A potato growth monitoring system and method based on UAV multispectral imaging

Abstract

This invention discloses a potato growth monitoring system and method based on UAV multispectral imaging. The system includes an input module, a control module, and an output module. The input module collects and receives multi-source data; the control module preprocesses the input multi-source data, extracts and co-optimizes spectral and texture features, and adaptively selects, dynamically switches, and fuses candidate models based on growth stage identification results to generate a spatial distribution map of potato chlorophyll; the output module outputs the monitoring results. Compared to existing technologies, this invention constructs a multi-model adaptive inversion mechanism for different growth stages, which can automatically match the optimal estimation model according to changes in potato growth stages, improving the accuracy, stability, and environmental adaptability of chlorophyll inversion. This system and method can achieve rapid, non-destructive, and visual monitoring of potato growth at a regional scale, providing reliable technical support for growth assessment, precise nitrogen fertilizer control, and smart agricultural management.

Description

Technical Field

[0001] This invention relates to the fields of precision agriculture and agricultural remote sensing technology for field crops, specifically to a potato growth monitoring system and method based on UAV multispectral imaging. Background Technology

[0002] The chlorophyll content of potatoes is a key indicator reflecting their photosynthetic capacity and nitrogen nutrition level. Therefore, it is of great significance for rapid and non-destructive monitoring of their growth in precision agricultural management and scientific research.

[0003] Traditional chlorophyll measurement methods primarily rely on SPAD (Soil and Plant Analyzer Development) instruments for point-based measurements. While the results are accurate, these methods are manual, have limited sample sizes, and struggle to reflect differences in crop canopy growth. In recent years, the development of UAV multispectral remote sensing technology has provided a high spatiotemporal resolution means for monitoring crop canopy parameters. It can reflect crop physiological states at the spectral level, providing data support for precision farmland management and has become an important technological direction for crop growth monitoring. Existing research and inventions indicate that UAV multispectral or hyperspectral technology has already been applied in the field of crop growth estimation.

[0004] However, traditional chlorophyll measurement relies primarily on manual spot measurements using SPAD instruments, resulting in limited sample sizes, low measurement efficiency, and difficulty in rapidly monitoring large-scale potato populations. While some technologies, such as UAV multispectral technology, provide high spatiotemporal resolution data for crop growth monitoring, existing devices often focus on image acquisition or vegetation index calculation, lacking systematic processing capabilities for multispectral images. Common methods typically rely on a single vegetation index for estimation, failing to fully integrate canopy structure and texture information; some systems incorporate texture features but lack feature selection and optimization modules, leading to severe feature redundancy and unstable model predictions. Furthermore, existing monitoring devices generally lack model selection mechanisms for different potato growth stages; inversion models are mostly single-stage trained, unable to dynamically adjust according to growth stage changes, resulting in outputs that are sensitive to environmental differences and have insufficient generalization ability. Overall, there is currently a lack of a complete device and system that integrates multispectral image processing, feature co-optimization, growth stage identification, and adaptive model switching, which limits the effectiveness of UAVs in quantitative monitoring of potato chlorophyll. Therefore, it is necessary to develop a potato chlorophyll estimation device with a complete structure, integrated functions, and the ability to improve inversion accuracy and stability, in order to meet the needs of precision agriculture for efficient and reliable monitoring methods. Summary of the Invention

[0005] To address the shortcomings of existing technologies for monitoring potato chlorophyll content, which primarily rely on manual point measurement using SPAD instruments, resulting in low measurement efficiency and insufficient sample representativeness, as well as the insufficient accuracy and stability of UAV multispectral inversion methods under different growth stages, this invention proposes a potato growth monitoring system and method based on UAV multispectral imagery. The system integrates multi-source information acquisition from the potato canopy, spectral-texture feature extraction, feature co-optimization, growth stage identification, and adaptive switching between multiple models by setting up input, control, and output modules. By embedding feature optimization mechanisms and dynamic model selection strategies into the control module, this invention can automatically match the optimal inversion model according to different growth stages, improving the accuracy and stability of chlorophyll estimation. This system enables rapid and non-destructive monitoring of potato growth at the regional scale, providing efficient and reliable technical support for potato growth assessment, nitrogen fertilizer regulation, and precision agricultural management.

[0006] To achieve this goal, the present invention adopts the following technical solution.

[0007] This invention discloses a potato growth monitoring system based on UAV multispectral imaging, comprising an input module, a control module, and an output module. The input module collects and receives multi-source data; the control module preprocesses the input multi-source data, extracts and co-optimizes spectral and texture features, and adaptively selects, dynamically switches, and fuses candidate models based on growth stage identification results to generate a spatial distribution map of potato chlorophyll; the output module outputs the monitoring results.

[0008] Furthermore, the potato growth monitoring system based on UAV multispectral imaging disclosed in this invention includes multi-source data such as UAV multispectral images of potato canopy, radiation correction parameters, aerial flight parameters and exposure parameters, measured ground SPAD values, and agronomic phenological information. The SPAD value is an index characterizing the relative chlorophyll content of potato leaves, used to construct and correct the potato relative chlorophyll content inversion model; the agronomic phenological information includes planting date, accumulated temperature, canopy coverage, and growth stage thresholds.

[0009] Furthermore, in the potato growth monitoring system based on UAV multispectral imaging disclosed in this invention, the control module includes a preprocessing unit for preprocessing the UAV multispectral images in the input multi-source data, wherein the preprocessing includes: Register and stitch together the multispectral images of the UAV, and then crop them according to the target area to obtain an orthophoto image in a unified coordinate system. Radiometric correction is performed on orthophotos using a radiometric correction plate. The radiometric correction formula is as follows: , in, The pixel brightness values ​​of the calibration board obtained from the remote sensing image of the calibration board captured by the UAV. To calibrate the reflectivity of the plate, The DN values ​​of potato canopy obtained from drone photography of the test area. The reflectance of the potato canopy.

[0010] Furthermore, in the potato growth monitoring system based on UAV multispectral imaging disclosed in this invention, the control module includes a spectral feature extraction unit, a texture feature extraction unit, and a feature fusion unit. The spectral feature extraction unit is used to extract vegetation indices from orthophotos to characterize the spectral response of the potato canopy. The texture feature extraction unit is used to extract texture features based on the gray-level co-occurrence matrix from orthophotos to characterize canopy structure features and spatial heterogeneity. The feature fusion unit is used to fuse the spectral response features and texture features to construct an initial joint feature set.

[0011] Furthermore, in the potato growth monitoring system based on UAV multispectral imaging disclosed in this invention, the control module includes a feature selection and optimization unit and a model building unit. The feature selection and optimization unit comprises a correlation analysis subunit, a maximum correlation and minimum redundancy subunit, and a genetic algorithm optimization subunit. The correlation analysis subunit uses Pearson analysis to evaluate the correlation between each feature in the initial joint feature set and the measured SPAD values ​​on the ground, eliminating features with correlations below a predetermined threshold. The maximum correlation and minimum redundancy unit reduces information redundancy among features, obtaining a subset of candidate features with strong information complementarity. The genetic algorithm unit performs combinatorial optimization search on the candidate feature subsets, determining the optimal feature subset based on the performance of the inversion model. The model building unit constructs a chlorophyll relative content inversion model, establishing a mapping relationship between spectral-texture features and SPAD values.

[0012] Furthermore, in the potato growth monitoring system based on UAV multispectral imaging disclosed in this invention, the control module includes a model selection and switching unit. This unit comprises a candidate model library, a stage identifier, a performance evaluator, a confidence evaluator, a switching controller, and a fusion unit. The candidate model library pre-configures regression models such as SVR, Lasso, XGBoost, GPR, and RF, and stores their phased training parameters; The stage identifier is used to determine the current potato growth stage based on the spectral response characteristics, texture characteristics, and agronomic phenological information, and generate a growth period label; The performance evaluator is used to assess the current fit of each candidate model based on the candidate model’s historical validation error and online residual statistics. The confidence estimator is used to output a measure of the uncertainty of each model in the current prediction results; The switching controller is used to calculate the comprehensive competency score of each candidate model and select the optimal model. The fusion unit is used to perform confidence-weighted fusion of the outputs of multiple candidate models when the score difference between multiple candidate models is less than a predetermined threshold.

[0013] The potato growth monitoring system based on UAV multispectral imaging as described in claim 6 is characterized in that the switching controller is used to perform a comprehensive competence score on each candidate model based on phenological consistency index, model validation error, uncertainty index, and online residual index, and selects the candidate model with the highest comprehensive competence score as the current optimal model. The formula for calculating the comprehensive competence score is as follows: , in, The phenological consistency score is used to reflect the degree of matching between the model and the current reproductive period; This represents the normalized cross-validation error of the model during the training phase. This is a normalized value for uncertainty; For the most recent window's online relative RMSE; , , and The weighting coefficients are set based on the cross-validation results and error sensitivity analysis of historical samples, and are corrected as adjustable parameters under different plots or growth stages. The optimal model is determined as follows: .

[0014] In addition, in the potato growth monitoring system based on UAV multispectral data disclosed in this invention, the model selection and switching are set with model switching trigger conditions. The model switching trigger conditions are used to constrain the comprehensive competence score difference between the candidate optimal model and the current model. When the score difference is greater than a preset hysteresis threshold for k consecutive time phases or k consecutive frames, the control system switches from the current model to the candidate optimal model.

[0015] This invention also discloses a method for monitoring potato growth based on UAV multispectral imaging, the method comprising the following steps: A. Collect and receive multi-source data, including multispectral images of potato canopy taken by UAV, radiometric correction parameters, aerial flight and exposure parameters, measured SPAD values, and agronomic phenological information. SPAD values ​​are an index characterizing the relative chlorophyll content of potato leaves, used to construct and calibrate the potato relative chlorophyll content inversion model; agronomic phenological information includes planting date, accumulated temperature, canopy coverage, and growth stage thresholds. B. Preprocess the input multi-source data, extract and co-optimize spectral and texture features, and combine the results of the fertility period identification to adaptively select, dynamically switch and fuse candidate models to generate a spatial distribution map of potato chlorophyll. C. Output monitoring results.

[0016] Furthermore, the potato growth monitoring method based on UAV multispectral imaging disclosed in this invention preprocesses the input multi-source data, extracts and co-optimizes spectral and texture features, and combines the results of growth stage identification to adaptively select, dynamically switch, and fuse candidate models to generate a potato chlorophyll spatial distribution map, including: Preprocessing involves registering and stitching the multispectral images of the potato canopy from the UAV, and cropping them according to the target area to obtain an orthorectified image with unified coordinates and complete coverage. Radiometric correction is then performed on the orthorectified image using a radiometric correction plate. The radiometric correction formula is as follows: , in, The pixel brightness values ​​of the calibration board obtained from the remote sensing image of the calibration board captured by the UAV. To calibrate the reflectivity of the plate, The DN values ​​of potato canopy obtained from drone photography of the test area. The reflectance of the potato canopy.

[0017] Spectral and texture feature extraction involves extracting vegetation index features that characterize the spectral response of the potato canopy from the orthophoto and extracting texture features based on the gray-level co-occurrence matrix to characterize canopy structure features and spatial heterogeneity.

[0018] Feature co-optimization involves fusing the vegetation index features and texture features to construct an initial joint feature set, and then filtering the initial joint feature set through correlation analysis, maximum correlation and minimum redundancy processing, and genetic algorithm combination optimization to obtain the optimal feature subset.

[0019] The adaptive selection, dynamic switching, and strategy fusion of models involve comprehensively scoring each candidate model based on its historical validation error, the uncertainty of the current prediction results, and the online residual statistics, and dynamically selecting the current optimal model. When the score difference between multiple candidate models is less than a predetermined threshold, the output results of multiple candidate models are weighted and fused.

[0020] Generating a spatial distribution map of potato chlorophyll involves estimating the relative chlorophyll content in the target area using the selected optimal model or fusion model, generating a spatial distribution map of potato chlorophyll, and outputting the monitoring results.

[0021] Compared with the prior art, the present invention has the following beneficial effects: This invention provides a potato growth monitoring system and method based on UAV multispectral imagery. It integrates multi-source data input, spectral and texture feature co-optimization, multi-model adaptive switching, and visualization output. This enables rapid inversion of the relative chlorophyll content in the potato canopy at different growth stages and visualized monitoring of potato growth. By introducing a feature optimization mechanism and a model adaptive selection strategy tailored to growth stage differences, the accuracy, stability, and environmental adaptability of chlorophyll estimation are improved, enhancing the timeliness and non-destructive nature of regional-scale monitoring. This provides reliable technical support for potato growth assessment, precise nitrogen fertilizer control, and precision agricultural management. Its key technical points are as follows: 1. Spectral-Texture Dual Feature Collaborative Optimization Mechanism. This invention constructs a joint feature system integrating vegetation index and texture features in the control module, and performs step-by-step screening and combination optimization of features through Pearson correlation analysis, maximum correlation minimum redundancy (mRMR) method, and genetic algorithm (GA), thereby reducing feature redundancy, enhancing the correlation between features and the relative chlorophyll content of potatoes, improving the effectiveness and representativeness of input features, and thus improving the accuracy, stability, and generalization ability of the inversion model.

[0022] 2. A multi-model adaptive inversion mechanism for multiple growth stages. This invention sets up a candidate model library and a model selection and switching unit in the control module. By integrating multiple machine learning models such as Lasso, SVR, XGBoost, and GPR, it uses information such as growth stage identification results, model validation errors, uncertainty measures, and online residuals as the basis for model evaluation. It comprehensively assesses the competence of different candidate models and dynamically selects or merges the current optimal inversion model. This mechanism can adapt to the differences in growth at different potato growth stages, achieve adaptive matching of inversion models, and improve the accuracy and adaptability of chlorophyll estimation results under different temporal and plot conditions.

[0023] 3. Visualization of monitoring results and support for agricultural decision-making. The output module of this invention can generate chlorophyll spatial distribution maps, multi-temporal growth change maps, and overlay results of growth period information, realizing intuitive expression and dynamic monitoring of potato population growth; at the same time, it can also output model switching records, estimation confidence levels, and abnormal area identification results, providing visualized and quantitative technical support for potato growth assessment, nitrogen fertilizer regulation, field management, and precision agricultural decision-making.

[0024] Furthermore, by simultaneously receiving UAV multispectral imagery, ground-based SPAD measurement data, and agronomic phenological information through the input module, the integrity of the monitoring data and its adaptability to different growth stages are ensured. The control module performs collaborative optimization of spectral and texture features from multiple data sources, and adaptively selects and switches models for different growth stages, achieving accurate estimation of relative chlorophyll content and dynamic characterization of growth information. The output module generates chlorophyll spatial distribution maps, multi-temporal growth change maps, and results of growth stage identification and model switching, enabling intuitive display and dynamic monitoring of potato population growth. Compared with traditional methods relying on SPAD point measurements, this invention offers advantages such as non-destructive, rapid, and continuous monitoring at the regional scale, improving the accuracy, stability, and model generalization ability of chlorophyll estimation. It can provide reliable technical support for precision fertilization, nitrogen regulation, and growth monitoring, and has significant scientific research application value and prospects. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of a potato growth detection system based on UAV multispectral imagery according to a specific embodiment of the present invention.

[0026] Figure 2 This is a schematic diagram of a feature extraction method in a potato growth detection system based on UAV multispectral imagery according to a specific embodiment of the present invention.

[0027] Figure 3 This is a schematic diagram of UAV multispectral image data processing in a potato growth detection system based on UAV multispectral imagery according to a specific embodiment of the present invention. Detailed Implementation

[0028] The present invention will now be described in detail with reference to the accompanying drawings.

[0029] The following detailed exemplary embodiments are disclosed. However, the specific structural and functional details disclosed herein are merely for the purpose of describing exemplary embodiments.

[0030] However, it should be understood that the present invention is not limited to the specific exemplary embodiments disclosed, but covers all modifications, equivalents, and substitutions falling within the scope of this disclosure. Throughout the description of the drawings, the same reference numerals denote the same elements.

[0031] Referring to the accompanying drawings, the structures, proportions, sizes, etc., depicted in the drawings are merely for illustrative purposes to aid those skilled in the art in understanding and reading the content disclosed herein. They are not intended to limit the conditions under which the invention can be implemented and therefore have no substantial technical significance. Any modifications to the structure, changes in proportions, or adjustments to the size, without affecting the effects and objectives achieved by the invention, should still fall within the scope of the technical content disclosed herein. Furthermore, the positional limitations used in this specification are merely for clarity of description and are not intended to limit the scope of the invention. Changes or adjustments to their relative relationships, without substantially altering the technical content, should also be considered within the scope of the invention's implementation.

[0032] It should also be understood that the term “and / or” as used herein includes any and all combinations of one or more of the related listed items. Furthermore, it should be understood that when a component or unit is referred to as “connected” or “coupled” to another component or unit, it may be directly connected or coupled to the other component or unit, or there may be intermediate components or units. In addition, other words used to describe the relationship between components or units should be understood in the same manner (e.g., “between” versus “directly between,” “adjacent” versus “directly adjacent,” etc.).

[0033] Figure 1 This is a schematic diagram of a potato growth detection system based on UAV multispectral imagery according to a specific embodiment of the present invention. Figure 1 As shown in the figure, a potato growth monitoring system based on UAV multispectral imaging is disclosed in a specific embodiment of the present invention, including an input module, a control module, and an output module. The input module is used to collect and receive multi-source data; the control module is used to preprocess the input multi-source data, extract and collaboratively optimize spectral and texture features, and adaptively select, dynamically switch, and fuse candidate models based on the growth stage identification results to generate a spatial distribution map of potato chlorophyll; the output module is used to output the monitoring results. Specific implementation steps are shown below: (1) Step 1: Input data acquisition

[0034] In this embodiment, the input module receives multi-source data including potato canopy UAV multispectral imagery, radiometric correction plate parameters, aerial flight and exposure parameters, measured ground SPAD values, and agronomic phenological information. The UAV multispectral imagery is used to characterize the remote sensing response characteristics of the potato canopy at different growth stages; the measured ground SPAD values ​​are used for model training, validation, and accuracy correction; the agronomic phenological information includes planting date, accumulated temperature, canopy coverage, and growth stage thresholds, used to assist in identifying the current growth stage and support adaptive model selection. This step, by jointly inputting remote sensing imagery, ground-measured information, and agronomic phenological information into the system, can simultaneously reflect the spectral differences, structural differences, and growth stage differences of the potato canopy, providing a complete data foundation for subsequent chlorophyll relative content inversion.

[0035] (2) Step 2: Multispectral image preprocessing

[0036] The control module first performs image preprocessing, feature extraction, and feature optimization on the UAV multispectral image data, such as... Figure 3 As shown. Specifically, image preprocessing includes image stitching, correction, and cropping operations for each band, which can reduce positional deviations and radiometric differences between images of different bands, improve the comparability of image data from different time phases and different plots, and provide standardized input for subsequent feature extraction. Among these, radiometric correction corrects the orthorectified image according to the parameters of the radiometric correction plate to obtain canopy reflectance data for subsequent analysis. The radiometric correction formula is: , in, The pixel brightness values ​​of the calibration board obtained from the remote sensing image of the calibration board captured by the UAV. To calibrate the reflectivity of the plate, The DN values ​​of potato canopy obtained from drone photography of the test area. The reflectance of the potato canopy.

[0037] (3) Step 3: Feature extraction

[0038] After image preprocessing, the control module performs feature extraction on the potato canopy image. This feature extraction includes two parts: spectral feature extraction and texture feature extraction. Specifically, the spectral feature extraction unit extracts vegetation index features from the multispectral orthophoto image that characterize the spectral response of the potato canopy. The vegetation index calculation formula is shown in the table below.

[0039]

[0040] in, , , , These represent the reflectance of the red band, green band, near-infrared band, and red-edge band, respectively.

[0041] The texture feature extraction unit extracts eight texture feature parameters based on the gray-level co-occurrence matrix (GLCM): mean (MEA), variance (VAR), homogeneity (HOM), angular second moment (ASM), contrast (CON), dissimilarity (DIS), entropy (ENT), and correlation (COR), to characterize the structural features and spatial heterogeneity of the potato canopy. Subsequently, the feature fusion unit fuses the extracted spectral and texture features to construct an initial joint feature set. By simultaneously utilizing spectral and texture information, the system's comprehensive ability to characterize differences in potato growth is enhanced.

[0042] (4) Step 4: Joint feature selection and collaborative optimization Since the initial joint feature set usually has problems such as high dimensionality, strong correlation between features and high redundancy, this implementation method sets a feature screening and optimization unit in the control module to perform step-by-step screening and combination optimization of the joint feature set.

[0043] First, the correlation analysis sub-unit uses Pearson correlation analysis to assess the correlation between each feature and the ground SPAD value. The formula for calculating the Pearson correlation coefficient is as follows: , in, Let x be the covariance of variables x and y. Let x be the standard deviation of the variable. Let y be the standard deviation of the variable y.

[0044] Subsequently, the maximum correlation minimum redundancy subunit uses the maximum correlation minimum redundancy method to further screen the retained features, reducing redundancy between features while maintaining high correlation between features and the target variable. Its evaluation objective can be expressed as: , , , Where S is the candidate feature subset, Representation of features Mutual information with the target variable y This step represents the mutual information between features. It yields a subset of candidate features with strong complementary information.

[0045] Building upon this, the genetic algorithm optimization subunit further performs combinatorial optimization search on the candidate feature subsets. Specifically, the candidate feature subsets are encoded as chromosomes, and the optimal feature combination is iteratively searched through selection, crossover, and mutation operations. The fitness function is then used as the performance of the inversion model. The fitness function can be defined as: , Wherein, RMSE is the root mean square error of the inversion model constructed using the current feature subset on the validation set. To prevent the use of tiny constants with a denominator of zero, the above-described step-by-step screening and optimization process yields a feature subset with optimal overall performance.

[0046] (5) Step 5: Construction of candidate inversion model The optimal feature subset output by the feature selection and optimization unit is used as model input and fed into the model building unit to construct a potato chlorophyll relative content inversion model, establishing a mapping relationship between spectral-texture joint features and SPAD values. In this embodiment, the model building unit can construct multiple candidate regression models, including SVR regression models, Lasso regression models, RF regression models, XGBoost regression models, and GPR regression models. After training, each candidate model is stored in the candidate model library, and its training parameters at different growth stages are saved for subsequent runtime use.

[0047] During model training, ground-based SPAD measurement data are used as supervision information, and the coefficient of determination can be employed. The root mean square error (RMSE) and relative root mean square error (rRMSE) are used to evaluate model performance. The calculation formulas are as follows: , , .

[0048] (6) Step 6: Reproductive period identification To enhance model adaptability under different growth stages, the stage identifier in the control module determines the current growth stage of the potato based on the spectral features, texture features, and agronomic phenological information extracted from the current image, and generates a growth stage label. The growth stage label can be jointly determined by the planting date, accumulated temperature, canopy coverage, and temporal image features.

[0049] (7) Step 7: Model adaptive selection, dynamic switching and policy fusion During system operation, the performance evaluator assesses the current suitability of each candidate model based on its historical validation error and online residual statistics; the confidence evaluator outputs a measure of the uncertainty of the current prediction results for each candidate model; and the switching controller performs a comprehensive competency score on each candidate model based on phenological consistency index, model validation error, uncertainty index, and online residual index, and selects the current optimal model. The formula for calculating the comprehensive competency score is as follows: , in, The phenological consistency score is used to reflect the degree of matching between the model and the current reproductive period; This represents the normalized cross-validation error of the model during the training phase. This is a normalized value for uncertainty; For the most recent window's online relative RMSE; , , and The weighting coefficients are set based on the cross-validation results and error sensitivity analysis of historical samples, and are corrected as adjustable parameters under different plots or growth stages. The optimal model is determined as follows: .

[0050] When the score difference among multiple candidate models is less than a predetermined threshold τ, the fusion unit performs a confidence-weighted fusion of the outputs of the multiple candidate models. Let the set of models participating in the fusion be Ω, then the fusion weight of the m-th model is: .

[0051] (8) Step 8: Generation and output of chlorophyll spatial distribution map After obtaining the current optimal model or fusion model, the control module estimates the relative chlorophyll content of the target area pixel-by-pixel or region-by-region based on the model, generating a spatial distribution map of potato chlorophyll. The output module displays, stores, and exports the generated monitoring results, including the chlorophyll spatial distribution map, multi-temporal growth change map, growth stage identification results, model switching records, estimation confidence levels, and abnormal area identification results. Through this output module, the spatial distribution and temporal variation characteristics of potato population growth can be intuitively displayed, providing a visual basis for growth assessment, nitrogen fertilizer regulation, and field management.

[0052] Through the above eight steps, this invention achieves a complete monitoring process from multi-source data input, image preprocessing, feature co-optimization, model adaptive selection to chlorophyll spatial distribution map output. Compared with the traditional method of relying on SPAD instruments for point measurement, this embodiment can achieve rapid, non-destructive, and continuous regional-scale monitoring of the relative chlorophyll content of potatoes at different growth stages.

[0053] The foregoing description illustrates and describes several preferred embodiments of the present invention. However, as mentioned above, it should be understood that the present invention is not limited to the forms disclosed in this specification and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the inventive concept described in this specification through the foregoing teachings or techniques or knowledge in related fields. Any modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A potato growth monitoring system based on UAV multispectral imaging, comprising an input module, a control module, and an output module, characterized in that, The input module is used to collect and receive data from multiple sources; The control module is used to preprocess the input multi-source data, extract and co-optimize spectral and texture features, and combine the results of the growth period identification to adaptively select, dynamically switch and fuse candidate models to generate a spatial distribution map of potato chlorophyll. The output module is used to output the monitoring results.

2. The potato growth monitoring system based on UAV multispectral imaging as described in claim 1, characterized in that, The multi-source data includes multispectral images of potato canopy from UAVs, radiation correction plate parameters, aerial flight parameters and exposure parameters, measured ground SPAD values, and agronomic phenological information.

3. The potato growth monitoring system based on UAV multispectral imaging as described in claim 2, characterized in that, The control module includes a preprocessing unit for preprocessing UAV multispectral canopy images from the input multi-source data. The preprocessing includes: The multispectral images from the UAV are registered and stitched together across different bands, and then cropped according to the target area to obtain an orthorectified image in a unified coordinate system. Radiometric correction is then performed on the orthorectified image using a radiometric correction plate; the radiometric correction formula is as follows: ， in, The pixel brightness values ​​of the calibration board obtained from the remote sensing image of the calibration board captured by the UAV. To calibrate the reflectivity of the plate, The DN values ​​of potato canopy obtained from drone photography of the test area. The reflectance of the potato canopy.

4. The potato growth monitoring system based on UAV multispectral imaging as described in claim 3, characterized in that, The control module includes a spectral feature extraction unit, a texture feature extraction unit, and a feature fusion unit. The spectral feature extraction unit is used to extract vegetation indices from orthophotos to characterize the spectral response of the potato canopy. The texture feature extraction unit is used to extract texture features based on the gray-level co-occurrence matrix from orthophotos to characterize canopy structure features and spatial heterogeneity. The feature fusion unit is used to fuse the spectral response features and texture features to construct an initial joint feature set.

5. The potato growth monitoring system based on UAV multispectral imaging as described in claim 4, characterized in that, The control module includes a feature selection and optimization unit and a model building unit. The feature selection and optimization unit comprises a correlation analysis subunit, a maximum correlation and minimum redundancy subunit, and a genetic algorithm optimization subunit. The correlation analysis subunit uses Pearson analysis to evaluate the correlation between each feature in the initial joint feature set and the measured SPAD values ​​on the ground, eliminating features with correlation below a predetermined threshold. The maximum correlation and minimum redundancy unit reduces information redundancy among features, obtaining a subset of candidate features with strong information complementarity. The genetic algorithm unit performs combinatorial optimization search on the candidate feature subsets, determining the optimal feature subset based on the performance of the inversion model. The model building unit constructs a chlorophyll relative content inversion model, establishing a mapping relationship between spectral-texture features and SPAD values.

6. The potato growth monitoring system based on UAV multispectral imaging as described in claim 4, characterized in that, The control module includes a model selection and switching unit, which comprises a candidate model library, a stage identifier, a performance evaluator, a confidence evaluator, a switching controller, and a fusion unit. The candidate model library pre-configures regression models such as SVR, Lasso, XGBoost, GPR, and RF, and stores their phased training parameters; The stage identifier is used to determine the current potato growth stage based on the spectral response characteristics, texture characteristics, and agronomic phenological information, and generate a growth period label; The performance evaluator is used to assess the current fit of each candidate model based on the candidate model’s historical validation error and online residual statistics. The confidence estimator is used to output a measure of the uncertainty of each model in the current prediction results; The switching controller is used to calculate the comprehensive competency score of each candidate model and select the optimal model. The fusion unit is used to perform confidence-weighted fusion of the outputs of multiple candidate models when the score difference between multiple candidate models is less than a predetermined threshold.

7. The potato growth monitoring system based on UAV multispectral imaging as described in claim 6, characterized in that, The switching controller is used to calculate the overall competence score of each candidate model based on phenological consistency index, model validation error, uncertainty index, and online residual index, and selects the candidate model with the highest overall competence score as the current optimal model. The formula for calculating the overall competence score is as follows: ， in, The phenological consistency score is used to reflect the degree of matching between the model and the current reproductive period; This represents the normalized cross-validation error of the model during the training phase. This is a normalized value for uncertainty; For the most recent window's online relative RMSE; , , and The weighting coefficients are set based on the cross-validation results and error sensitivity analysis of historical samples, and are corrected as adjustable parameters under different plots or growth stages. The method for determining the optimal model is as follows: 。 8. The potato growth monitoring system based on UAV multispectral imaging as described in claim 6, characterized in that, The model selection and switching unit is equipped with a model switching trigger condition, which is used to constrain the difference in comprehensive competence score between the candidate optimal model and the current model. When the score difference is greater than a preset hysteresis threshold for k consecutive time phases or k consecutive frames, the control module switches from the current model to the candidate optimal model.

9. A method for monitoring potato growth based on UAV multispectral imaging, characterized in that, The method includes the following steps: A. Collect and receive multi-source data, including multispectral images of potato canopy taken by UAV, radiometric correction parameters, aerial flight and exposure parameters, measured SPAD values, and agronomic phenological information. SPAD values ​​are an index characterizing the relative chlorophyll content of potato leaves, used to construct and calibrate the potato relative chlorophyll content inversion model; agronomic phenological information includes planting date, accumulated temperature, canopy coverage, and growth stage thresholds. B. Preprocess the input multi-source data, extract and co-optimize spectral and texture features, and combine the results of the fertility period identification to adaptively select, dynamically switch and fuse candidate models to generate a spatial distribution map of potato chlorophyll. C. Output monitoring results.

10. The method for monitoring potato growth based on UAV multispectral imaging as described in claim 9, characterized in that, The input multi-source data is preprocessed, spectral and texture features are extracted and co-optimized, and candidate models are adaptively selected, dynamically switched and fused based on the fertility period identification results to generate a spatial distribution map of potato chlorophyll, including: Preprocessing involves registering and stitching the multispectral images of the potato canopy from the UAV, and cropping them according to the target area to obtain an orthorectified image with unified coordinates and complete coverage. Radiometric correction is then performed on the orthorectified image using a radiometric correction plate. The radiometric correction method is as follows: ， in, The pixel brightness values ​​of the calibration board obtained from the remote sensing image of the calibration board captured by the UAV. To calibrate the reflectivity of the plate, The DN values ​​of potato canopy obtained from drone photography of the test area. The reflectance of the potato canopy; Spectral and texture feature extraction involves extracting vegetation index features that characterize the spectral response of potato canopy from the orthophoto image and extracting texture features based on the gray-level co-occurrence matrix to characterize canopy structure features and spatial heterogeneity. Feature co-optimization involves fusing the vegetation index features and texture features to construct an initial joint feature set, and then filtering the initial joint feature set through correlation analysis, maximum correlation and minimum redundancy processing, and genetic algorithm combination optimization to obtain the optimal feature subset. The adaptive selection, dynamic switching, and strategy fusion of models involve comprehensively scoring each candidate model based on its historical validation error, the uncertainty of the current prediction results, and the online residual statistics, and dynamically selecting the current optimal model; when the score difference between multiple candidate models is less than a predetermined threshold, the output results of multiple candidate models are weighted and fused. Generating a spatial distribution map of potato chlorophyll involves estimating the relative chlorophyll content in the target area using the selected optimal model or fusion model, generating a spatial distribution map of potato chlorophyll, and outputting the monitoring results.

Images