensure that they are aligned in the same geographic coordinate system. This usually involves image registration algorithms, such as image registration or resampling techniques; if the data set includes images from different times, ensure that the time series data is aligned in chronological order to facilitate time series analysis. The spatially and temporally aligned data sets are merged to generate a standardized multispectral data set, including surface reflectance data of hyperspectral images, ratio enhancement data of visible light images, and precise temperature data of infrared images.

[0098] The present invention eliminates the influence of terrain undulation and photography angle on the image through geometric distortion correction, and ensures the accuracy of image data in geographic space. This is crucial for application scenarios that require accurate geographic location information (such as map making, object positioning, etc.); radiation correction that converts digital values ​​into reflectance and radiance values ​​enables image data acquired by different sensors at different times to be effectively compared and analyzed because they now have a unified measurement standard. Atmospheric scattering and absorption change the propagation path and intensity of light, thereby affecting the quality and spectral characteristics of the image. By eliminating these effects, reflectance data that is closer to the actual surface conditions can be obtained, which improves the accuracy of remote sensing data in applications such as vegetation index calculation and land cover classification. Color balance adjustment of visible light images ensures the naturalness and consistency of colors in the image, which is very important for intuitive interpretation and professional analysis of the image. Contrast enhancement processing improves the visual distinguishability between different object features in the image, helps to highlight important landforms or vegetation features, and lays a good foundation for further image interpretation and feature extraction. Converting infrared signals into actual temperature values ​​provides direct data support for analyzing surface temperature distribution, which is crucial for applications such as heat distribution research and urban heat island effect analysis. Emissivity correction takes into account the differences in thermal radiation characteristics of different objects, improves the accuracy of temperature measurement, and ensures the comparability of temperature measurement results of different objects. By aligning the surface reflectance data, visible light ratio enhancement data, and precise temperature distribution data in space and time series, the consistency and comparability between multi-phase and multi-sensor data sets are ensured, which makes multi-dimensional and multi-angle analysis possible. Standardized multispectral data sets provide high-quality basic data for subsequent data processing, feature extraction, and information synthesis, which helps to improve the efficiency and accuracy of remote sensing monitoring and evaluation. Overall, these steps are aimed at improving the quality, accuracy, and availability of remote sensing data, and providing solid and reliable data support for various remote sensing applications.

[0099] Preferably, step S2 comprises the following steps:

[0100] Step S21: performing color space conversion on the visible light image in the multispectral data set, from RGB space to HSV space, and extracting hue, saturation and brightness features, thereby obtaining color features of the visible light image;

[0101] Step S22: extracting texture features of visible light images in the multispectral data set using a gray level co-occurrence matrix method, and calculating contrast, entropy, energy, and homogeneity to obtain texture statistics;

[0102] Step S23: performing crop appearance feature recognition based on deep learning according to the color features of the visible light image and the texture statistics, thereby obtaining appearance feature data, wherein the crop appearance feature recognition includes leaf morphology and plant type structure;

[0103] Step S24: performing crop growth density analysis on the visible light image in the multispectral data set by using image segmentation and density estimation algorithms, thereby obtaining a density distribution map;

[0104] Step S25: fusing the appearance feature data and the density distribution map to obtain morphological feature data;

[0105] Step S26: Analyze the thermal characteristics of crops on the infrared images in the multispectral data set, and identify water stress and temperature anomalies, so as to obtain thermal characteristic data.

[0106] As an embodiment of the present invention, refer to Figure 3 As shown, Figure 1 Detailed step flow diagram of step S2 in the embodiment of the present invention, step S2 includes the following steps:

[0107] Step S21: performing color space conversion on the visible light image in the multispectral data set, from RGB space to HSV space, and extracting hue, saturation and brightness features, thereby obtaining color features of the visible light image;

[0108] The embodiment of the present invention converts the visible light image from the RGB (red, green, blue) color space to the HSV (hue, saturation, brightness) color space. Hue indicates the type of color, and different color types can be identified through hue features; saturation indicates the purity or intensity of the color, and the higher the saturation, the purer the color; brightness indicates the brightness of the color, and the higher the brightness, the brighter the image; after extracting these features, the color information of each pixel can be obtained for subsequent analysis.

[0109] Step S22: extracting texture features of visible light images in the multispectral data set using a gray level co-occurrence matrix method, and calculating contrast, entropy, energy, and homogeneity to obtain texture statistics;

[0110] The embodiment of the present invention calculates the gray value frequency of each pair of pixels in the visible light image, generates a gray level co-occurrence matrix by a sliding window method, and common directions include 0°, 45°, 90°, and 135°. The texture statistics for measuring the contrast of the texture, measuring the complexity of the texture, reflecting the uniformity of the texture, and measuring the smoothness of the texture are calculated. The texture feature statistics are extracted using the above calculation method to obtain the texture features of the visible light image.

[0111] Step S23: performing crop appearance feature recognition based on deep learning according to the color features of the visible light image and the texture statistics, thereby obtaining appearance feature data, wherein the crop appearance feature recognition includes leaf morphology and plant type structure;

[0112] The embodiment of the present invention uses labeled crop image data to train a deep learning model (such as a convolutional neural network (CNN)) to identify the appearance characteristics of crops, including leaf morphology and plant structure. Label data is used during training to optimize the accuracy of the model; the trained deep learning model is used to perform feature recognition on visible light images to obtain appearance feature data of the leaf morphology and plant structure of the crops. Based on the output of the model, the appearance feature data of each crop is extracted, including the shape, size, and distribution of the leaves and the structural characteristics of the plant type.

[0113] Step S24: performing crop growth density analysis on the visible light image in the multispectral data set by using image segmentation and density estimation algorithms, thereby obtaining a density distribution map;

[0114] The embodiment of the present invention uses an image segmentation algorithm (such as threshold segmentation, K-means clustering, deep learning segmentation model, etc.) to segment the crop area in the visible light image; in the segmented image, the growth density of the crop is calculated according to the number of pixels in the segmented area, and a density estimation algorithm (such as an estimation method based on point density) is used to generate a density distribution map to represent the crop density in different areas. The density calculation result is visualized as a density distribution map to show the crop density in different areas.

[0115] Step S25: fusing the appearance feature data and the density distribution map to obtain morphological feature data;

[0116] The embodiment of the present invention fuses appearance feature data (such as leaf morphology and plant type structure) with the density distribution map. A variety of fusion techniques, such as weighted average and principal component analysis (PCA), can be used to integrate different types of data into a unified feature data set. Through data fusion, comprehensive morphological feature data is obtained, which combines the appearance characteristics and growth density information of crops.

[0117] Step S26: Analyze the thermal characteristics of crops on the infrared images in the multispectral data set, and identify water stress and temperature anomalies, so as to obtain thermal characteristic data.

[0118] The embodiment of the present invention performs thermal feature analysis on infrared images to identify temperature anomalies and water stress of crops. Thermal feature extraction algorithms (such as temperature difference detection, thermal anomaly detection, etc.) can be used to analyze temperature changes and water conditions in images. The thermal features of crops in infrared images are analyzed to identify water stress areas. By comparing normal and abnormal temperature patterns, it is determined whether water stress exists; abnormal temperature areas of crops are identified to detect whether there are abnormal high or low temperatures, which may indicate diseases or environmental problems. After the analysis results are obtained, thermal feature data is generated to represent the thermal features of crops and their possible abnormal conditions.

[0119] The present invention converts visible light images from RGB color space to HSV space, which can analyze and process color information more intuitively, because HSV color space separates the three basic attributes of color (hue, saturation, brightness), which is convenient for subsequent color feature extraction. The texture features can be extracted by using the grayscale co-occurrence matrix method, which can quantify the roughness of the image, the uniformity of the texture, etc., which is of great significance for the analysis and classification of the surface texture of crops. Based on deep learning technology, the appearance feature recognition of crops can automatically identify features such as leaf morphology and plant type structure, which is crucial for applications such as pest detection and variety identification. The crop growth density is analyzed by image segmentation and density estimation algorithm, and the obtained density distribution map provides important information for evaluating crop growth status and field management. The appearance feature data and the density distribution map are fused to obtain morphological feature data. This fusion can provide more comprehensive crop feature information and help improve the accuracy of crop monitoring and analysis. Thermal feature analysis of infrared images can identify water stress and temperature anomalies of crops, which is of great value for timely understanding of crop growth status and guiding irrigation and other agricultural activities. Through these steps, the growth conditions of crops can be comprehensively analyzed and evaluated, providing a scientific basis for agricultural production and improving the efficiency and quality of agricultural production.

[0120] Preferably, step S26 comprises the following steps:

[0121] Step S261: performing temperature calibration processing on the infrared image in the multispectral data set to obtain temperature calibration data;

[0122] The embodiment of the present invention extracts infrared radiation signals from infrared images. The signals in the infrared images usually need to be converted according to the calibration coefficients of the sensor to obtain the actual radiation brightness value. The radiation brightness value is converted into a temperature value using the radiation calibration coefficient of the infrared sensor; if there are multiple images or data at multiple time points, calibration is required to ensure consistency between different data, and any necessary calibration algorithms (such as adjustments in time and space) are applied to correct the temperature data; after the calibration process, the obtained temperature calibration data will more accurately reflect the actual temperature distribution in the image.

[0123] Step S262: Calculate the difference between the crop canopy temperature and the ambient temperature based on the temperature calibration data, and evaluate the water stress degree, thereby obtaining water stress degree data;

[0124] The embodiment of the present invention extracts the temperature value of the crop canopy from the temperature calibration data, usually by selecting a region of interest (ROI) or defining a specific area in the image. The ambient temperature data corresponding to the crop canopy is obtained. The ambient temperature can be obtained by selecting a surrounding area in the image or through ground sensor data. A threshold standard for water stress (such as a critical value of the temperature difference) is set. A large temperature difference may indicate that the crops are under water stress. Based on the calculated temperature difference, the degree of water stress is evaluated and the corresponding area is marked to generate water stress degree data.

[0125] Step S263: using a clustering algorithm to identify temperature anomaly regions on the precise temperature distribution data in the multispectral data set, and marking the abnormal regions, thereby obtaining temperature anomaly region data;

[0126] The embodiment of the present invention uses a clustering algorithm (such as K-means clustering, DBSCAN, etc.) to analyze the precise temperature distribution data and identify the abnormal temperature area; the clustering algorithm groups the temperature data according to similarity to help identify the abnormal area. According to the clustering results, the abnormal temperature area is marked. These areas may show abnormal high or low temperature characteristics; the identified abnormal temperature area is marked and recorded in the data set to generate abnormal temperature area data.

[0127] Step S264: Integrate the water stress degree data and the temperature abnormality area data to obtain thermal characteristic data.

[0128] The embodiment of the present invention integrates the water stress degree data with the temperature anomaly area data, and can merge the features of the two data sets through data fusion techniques, such as weighted average, joint analysis, etc. According to the application requirements, a suitable data fusion method is selected. For example, the water stress degree and the temperature anomaly area can be superimposed to generate a layer containing comprehensive thermal features. The integrated data set will contain information on water stress and temperature anomaly, providing a comprehensive view of the thermal characteristics of crops, and the generated thermal feature data will be used for subsequent analysis or decision support.

[0129] The present invention can ensure the accuracy and reliability of infrared image data through temperature calibration processing, and provide an accurate temperature data basis for subsequent analysis. This step is crucial for infrared images in multispectral data sets because it directly affects the accuracy of water stress assessment. Calculating the difference between the canopy temperature of crops and the ambient temperature and assessing the degree of water stress can help to timely discover the moisture status of crops and provide a scientific basis for irrigation decisions. Indicators such as water stress index (CWSI) can reflect the energy exchange between crops and the atmosphere and are important indicators for diagnosing crop water shortage. Using clustering algorithms to analyze precise temperature distribution data, temperature anomaly areas can be identified, which may indicate problems in crop growth, such as pests and diseases or nutrient deficiencies. Clustering algorithms can process large-scale data sets, identify data samples that are different from normal patterns, and are highly efficient in anomaly detection. Integrating water stress data and temperature anomaly area data to form a comprehensive thermal feature data set provides a comprehensive perspective for monitoring and evaluating crop growth conditions. Data integration helps to improve the accuracy of data analysis and the efficiency of decision support. Through these steps, the moisture status and growth environment of crops can be monitored and evaluated more accurately, and irrigation and field management measures can be adjusted in time, thereby improving crop yield and quality and achieving the goal of precision agriculture.

[0130] Preferably, step S3 comprises the following steps:

[0131] Step S31: extract key band information using the hyperspectral image in the multispectral data set, and fuse the key band information with the morphological feature data at the pixel level to generate an enhanced morphological feature map, wherein the key band information includes characteristic bands of near infrared, red edge and short-wave infrared;

[0132] The embodiment of the present invention extracts key band information from the hyperspectral image, including near infrared (NIR), red edge (RE), and short wave infrared (SWIR) bands; usually, the hyperspectral image data will have multiple bands, and it is necessary to select the bands that are sensitive to the morphological characteristics of crops; extract the data of these bands through spectral analysis tools or scripts to generate images of the corresponding bands. The extracted key band information is fused with the morphological feature data (such as the morphological feature data in step S25) at the pixel level, and image fusion techniques such as weighted fusion and principal component analysis (PCA) can be used. Through the fusion algorithm, the key band information and the morphological feature data are combined to generate an enhanced morphological feature map, which will contain more crop details and morphological features.

[0133] Step S32: performing spatial resolution enhancement processing on the thermal feature data, and using the spatial information in the hyperspectral image to guide filtering, thereby obtaining high-resolution thermal feature map data;

[0134] The embodiment of the present invention performs spatial resolution improvement processing on thermal signature data, and can use interpolation methods such as bilinear interpolation, cubic interpolation, etc. to improve the spatial resolution of the thermal signature map. Using the spatial information in the hyperspectral image as a guide, a guided filtering algorithm is applied to improve the resolution of the thermal signature map. The guided filtering algorithm uses the detail information of the hyperspectral image to adjust the pixel values ​​in the thermal signature map to improve the image quality and resolution. After the guided filtering process, the obtained thermal signature map will have a higher spatial resolution and clearer detail information.

[0135] Step S33: superimposing the enhanced morphological feature map and the high-resolution thermal feature map at the feature level to generate an enhanced multispectral data set;

[0136] The embodiment of the present invention superimposes the enhanced morphological feature map and the high-resolution thermal feature map at the feature level. Image fusion techniques such as weighted superposition and linear superposition can be used to combine the two layers. Through the superposition process, an enhanced multispectral data set containing enhanced morphological features and high-resolution thermal features is generated.

[0137] Step S34: extracting reflectance of key band information according to the hyperspectral image data and the enhanced morphological feature map, thereby obtaining spectral reflectance data;

[0138] The embodiment of the present invention extracts the reflectance values ​​of key bands (near infrared, red edge, short-wave infrared) from the hyperspectral image. The reflectance data is radiation-corrected data, which represents the spectral reflectance characteristics of each pixel. According to the band information extracted in step S31, the reflectance of each band is calculated, and the corresponding spectral reflectance data is generated.

[0139] Step S35: Calculate the normalized vegetation index and the enhanced vegetation index based on the enhanced multispectral data set and the spectral reflectance data to obtain vegetation index data, where the normalized vegetation index = (near infrared band reflectance - red light band reflectance) / (near infrared band reflectance + red light band reflectance), and the enhanced vegetation index = gain factor * ((near infrared band reflectance - red light band reflectance) / (near infrared band reflectance + aerosol correction factor * red light band reflectance - aerosol correction factor * blue light band reflectance + soil adjustment factor)).

[0140] The calculation formula of the Normalized Difference Vegetation Index (NDVI) in the embodiment of the present invention is: , where NIR is the reflectance of the near-infrared band and Red is the reflectance of the red light band; the formula is used to calculate NDVI to generate NDVI data, which can reflect the health and density of vegetation. The Enhanced Vegetation Index (EVI) calculation formula is: , where Gain is the gain factor, C1 and C2 are aerosol correction coefficients, L is the soil adjustment factor, NIR is the near-infrared band reflectance, Red is the red light band reflectance, and Blue is the blue light band reflectance. Applying the formula to calculate EVI and generate EVI data helps to more accurately assess the status of vegetation, especially in cases of high vegetation cover and large aerosol influence.

[0141] The present invention can generate a more accurate and detailed enhanced morphological feature map by extracting the key band information of the hyperspectral image and fusing it with the morphological feature data at the pixel level. This helps to more accurately identify and analyze the morphological features of crops, such as leaf structure and growth density. By performing spatial resolution enhancement processing on the thermal feature data and using the spatial information of the hyperspectral image for guided filtering, higher-definition thermal feature map data can be obtained. This helps to more carefully observe and analyze the thermal characteristics of crops, such as temperature distribution and water stress. By superimposing the enhanced morphological feature map and the high-resolution thermal feature map at the feature level, a more comprehensive enhanced multispectral data set can be generated. This data set can provide richer information and help to conduct a more in-depth analysis of the growth status of crops. Reflectance extraction of key band information based on hyperspectral image data and enhanced morphological feature maps can obtain more accurate spectral reflectance data. These data are crucial for understanding the spectral characteristics of crops and calculating vegetation indices. Based on the enhanced multispectral data set and spectral reflectance data, the normalized vegetation index (NDVI) and enhanced vegetation index (EVI) are calculated. These vegetation indices can reflect the growth and health of vegetation, which is of great significance for crop monitoring and management. Through these steps, the accuracy and efficiency of crop monitoring can be significantly improved, providing a scientific basis for agricultural production and helping to achieve the goal of precision agriculture.

[0142] Preferably, step S4 comprises the following steps:

[0143] Step S41: Statistically analyzing the vegetation index data to identify the changing trend of the vegetation index based on the time series and spatial distribution, thereby obtaining vegetation dynamic feature data;

[0144] The embodiment of the present invention extracts time series data from the vegetation index data set to ensure that the data covers a sufficiently long period of time to capture the changes in the vegetation index; uses time series analysis methods (such as moving average, exponential smoothing, seasonal decomposition, etc.) to identify the changing trend of the vegetation index, and the analysis may include seasonal changes, long-term trends, and periodic fluctuations; calculates statistical indicators such as mean, variance, trend line slope, etc. to quantify the changing characteristics of the vegetation index. The vegetation index data is distributed and analyzed according to spatial position to create a spatial distribution map; spatial statistical methods (such as spatial autocorrelation analysis, hot spot analysis, etc.) are applied to identify the spatial distribution trend of the vegetation index in different regions; combines the time series and spatial distribution analysis results to generate vegetation dynamic feature data to describe the changing characteristics of vegetation in time and space.

[0145] Step S42: constructing a semi-theoretical model between spectral index and crop physiological parameters according to a preset sample farmland area, thereby obtaining a physiological parameter estimation model;

[0146] The embodiment of the present invention selects a representative farmland area as a sample area to ensure that the vegetation index and physiological parameter data of the sample area can cover various growth conditions of the target crops; collects vegetation index and corresponding crop physiological parameter data in the sample area, such as chlorophyll content, plant height, biomass, etc. A suitable semi-theoretical model is selected to describe the relationship between the spectral index and the physiological parameters. Common models include linear regression models, nonlinear regression models, mixed models, etc.; the model is trained using data from the sample area, and the model parameters are optimized to accurately fit the relationship between the spectral index and the physiological parameters; the accuracy and stability of the model are verified by cross-validation or independent test sets to ensure the reliability of the model. Based on the training and verification results, a physiological parameter estimation model is generated for subsequent physiological parameter estimation.

[0147] Step S43: inputting the vegetation dynamic characteristic data into the physiological parameter estimation model to estimate the crop physiological parameters, thereby obtaining the crop physiological parameter data;

[0148] The embodiment of the present invention uses vegetation dynamic characteristic data (such as NDVI or EVI) as input and provides it to the physiological parameter estimation model; uses the pre-built physiological parameter estimation model to process the input data and calculate the corresponding crop physiological parameters. Based on the model output results, the crop physiological parameter data, such as chlorophyll content, plant biomass, etc., are generated.

[0149] Step S44: performing time series analysis on the ground environment monitoring data, extracting the daily, weekly and seasonal variation characteristics of temperature, humidity, soil moisture and light intensity, thereby obtaining environmental variation characteristic data;

[0150] The embodiment of the present invention extracts data on temperature, humidity, soil moisture and light intensity from ground environmental monitoring data, ensuring that the data covers a sufficient time range for detailed analysis; analyzes the daily variation characteristics of each environmental factor, such as daily temperature fluctuations, humidity changes, etc., and can use periodic analysis methods (such as Fourier transform) to identify intra-day variation patterns; analyzes the weekly and seasonal variation characteristics of environmental factors, and can apply seasonal decomposition methods (such as STL decomposition) to identify and quantify these changes. Integrate the results of daily, weekly and seasonal changes to generate environmental change feature data to describe the time variation characteristics of environmental factors.

[0151] Step S45: Using the Kriging method to perform spatial interpolation on discrete ground monitoring point data according to the environmental change characteristic data, to generate a continuous environmental factor distribution map;

[0152] The embodiment of the present invention organizes discrete ground monitoring point data, including the measured values ​​and geographic coordinates of environmental factors; and uses Kriging to perform spatial interpolation on discrete data. Kriging is an interpolation method based on spatial autocorrelation, which can generate a smooth and continuous distribution map of environmental factors; and generates an environmental factor distribution map through the interpolation results to show the spatial changes of environmental factors in the entire study area.

[0153] Step S46: local microclimate characteristics are determined based on the pre-acquired meteorological data and the environmental factor distribution map, thereby generating crop growth environment characteristic data.

[0154] The embodiment of the present invention collects pre-acquired meteorological data, such as temperature, humidity, wind speed, etc., to ensure that these data match the environmental factor distribution map; combines the meteorological data and the environmental factor distribution map to analyze the local microclimate characteristics. Spatial analysis methods (such as buffer zone analysis and spatial regression analysis) can be used to identify the impact of climate conditions on crop growth. Based on the analysis results, crop growth environment characteristic data is generated to describe the local climate conditions that affect crop growth.

[0155] The present invention identifies the vegetation index change trend in time series and spatial distribution by statistically analyzing vegetation index data, which helps to understand the vegetation growth status and seasonal change law, and provides important information for crop growth monitoring and ecological environment assessment. According to the relationship between the spectral index and the crop physiological parameters of the sample farmland area, a semi-theoretical model is constructed, which helps to convert remote sensing data into quantitative indicators of crop physiological state and improve the accuracy of crop growth status monitoring. The physiological parameter estimation model is input with vegetation dynamic characteristic data to obtain crop physiological parameter data, which can provide crop nutritional status, growth potential and pest risk assessment for agricultural production. Time series analysis is performed on ground environmental monitoring data to extract the change characteristics of environmental factors such as temperature, humidity, soil moisture and light intensity, which helps to identify the seasonal and daily change laws of crop growth environment. The ground monitoring point data is spatially interpolated using the Kriging method to generate a continuous environmental factor distribution map, which can provide a more detailed spatial distribution of environmental conditions and provide accurate environmental information for agricultural planning and management. Combined with the pre-acquired meteorological data and environmental factor distribution map, local microclimate characteristics analysis is performed to generate crop growth environment characteristic data, which helps to evaluate and optimize the microclimate conditions for crop growth and improve crop yield and quality. Through these steps, the physiological state and growth environment of crops can be comprehensively analyzed and evaluated, providing a scientific basis for precise management and decision-making of agricultural production.

[0156] Preferably, step S5 comprises the following steps:

[0157] Step S51: spatially registering the enhanced multispectral dataset, crop physiological parameter data, and growth environment characteristic data, and uniformly reprojecting them to the same geographic coordinate system, thereby obtaining a spatially consistent multi-source dataset;

[0158] Embodiment of the present invention: Obtain enhanced multispectral data sets, crop physiological parameter data and growth environment characteristic data, ensure that these data are derived from different sensors or monitoring equipment, and have different geographic reference information. Select a main reference data set (usually the data with higher resolution or the most accurate) as the registration benchmark, and other data sets will be aligned with this reference data set. Apply spatial registration algorithms (such as registration based on feature point matching, registration based on similarity transformation, etc.) to align all data sets to the coordinate system of the reference data, and software tools such as GIS (Geographic Information System) software or image processing libraries can be used to perform registration. Determine a unified geographic coordinate system (such as WGS84, UTM, etc.); reproject all registered data sets to a unified geographic coordinate system, which can be achieved through GIS software or projection tools to ensure that all data sets have consistent geographic locations in the same coordinate system. Ensure that all data are aligned in the same geographic coordinate system and generate a spatially consistent multi-source data set.

[0159] Step S52: matching and interpolating the data at the same time node on the multi-source data sets, thereby forming a comprehensive data set that is consistent in time and space;

[0160] The embodiment of the present invention identifies the corresponding time nodes in all data sets, ensuring that the time points are the same or close in all data sets for effective spatiotemporal matching. Match the time nodes of all data, and for data at different time points, interpolation methods (such as linear interpolation, spline interpolation, etc.) can be applied to adjust them to the same time nodes. Select a suitable interpolation algorithm for spatiotemporal interpolation of data, which can include spatial interpolation (such as Kriging, inverse distance weighting) and time interpolation (such as time series interpolation). Interpolate the data to generate consistent data at the same time node, ensuring that the interpolated data can cover the entire area and maintain spatiotemporal consistency. The processed data are synthesized into a comprehensive data set that is consistent in time and space, including image features, spectral features, physiological parameters, and environmental features.

[0161] Step S53: extracting image features, spectral features, physiological parameter features and environmental features from the comprehensive data set, thereby obtaining a comprehensive feature data set;

[0162] The embodiment of the present invention applies image processing technology to extract image features, such as texture, shape, edge, etc. Filters, edge detection algorithms, etc. can be used; spectral features, such as vegetation index, spectral reflectance, etc. are extracted from the comprehensive data set. Use the vegetation index (such as NDVI, EVI) and other spectral indicators calculated previously. Extract the physiological parameter characteristics of crops, such as chlorophyll content, biomass, etc., and extract these parameters from the previously obtained physiological parameter data. Extract environmental characteristics, such as temperature, humidity, soil moisture and light intensity, and analyze important indicators in the environmental change characteristic data; integrate image features, spectral features, physiological parameter features and environmental characteristics to generate a comprehensive feature data set.

[0163] Step S54: using the pre-trained pest and disease identification model to identify the growth abnormalities and potential pest and disease risks of crops in the comprehensive feature data set, and marking the growth abnormality areas and potential pest and disease risk areas, thereby obtaining abnormal area marking data;

[0164] The embodiment of the present invention selects a pre-trained pest and disease identification model, which is usually a model based on deep learning (such as a convolutional neural network CNN) or a machine learning model (such as a support vector machine SVM); the comprehensive feature data set is input into the pre-trained model for processing. Ensure that the format and input requirements of the data meet the requirements of the model. Use the model to process the comprehensive feature data set to identify abnormal growth areas and potential pest and disease risk areas, and the model will output the labels of these areas; based on the recognition results of the model, the abnormal areas and pest and disease risk areas are labeled to generate corresponding abnormal area labeling data.

[0165] Step S55: Generate decision suggestions based on preset expert knowledge and abnormal area marking data, so as to obtain decision suggestion data.

[0166] The embodiments of the present invention organize and integrate expert knowledge related to crop growth and pest and disease control, including the characteristics, control measures and best practices of pests and diseases. Rules and processes for formulating decision recommendations are based on expert knowledge. These rules may include how to deal with specific types of growth abnormalities or pest and disease risks. Based on the abnormal area marking data and preset expert knowledge, specific decision recommendations are generated, such as recommended treatment measures, management strategies to be adopted, etc.; final decision recommendation data is generated and provided to farm managers or agricultural experts for actual operation.

[0167] The present invention ensures the consistency of data in geographic space by spatially registering enhanced multispectral data sets, crop physiological parameter data and growth environment characteristic data, which is convenient for accurate data analysis and comparison. Data matching and interpolation processing at the same time node form a comprehensive data set that is consistent in time and space, which provides reliable data support for time series analysis and dynamic monitoring. Image features, spectral features, physiological parameter features and environmental features are extracted to obtain a comprehensive feature data set, which enriches the dimension of the data and provides more valuable information for in-depth analysis. The comprehensive feature data set is analyzed using a pre-trained pest and disease recognition model to automatically identify abnormal growth of crops and potential pest and disease risk areas, thereby improving the efficiency and accuracy of pest and disease monitoring. Combining expert knowledge and abnormal area marking data, decision recommendations are generated to provide scientific decision support for agricultural production management, helping farmers to take prevention and control measures in a timely manner and reduce losses. Through these steps, an efficient and accurate crop pest and disease monitoring and decision support system can be constructed to achieve real-time monitoring and intelligent management of crop growth conditions.

[0168] The present invention also provides a decision-making system for multi-source information fusion, which is used to execute the above-mentioned decision-making method for multi-source information fusion. The decision-making system for multi-source information fusion comprises:

[0169] The data acquisition module is used to obtain the hyperspectral image data, visible light image data and infrared image data collected by the UAV, and collect the ground environment monitoring data collected by the ground sensor; perform image preprocessing on the hyperspectral image data, visible light image data and infrared image data to obtain a standardized multispectral data set, wherein the image preprocessing includes geometric correction, radiation correction and atmospheric correction;

[0170] The feature extraction module is used to extract color and texture features from visible light images in the multispectral data set, and identify the appearance features and growth density of crops to obtain morphological feature data; analyze the thermal features of crops from infrared images in the multispectral data set, and identify water stress and temperature anomalies to obtain thermal feature data;

[0171] The multi-source data fusion module is used to integrate the morphological feature data and thermal feature data according to the hyperspectral images in the multispectral data set, so as to generate an enhanced multispectral data set; the normalized vegetation index and the enhanced vegetation index are calculated according to the enhanced multispectral data set, so as to obtain vegetation index data;

[0172] The physiological parameter estimation module is used to estimate the crop physiological parameters based on the spectral reflectance characteristics according to the vegetation index data, so as to obtain the crop physiological parameter data; and to use the ground environment monitoring data to perform microclimate analysis, so as to generate the crop growth environment characteristic data;

[0173] The intelligent decision-making module is used to perform spatiotemporal registration and correlation analysis on the enhanced multispectral data set, crop physiological parameter data, and growth environment characteristic data to generate a comprehensive characteristic data set; the pre-trained pest and disease identification model is used to identify the growth anomalies and potential pest and disease risks of crops in the comprehensive characteristic data set, and decision recommendations are generated based on preset expert knowledge to obtain decision recommendation data;

[0174] The irrigation and fertilization decision-making assistance module is used to evaluate the irrigation amount based on humidity and crop water demand according to the decision recommendation data, and to evaluate the nutrients required for crop growth, so as to generate an irrigation prescription map to assist engineers in making irrigation and fertilization decisions.

[0175] The present invention collects multi-source data through unmanned aerial vehicles and ground sensors, and after image preprocessing steps such as geometric correction, radiation correction, and atmospheric correction, the standardization and quality of the data are ensured, providing accurate basic data for subsequent analysis. Feature extraction of visible light images and infrared images, including color, texture, thermal features, etc., helps to identify the appearance characteristics, growth density, water stress and temperature anomalies of crops, and provides important information for evaluating crop growth conditions. Integrate hyperspectral image data, morphological feature data and thermal feature data to generate enhanced multispectral data sets, and calculate vegetation indexes, providing a scientific basis for estimating crop physiological parameters. Combined with ground environmental monitoring data and meteorological data, microclimate analysis and pest and disease identification are carried out, which helps to timely discover crop growth abnormalities and potential pest and disease risks, and provides scientific decision support for agricultural production. Through spatiotemporal registration and association analysis, a comprehensive feature data set is generated, and a pest and disease identification model is used for risk identification, and decision-making recommendations are generated in combination with expert knowledge, which improves the intelligence and precision level of agricultural management. Based on the irrigation volume assessment and nutrient assessment of decision-making recommendation data, accurate irrigation prescription maps can be generated to guide engineers to carry out precise irrigation and fertilization, which helps to improve the utilization efficiency of water resources and fertilizers, reduce waste, and promote healthy growth of crops. Automated data collection and analysis reduce manpower requirements and improve the efficiency of monitoring and decision-making; through accurate data analysis, more accurate irrigation and fertilization can be achieved, reducing resource waste; pest and disease identification models can detect problems early and reduce crop losses; decision support systems based on comprehensive data provide scientific decision-making basis and improve the accuracy of decision-making; optimizing resource utilization and reducing environmental impact help achieve sustainable agricultural development. The entire system realizes comprehensive monitoring and analysis of crop growth conditions through high-tech means, which helps to improve the level of intelligence in agricultural production, optimize resource utilization, reduce costs, increase output, and ultimately achieve sustainable agricultural development.

[0176] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is therefore intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0177] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A decision-making method for multi-source information fusion, characterized in that: The following steps are involved: Step S1: Acquire hyperspectral image data, visible light image data and infrared image data collected by the UAV, and collect ground environment monitoring data collected by ground sensors; perform image preprocessing on the hyperspectral image data, visible light image data and infrared image data to obtain a standardized multispectral data set, wherein the image preprocessing includes geometric correction, radiation correction and atmospheric correction; Step S2: extracting color and texture features from the visible light images in the multispectral data set, and identifying the appearance features and growth density of the crops, thereby obtaining morphological feature data; analyzing the thermal features of the crops from the infrared images in the multispectral data set, and identifying water stress and temperature anomalies, thereby obtaining thermal feature data; Step S3: performing image integration on the morphological feature data and the thermal feature data according to the hyperspectral image in the multispectral dataset, thereby generating an enhanced multispectral dataset; The normalized vegetation index and the enhanced vegetation index are calculated according to the enhanced multispectral data set to obtain vegetation index data. Step S3 includes: Step S31: extract key band information using the hyperspectral image in the multispectral data set, and fuse the key band information with the morphological feature data at the pixel level to generate an enhanced morphological feature map, wherein the key band information includes characteristic bands of near infrared, red edge and short-wave infrared; Step S32: performing spatial resolution enhancement processing on the thermal feature data, and using the spatial information in the hyperspectral image to guide filtering, thereby obtaining high-resolution thermal feature map data; Step S33: superimposing the enhanced morphological feature map and the high-resolution thermal feature map at the feature level to generate an enhanced multispectral data set; Step S34: extracting reflectance of key band information according to the hyperspectral image data and the enhanced morphological feature map, thereby obtaining spectral reflectance data; Step S35: Calculate the normalized vegetation index and the enhanced vegetation index based on the enhanced multispectral data set and the spectral reflectance data, so as to obtain vegetation index data, wherein the normalized vegetation index = (near infrared band reflectance - red light band reflectance) / (near infrared band reflectance + red light band reflectance), enhanced vegetation index = gain factor * ((near infrared band reflectance - red light band reflectance) / (near infrared band reflectance + aerosol correction factor * red light band reflectance - aerosol correction factor * blue light band reflectance + soil adjustment factor)); Step S4: Estimating crop physiological parameters based on spectral reflectance characteristics according to vegetation index data, thereby obtaining crop physiological parameter data; performing microclimate analysis using ground environment monitoring data, thereby generating crop growth environment characteristic data; Step S5: performing spatiotemporal registration and correlation analysis on the enhanced multispectral data set, crop physiological parameter data, and growth environment characteristic data to generate a comprehensive characteristic data set; using a pre-trained pest and disease recognition model to identify abnormal growth and potential pest and disease risks of crops in the comprehensive characteristic data set, and generating decision recommendations based on preset expert knowledge, thereby obtaining decision recommendation data; Step S6: Evaluate the irrigation amount based on humidity and crop water demand according to the decision recommendation data, and evaluate the nutrients required for crop growth, so as to generate an irrigation prescription map to assist engineers in making irrigation and fertilization decisions.

2. The decision-making method of multi-source information fusion according to claim 1, characterized in that: Step S1 includes the following steps: Step S11: using a drone equipped with a multispectral sensor to collect hyperspectral image data, visible light image data, and infrared image data; Step S12: collecting temperature, humidity, soil moisture and light intensity information through a ground sensor network deployed on the target farmland, thereby obtaining ground environment monitoring data; Step S13: performing integrity detection on the hyperspectral image data, the visible light image data and the infrared image data, and removing or marking areas where the degree of influence of clouds and shadows exceeds a predetermined threshold, thereby obtaining hyperspectral image marking data, visible light image marking data and infrared image marking data; Step S14: performing image preprocessing on the hyperspectral image labeling data, the visible light image labeling data and the infrared image labeling data, so as to obtain a standardized multispectral data set, wherein the image preprocessing includes geometric correction, radiation correction and atmospheric correction.

3. The decision-making method of multi-source information fusion according to claim 2 is characterized in that: Step S13 includes the following steps: Step S131: performing pixel-level integrity scanning on the hyperspectral image data, visible light image data, and infrared image data to identify pixels with missing or damaged data, thereby obtaining incomplete pixel data; Step S132: using the shadow detection model and the cloud detection model to identify the cloud layer and the shadow area of ​​the hyperspectral image data, the visible light image data and the infrared image data, thereby obtaining the cloud layer and the shadow area data; Step S133: Calculate the ratio of cloud and shadow coverage areas in each image based on the cloud and shadow area data, and mark the images whose coverage ratio exceeds a predetermined threshold as low-quality data; Step S134: For low-quality data, if the coverage ratio exceeds a preset critical threshold, the image is removed from the data set; if the coverage ratio is between the preset threshold and the critical threshold, a quality mark is added to the metadata to obtain marked image data; Step S135: using an image restoration algorithm to fill in missing or damaged pixels, small cloud areas and shadow areas according to the incomplete pixel data and the marked image data, thereby obtaining hyperspectral image marked data, visible light image marked data and infrared image marked data.

4. The decision-making method of multi-source information fusion according to claim 3 is characterized in that: Step S14 includes the following steps: Step S141: Correcting the geometric distortion of the hyperspectral image marker data due to the terrain and the shooting angle, converting the digital values ​​of the hyperspectral image marker data into reflectance and radiance values, thereby obtaining the reflectance or radiance data of the hyperspectral image after radiation correction; Step S142: Eliminate the influence of atmospheric scattering and absorption according to the reflectivity or radiance data, so as to obtain surface reflectivity data; Step S143: performing color balance adjustment based on the actual object color on the visible light image mark data, and performing contrast enhancement processing, so as to obtain visible light contrast enhancement data; Step S144: converting the infrared signal in the infrared image tag data into an actual temperature value, and performing emissivity correction based on the emissivity difference of different ground objects, so as to obtain accurate temperature distribution data; Step S145: aligning the surface reflectance data, visible light ratio enhancement data and precise temperature distribution data based on ground space and time series, thereby obtaining a standardized multispectral data set.

5. The decision-making method of multi-source information fusion according to claim 4 is characterized in that: Step S2 includes the following steps: Step S21: performing color space conversion on the visible light image in the multispectral data set, from the RGB space to the HSV space, and extracting hue, saturation and brightness features, thereby obtaining color features of the visible light image; Step S22: extracting texture features of visible light images in the multispectral data set using a gray level co-occurrence matrix method, and calculating contrast, entropy, energy, and homogeneity to obtain texture statistics; Step S23: performing crop appearance feature recognition based on deep learning according to the color features of the visible light image and the texture statistics, thereby obtaining appearance feature data, wherein the crop appearance feature recognition includes leaf morphology and plant type structure; Step S24: performing crop growth density analysis on the visible light image in the multispectral data set by using image segmentation and density estimation algorithms, thereby obtaining a density distribution map; Step S25: fusing the appearance feature data and the density distribution map to obtain morphological feature data; Step S26: Analyze the thermal characteristics of crops on the infrared images in the multispectral data set, and identify water stress and temperature anomalies, so as to obtain thermal characteristic data.

6. The decision-making method of multi-source information fusion according to claim 5, characterized in that: Step S26 includes the following steps: Step S261: performing temperature calibration processing on the infrared image in the multispectral data set to obtain temperature calibration data; Step S262: Calculate the difference between the crop canopy temperature and the ambient temperature based on the temperature calibration data, and evaluate the water stress degree, thereby obtaining water stress degree data; Step S263: using a clustering algorithm to identify temperature anomaly regions on the precise temperature distribution data in the multispectral data set, and marking the abnormal regions, thereby obtaining temperature anomaly region data; Step S264: Integrate the water stress degree data and the temperature abnormality area data to obtain thermal characteristic data.

7. The decision-making method of multi-source information fusion according to claim 6, characterized in that: Step S4 includes the following steps: Step S41: Statistically analyzing the vegetation index data to identify the changing trend of the vegetation index based on the time series and spatial distribution, thereby obtaining vegetation dynamic feature data; Step S42: constructing a semi-theoretical model between spectral index and crop physiological parameters according to a preset sample farmland area, thereby obtaining a physiological parameter estimation model; Step S43: inputting the vegetation dynamic characteristic data into the physiological parameter estimation model to estimate the crop physiological parameters, thereby obtaining the crop physiological parameter data; Step S44: performing time series analysis on the ground environment monitoring data, extracting the daily, weekly and seasonal variation characteristics of temperature, humidity, soil moisture and light intensity, thereby obtaining environmental variation characteristic data; Step S45: Using the Kriging method to perform spatial interpolation on discrete ground monitoring point data according to the environmental change characteristic data, to generate a continuous environmental factor distribution map; Step S46: local microclimate characteristics are determined based on the pre-acquired meteorological data and the environmental factor distribution map, thereby generating crop growth environment characteristic data.

8. The decision-making method of multi-source information fusion according to claim 7, characterized in that: Step S5 includes the following steps: Step S51: spatially registering the enhanced multispectral dataset, crop physiological parameter data, and growth environment characteristic data, and uniformly reprojecting them to the same geographic coordinate system, thereby obtaining a spatially consistent multi-source dataset; Step S52: matching and interpolating the data at the same time node on the multi-source data sets, thereby forming a comprehensive data set that is consistent in time and space; Step S53: extracting image features, spectral features, physiological parameter features and environmental features from the comprehensive data set, thereby obtaining a comprehensive feature data set; Step S54: using the pre-trained pest and disease identification model to identify the growth abnormalities and potential pest and disease risks of crops in the comprehensive feature data set, and marking the growth abnormality areas and potential pest and disease risk areas, thereby obtaining abnormal area marking data; Step S55: Generate decision suggestions based on preset expert knowledge and abnormal area marking data, so as to obtain decision suggestion data.

9. A decision-making system for multi-source information fusion, characterized in that: The decision-making method for multi-source information fusion according to claim 1 is used to execute the decision-making system for multi-source information fusion, comprising: The data acquisition module is used to obtain the hyperspectral image data, visible light image data and infrared image data collected by the UAV, and collect the ground environment monitoring data collected by the ground sensor; perform image preprocessing on the hyperspectral image data, visible light image data and infrared image data to obtain a standardized multispectral data set, wherein the image preprocessing includes geometric correction, radiation correction and atmospheric correction; The feature extraction module is used to extract color and texture features from visible light images in the multispectral data set, and identify the appearance features and growth density of crops to obtain morphological feature data; analyze the thermal features of crops from infrared images in the multispectral data set, and identify water stress and temperature anomalies to obtain thermal feature data; The multi-source data fusion module is used to integrate the morphological feature data and thermal feature data according to the hyperspectral images in the multispectral data set, so as to generate an enhanced multispectral data set; the normalized vegetation index and the enhanced vegetation index are calculated according to the enhanced multispectral data set, so as to obtain vegetation index data; The physiological parameter estimation module is used to estimate the crop physiological parameters based on the spectral reflectance characteristics according to the vegetation index data, so as to obtain the crop physiological parameter data; and to use the ground environment monitoring data to perform microclimate analysis, so as to generate the crop growth environment characteristic data; The intelligent decision-making module is used to perform spatiotemporal registration and correlation analysis on the enhanced multispectral data set, crop physiological parameter data, and growth environment characteristic data to generate a comprehensive characteristic data set; the pre-trained pest and disease identification model is used to identify the growth anomalies and potential pest and disease risks of crops in the comprehensive characteristic data set, and decision recommendations are generated based on preset expert knowledge to obtain decision recommendation data; The irrigation and fertilization decision-making assistance module is used to evaluate the irrigation amount based on humidity and crop water demand according to the decision recommendation data, and to evaluate the nutrients required for crop growth, so as to generate an irrigation prescription map to assist engineers in making irrigation and fertilization decisions.

Images