A method for detecting the growth state of fritillaria cirrhosa under forest cultivation

Abstract

This application discloses a method for detecting the growth status of Fritillaria cirrhosa under forest cultivation. By constructing a multi-source data fusion architecture, it deeply couples physiological and electrical characteristics with optical features, breaking through the technical bottleneck of spectral aliasing and texture similarity in complex forest backgrounds. By introducing spatial geometric constraints and probabilistic graphical model reasoning, it elevates discrete pixel analysis to object-level understanding with cultivation semantics, effectively solving the problems of high false alarm rate and ambiguous positioning caused by interference from forest cover. By establishing a complete technical chain from data acquisition, feature extraction, semantic optimization to prevention and control decision-making, it realizes the transformation of Fritillaria cirrhosa growth status from passive discovery to active early warning, and from experience judgment to quantitative assessment, improving the accuracy of early disease identification and spatial positioning precision.

Description

Technical Field

[0001] This application relates to the field of data detection technology, and in particular to a method for detecting the growth status of Fritillaria cirrhosa under forest cultivation. Background Technology

[0002] Fritillaria cirrhosa, a high-value and precious traditional Chinese medicine, is cultivated under strict wild-like environmental control with a canopy closure of 0.5 to 0.7. Under this ecological niche constraint, physiological state detection technology based on machine vision has become a core means of assessing its growth and health status, especially in the field of early disease diagnosis. For yellowing and brown spot symptoms caused by soil-borne diseases such as damping-off and root rot, image recognition algorithms can accurately extract and classify diseased leaves, forming a key technological link in the current intelligent detection of Fritillaria cirrhosa's growth status.

[0003] However, existing threshold segmentation methods based on hue saturation color space have limitations in practical applications. Because the spectral characteristics of ground cover such as fallen leaves, moss, and humus soil in the understory cultivation environment highly overlap with those of diseased leaves of Fritillaria cirrhosa, the algorithm suffers severe spectral confusion during disease identification. On the one hand, yellowed leaves, due to their hue value shifting towards the yellow end, are similar in hue to semi-decomposed fallen leaves, making them easily misjudged as background and resulting in missed detections. On the other hand, the brown characteristics of brown spot lesions are difficult to distinguish from bark fragments and exposed soil, causing diseased leaves to be incorrectly filtered out, hindering early warning of damping-off and root rot. This identification failure caused by the complex understory background and the similar-colored but different characteristics of disease symptoms severely restricts the technical realization of precise disease control in Fritillaria cirrhosa cultivation. Summary of the Invention

[0004] The main objective of this application is to provide a method for detecting the growth status of Fritillaria cirrhosa under forest cultivation, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, this application provides the following technical solution: A method for detecting the growth status of Fritillaria cirrhosa cultivated under forest cover, the specific steps of which are as follows: S1. Regional monitoring deployment: Divide the cultivation area into n detection areas, deploy sensor groups to collect multi-source data, and generate multispectral image parameter set, ambient light parameter set, physiological verification parameter set, spatial geometric parameter set and ground object spectral characteristic parameter set; The multispectral image parameter set includes: visible light image data, grayscale data after illumination normalization, and near-infrared channel data; The ambient light parameter set includes: spectral power distribution data and photosynthetically active radiation data; The physiological verification parameter set includes: capacitance data and conductivity data; The spatial geometric parameter set includes: depth point cloud data, plant spatial coordinate data, and prior data on cultivation density; The set of spectral feature parameters of ground features includes: spectral response feature data of diseased leaves, texture benchmark difference data of bark and soil, and scale feature data of moss and dead branches and fallen leaves in the forest. S2. Diseased leaf feature extraction: Convert the visible light image to the Lab color space, extract the yellowed diseased leaf area based on the blue-yellow axis chromaticity adaptive threshold, construct a hybrid probability model to optimize recognition, extract hue saturation features, and generate a set of disease-sensitive feature parameters; S3. Texture morphology optimization: Based on near-infrared data, a gray-scale spatial relationship matrix is ​​constructed to extract texture statistics. Color features are fused to distinguish brown spots from the background. Superpixel units are generated based on depth point clouds. The edges of the spots are optimized through morphological filtering to generate a texture morphology optimization parameter set. S4. Spatial semantic optimization: Based on the disease-sensitive feature parameter set and texture morphology optimization parameter set, an initial segmentation mask is constructed. A conditional random field model is established to integrate cultivation density to construct position potential energy and leaf geometry to construct shape potential energy. Capacitance value constraints are introduced to optimize labels and generate an accurate disease identification parameter set. S5. Prevention and Control Decision: Calculate the severity of diseases based on the accurate disease identification parameter set, generate targeted pesticide application and irrigation suggestions, and transmit them to the operating equipment.

[0006] Preferably, in step S2, the specific method for extracting the features of diseased leaves is as follows: S2.1. Convert the visible light data in the multispectral image parameter set to the Lab color space, extract the blue-yellow axis chromaticity components and calculate their mean and standard deviation to construct an adaptive threshold, extract pixels with blue-yellow axis chromaticity values ​​greater than the threshold as the initial candidate regions for yellowing diseased leaves, and simultaneously extract the diseased leaf clustering features of hue saturation joint distribution. S2.2 Construct a mixed probability model of two-dimensional chromaticity of green-red axis and blue-yellow axis, estimate model parameters through iterative algorithm and calculate the posterior probability of each pixel belonging to yellow leaf disease, remove low confidence pixels in the initial candidate area to optimize the recognition results, and integrate hue saturation clustering features and probability model parameters to generate a set of disease-sensitive feature parameters.

[0007] Preferably, step S2.1 is performed as follows: S2.11. Based on the spectral power distribution and photosynthetically active radiation data of the ambient light parameter set, the multispectral image parameter set is normalized and corrected and mapped to the Lab color space. The blue-yellow axis and green-red axis chromaticity components are extracted. The capacitance and conductivity data of the physiological verification parameter set are combined to distinguish between disease chlorosis and physiological water deficiency chlorosis. An adaptive chromaticity threshold model that integrates physiological criteria is constructed. Pixels that meet the threshold and conform to the prior distribution of cultivation density in the spatial geometric parameter set are extracted to form the initial candidate region for chlorotic leaves. S2.12. Based on the plant coordinates and depth point cloud data of the spatial geometric parameter set, a three-dimensional region of interest is delineated. The multispectral image parameter set is mapped to the hue saturation color space to construct a two-dimensional joint distribution. Combining the spectral response characteristics of diseased leaves from the ground object spectral feature parameter set, diseased leaf cluster centers and boundary features that conform to the cultivation space pattern are extracted within the three-dimensional region of interest. Hue saturation diseased leaf feature parameters that match the subsequent spatial context constraints are generated.

[0008] Preferably, step S2.2 is performed as follows: S2.21. Combine the chromaticity data of the green-red axis and the blue-yellow axis with the capacitance value distribution of the physiological verification parameter set, construct a dynamic probability model to distinguish between disease-induced yellowing and physiological water loss, optimize its parameter estimation, calculate the posterior probability of a pixel belonging to the diseased leaf category, and combine the prior of the cultivation density of the spatial geometric parameter set to remove isolated pixels that deviate from the plant distribution pattern, and generate a physiologically verified high-confidence yellowing diseased leaf identification region. S2.22. Construct a two-dimensional joint distribution of hue saturation to extract the cluster center and boundary features of diseased leaves. Integrate the optimized yellowed leaf regions, probability distribution parameters and cluster features. Dynamically adjust the feature weights based on the environmental light parameter set. Combine the ground object spectral feature parameter set to construct a multi-dimensional joint descriptor that is compatible with subsequent texture analysis and spatial context constraints, and generate a disease-sensitive feature parameter set.

[0009] Preferably, in step S3, the specific method for texture morphology optimization is as follows; S3.1. Targeting the unique parallel distribution of leaf veins and heterogeneous texture pattern of necrotic tissue in the brown spot lesions of the stem-clasping strip leaves of Fritillaria cirrhosa, a pixel-level gray-scale spatial relationship matrix is ​​constructed by calling the near-infrared channel data of the multispectral image parameter set and the spatial distribution of capacitance values ​​of the physiological verification parameter set. The surface roughness and anisotropy index, which characterize the degree of necrosis of lesion tissue and the difference in physiological water loss, are extracted. A color and texture joint feature descriptor is constructed by integrating the disease sensitive feature parameter set. Based on the ground spectral feature parameter set, brown spot lesions with similar hues but heterogeneous texture structure are identified from the background, and physiologically verified texture-constrained lesion candidate masks are generated. S3.2. Based on the surface normal vector of the depth point cloud of the spatial geometric parameter set, estimate the surface geometry of the stem-holding leaf, generate superpixel units that fit the bending shape of the leaf, and establish a spatial adjacency graph. Statistically measure the spatial distribution consistency of the joint features of color and texture within the unit. Adaptively construct a multi-scale morphological opening and closing operation structure element sequence based on the scale features of forest moss and dead leaves and the abnormal distribution of capacitance values ​​in the ground cover spectral feature parameter set. Cascade filter out small background interference and restore the continuity of brown spot disease edge. Construct a texture morphology optimization parameter set that is compatible with subsequent spatial context constraints.

[0010] Preferably, step S3.1 is performed as follows: S3.11. Call the near-infrared channel data of the multispectral image parameter set, combine it with the spatial distribution of capacitance values ​​of the physiological verification parameter set to construct a pixel-level gray-scale spatial relationship matrix, correct the texture directionality index according to the capacitance value gradient, extract the surface roughness and adaptive directionality index that characterize the difference between necrotic lesion tissue and physiological water loss, and establish a texture feature mapping that integrates physiological criteria. S3.12. Map the color features and texture index of the disease sensitive feature parameter set to the joint feature space. Based on the difference in texture benchmark between bark and soil in the ground object spectral feature parameter set, identify brown spots with similar hues but heterogeneous texture structures from the background, and generate texture-constrained spot candidate masks that are physiologically verified and adapted to subsequent spatial context constraints.

[0011] Preferably, step S3.2 is performed as follows: S3.21. Based on the spatial geometric parameter set, the differential geometry of the stem-holding leaf surface is analyzed by deep point cloud analysis. Adaptive superpixel units that fit the curvature change of the leaf are constructed and topological connections are established. The spatial distribution consistency of color and texture features within the unit is statistically analyzed to form a surface feature descriptor that is compatible with subsequent conditional random fields. S3.22. Based on the joint constraints of the small-scale features of the ground object spectral feature parameter set and the abnormal distribution of capacitance values ​​in the physiological verification parameter set, an adaptive multi-scale morphological structure element sequence is constructed to restore the continuity of brown spot lesion edges and filter out background interference, generating a texture morphology optimization parameter set coupled with spatial context constraints.

[0012] Preferably, in step S4, the spatial semantic optimization is performed in the following specific way: S4.1. Based on the posterior probability of the disease-sensitive feature parameter set and the region labeling of the texture morphology optimization parameter set, an initial configuration is constructed. A fully connected conditional random field model is established to define a univariate potential energy term for the fusion of color and texture confidence. Based on the cultivation density prior of the spatial geometric parameter set, a position potential energy function is constructed to penalize the spatial inconsistency of deviation from the norm grid. Based on the strip geometry of the stem-clamping leaf analyzed by the depth point cloud, a shape potential energy function is constructed to constrain the directional continuity of adjacent nodes, forming a joint potential energy model that integrates cultivation norms and morphological priors. S4.2. The message passing algorithm is used to iteratively optimize the node label allocation. The capacitance value of the physiological verification parameter set is introduced to construct the physiological activity constraint term. The non-living area labels with capacitance values ​​lower than the physiological activity threshold are forcibly corrected to suppress background misclassification. The accurate disease area markers are generated through iterative convergence. The severity level is calculated based on the proportion of diseased leaf pixels and the proportion of lesion area based on the optimized node labels. The spatial geometric parameter set of plant coordinates is integrated to locate the spatial distribution of the disease, forming a disease identification parameter set that includes disease category identifiers, severity quantification values ​​and spatial coordinates.

[0013] Preferably, step S4.1 is performed as follows: S4.11. By fusing the surface adaptive superpixel unit in the texture morphology optimization parameter set with the depth point cloud surface normal vector data in the spatial geometry parameter set, a geodesic node set that fits the bending morphology of the stem-clasping leaf is constructed. The initial weight configuration of the nodes is dynamically adjusted according to the ground object spectral feature parameter set and the physiological verification parameter set to form a graph node topology structure that is adapted to the plant surface geometry. S4.12. Analyze the spatial geometric parameter set of cultivation density prior and depth point cloud surface differential geometric features, construct a position potential energy function to penalize deviations from the standard planting grid spatial distribution, extract the direction of the stem-holding leaf strip geometric principal axis to construct an anisotropic shape potential energy function to constrain the continuity of the surface orientation, and couple them to form a joint potential energy model that integrates cultivation norms and morphological priors.

[0014] Preferably, step S4.2 is performed as follows: S4.21. Based on the mean field approximation, the message passing iterative optimization of node label configuration is implemented. The capacitance value of the physiological verification parameter set is integrated to construct a complementary evidence item of physiological activity and mapped to an energy correction function. According to the distribution characteristics of capacitance value below the physiological activity threshold, the labels of non-living areas are forcibly corrected to remove the diseased leaf category. The misclassification of background ground features with similar spectral characteristics but abnormal capacitance value is suppressed. The high-confidence accurate disease area markers are generated iteratively and converged. S4.22. Based on the optimized and converged node labels, the percentage of diseased leaf pixels and the percentage of diseased area are statistically analyzed and mapped to the plant-level diseased leaf rate and severity level index. The spatial geometric parameter set of plant coordinates and cultivation density priors are integrated to construct the disease spatial distribution topology, forming a disease identification parameter set containing disease category identifiers, severity quantification values ​​and precise spatial coordinates to support subsequent variable control decisions.

[0015] Compared with the prior art, the beneficial effects of the present invention are: 1. By constructing a multi-source data fusion architecture, physiological and electrical characteristics are deeply coupled with optical features, breaking through the technical bottleneck of spectral aliasing and texture similarity in complex forest backgrounds; by introducing spatial geometric constraints and probabilistic graphical model reasoning, discrete pixel analysis is elevated to object-level understanding with cultivation semantics, effectively solving the problems of high false alarm rate and ambiguous positioning caused by forest cover interference; by establishing a complete technical chain from data collection, feature extraction, semantic optimization to prevention and control decision-making, the transformation of Fritillaria cirrhosa growth status from passive discovery to active early warning, and from experience judgment to quantitative assessment has been realized, improving the accuracy of early disease identification and spatial positioning precision.

[0016] 2. By introducing a mechanism that integrates ambient light normalization correction and physiological electrical parameters, this method effectively overcomes the limitations of relying solely on visual criteria in complex forest lighting environments, achieving accurate differentiation between disease-related yellowing and physiological dehydration, and reducing the misjudgment rate. Simultaneously, by constructing a three-dimensional region of interest based on deep point clouds and extracting feature parameters adapted to the spatial context, a hierarchical feature extraction system from global monitoring to local analysis is established. This solves the problem of the disconnect between feature extraction and subsequent spatial reasoning in existing technologies, comprehensively improving the accuracy of early identification of yellowing leaves and the continuity of feature data, laying a data foundation for building a high-confidence disease identification model. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the steps of the method described in this application. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0019] The terms "first," "second," and "third" in this application are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first," "second," or "third" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified. All directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of this application are only used to explain the relative positional relationships and movements between components in a specific orientation (as shown in the figures). If the specific orientation changes, the directional indications also change accordingly. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0020] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0021] Example 1: Please refer to Figure 1 A method for detecting the growth status of Fritillaria cirrhosa under forest cultivation, the specific steps of which are as follows: S1. Regional monitoring deployment: Divide the cultivation area into n detection areas, deploy sensor groups to collect multi-source data, and generate multispectral image parameter set, ambient light parameter set, physiological verification parameter set, spatial geometric parameter set and ground object spectral characteristic parameter set; The multispectral image parameter set includes: visible light image data, grayscale data after illumination normalization, and near-infrared channel data; The ambient light parameter set includes: spectral power distribution data and photosynthetically active radiation data; The physiological verification parameter set includes: capacitance data and conductivity data; The spatial geometric parameter set includes: depth point cloud data, plant spatial coordinate data, and prior data on cultivation density; The set of spectral feature parameters of ground features includes: spectral response feature data of diseased leaves, texture benchmark difference data of bark and soil, and scale feature data of moss and dead branches and fallen leaves in the forest. S2. Diseased leaf feature extraction: Convert the visible light image to the Lab color space, extract the yellowed diseased leaf area based on the blue-yellow axis chromaticity adaptive threshold, construct a hybrid probability model to optimize recognition, extract hue saturation features, and generate a set of disease-sensitive feature parameters; S3. Texture morphology optimization: Based on near-infrared data, a gray-scale spatial relationship matrix is ​​constructed to extract texture statistics. Color features are fused to distinguish brown spots from the background. Superpixel units are generated based on depth point clouds. The edges of the spots are optimized through morphological filtering to generate a texture morphology optimization parameter set. S4. Spatial semantic optimization: Based on the disease-sensitive feature parameter set and texture morphology optimization parameter set, an initial segmentation mask is constructed. A conditional random field model is established to integrate cultivation density to construct position potential energy and leaf geometry to construct shape potential energy. Capacitance value constraints are introduced to optimize labels and generate an accurate disease identification parameter set. S5. Prevention and Control Decision: Calculate the severity of diseases based on the accurate disease identification parameter set, generate targeted pesticide application and irrigation suggestions, and transmit them to the operating equipment.

[0022] In this embodiment: Step S1 involves deploying regional monitoring to systematically divide the Fritillaria cirrhosa cultivation area under the forest into several detection units and configure multi-source sensor groups. This enables the synchronous acquisition of multi-dimensional data, including multispectral images, ambient light, physiological verification, spatial geometry, and spectral characteristics of ground objects. This solves the problem of spatiotemporal heterogeneity in data acquisition in complex forest habitats and overcomes the limitations of traditional manual inspection data acquisition, which is one-sided and lacks real-time performance. This establishes a standardized and systematic data foundation for subsequent intelligent analysis.

[0023] Step S2 extracts diseased leaf features, maps the visible light image to the Lab color space, and combines the capacitance value distribution of the physiological verification parameter set to construct an adaptive chromaticity threshold and mixing probability model. This effectively distinguishes the spectral aliasing of yellowed diseased leaves and forest litter, solves the false negative problem caused by relying solely on hue segmentation, realizes early disease identification based on physiological-visual joint verification, and improves the identification accuracy and confidence of yellowed diseased leaves.

[0024] Step S3 optimizes the texture morphology by constructing a grayscale spatial relationship matrix using near-infrared channel data to extract texture statistics, constructing a joint feature space by combining color features, and generating superpixel units that fit the geometry of the plant surface based on the depth point cloud. The edges of the lesions are optimized by multi-scale morphological filtering, which effectively solves the dilemma that brown spot lesions are difficult to distinguish from bark and soil in color space, and achieves accurate preservation of lesion boundaries and optimized reconstruction of regional integrity.

[0025] Step S4 optimizes spatial semantics by constructing a conditional random field model and integrating cultivation density priors with leaf geometric features to build a potential energy function. It introduces capacitance values ​​as physiological activity constraints to correct visual segmentation results, effectively suppressing misclassification of background features such as scattered fallen leaves and patches of moss. It solves the interference problem of objects with similar spectra but abnormal spatial distribution, achieving a leap from pixel-level classification to plant-level semantic understanding, and achieving accurate localization and spatial distribution analysis of diseased areas.

[0026] Step S5, through prevention and control decision-making, calculates the disease leaf rate and severity level based on the precise disease identification parameter set, generates a targeted application area map and precise irrigation suggestions, and transmits them to the operating equipment. This achieves closed-loop management from disease identification to variable control, solves the problems of pesticide waste and insufficient control precision caused by traditional uniform application, and achieves the management goals of reducing pesticide use, increasing efficiency, and precision cultivation.

[0027] Compared to existing monitoring methods that rely on single visual sensors or human experience, this solution overcomes the technical bottleneck of spectral aliasing and texture similarity in complex forest backgrounds by constructing a multi-source data fusion architecture that deeply couples physiological and electrical characteristics with optical features. By introducing spatial geometric constraints and probabilistic graphical model reasoning, discrete pixel analysis is elevated to object-level understanding with cultivation semantics, effectively solving the problems of high false alarm rate and ambiguous positioning caused by forest cover interference. By establishing a complete technical chain from data acquisition, feature extraction, semantic optimization to prevention and control decision-making, the solution realizes the transformation of Fritillaria cirrhosa growth status from passive discovery to active early warning, and from experience-based judgment to quantitative assessment, improving the accuracy of early disease identification and spatial positioning precision.

[0028] Example 2: Please refer to Figure 1 In step S2, the specific method for extracting diseased leaf features is as follows: S2.1. Convert the visible light data in the multispectral image parameter set to the Lab color space, extract the blue-yellow axis chromaticity components and calculate their mean and standard deviation to construct an adaptive threshold, extract pixels with blue-yellow axis chromaticity values ​​greater than the threshold as the initial candidate regions for yellowing diseased leaves, and simultaneously extract the diseased leaf clustering features of hue saturation joint distribution. S2.2 Construct a mixed probability model of two-dimensional chromaticity of green-red axis and blue-yellow axis, estimate model parameters through iterative algorithm and calculate the posterior probability of each pixel belonging to yellow leaf disease, remove low confidence pixels in the initial candidate area to optimize the recognition results, and integrate hue saturation clustering features and probability model parameters to generate a set of disease-sensitive feature parameters.

[0029] In this embodiment: Step S2.1 maps visible light data to the Lab color space and analyzes the blue-yellow axis chromaticity components, constructing an adaptive segmentation threshold based on statistical moments. This effectively separates the overlapping areas of yellowed leaves and forest litter in the hue space, forming initial candidate regions. The simultaneously extracted hue saturation joint distribution features provide multi-dimensional data support for subsequent refined identification, solving the screening bias problem caused by insufficient information from a single color channel.

[0030] Step S2.2 constructs a two-dimensional chromaticity mixing probability model and iteratively optimizes parameter estimation to calculate the posterior probability of each pixel belonging to the diseased leaf category, effectively eliminating low-confidence background pixels mixed in the initial candidate region due to spectral similarity. The disease-sensitive feature parameter set generated by fusing hue saturation clustering features achieves a leap from coarse screening to fine screening, reducing the false negative rate caused by traditional threshold segmentation.

[0031] Compared to traditional detection methods that rely solely on hue threshold segmentation, this step effectively addresses the identification failure caused by spectral overlap between yellowed leaves and litter in complex forest backgrounds through a progressive processing architecture involving color space transformation, adaptive threshold construction, and probabilistic model optimization. By introducing a two-dimensional chromaticity joint distribution and iterative optimization mechanism, precise localization and confidence quantification of diseased leaf areas are achieved, overcoming the bottleneck of difficulty in distinguishing similar colored features in existing technologies. This improves the accuracy and robustness of early identification of yellowing diseases, laying a high-fidelity data foundation for subsequent texture analysis and spatial semantic optimization.

[0032] Example 3: Please refer to Figure 1 The specific method for step S2.1 is as follows: S2.11. Based on the spectral power distribution and photosynthetically active radiation data of the ambient light parameter set, the multispectral image parameter set is normalized and corrected and mapped to the Lab color space. The blue-yellow axis and green-red axis chromaticity components are extracted. The capacitance and conductivity data of the physiological verification parameter set are combined to distinguish between disease chlorosis and physiological water deficiency chlorosis. An adaptive chromaticity threshold model that integrates physiological criteria is constructed. Pixels that meet the threshold and conform to the prior distribution of cultivation density in the spatial geometric parameter set are extracted to form the initial candidate region for chlorotic leaves. S2.12. Based on the plant coordinates and depth point cloud data of the spatial geometric parameter set, a three-dimensional region of interest is delineated. The multispectral image parameter set is mapped to the hue saturation color space to construct a two-dimensional joint distribution. Combining the spectral response characteristics of diseased leaves from the ground object spectral feature parameter set, diseased leaf cluster centers and boundary features that conform to the cultivation space pattern are extracted within the three-dimensional region of interest. Hue saturation diseased leaf feature parameters that match the subsequent spatial context constraints are generated.

[0033] In this embodiment: Step S2.11 implements light normalization correction based on the spectral power distribution and photosynthetically active radiation data of the ambient light parameter set, which effectively eliminates the uneven image brightness caused by light spots and shadows in the closed canopy environment. Combined with the capacitance and conductivity data of the physiological verification parameter set, it distinguishes between disease-related yellowing and physiological water-deficiency yellowing, and constructs an adaptive chromaticity threshold model that integrates physiological criteria. This solves the technical problem that it is difficult to distinguish between disease types and physiological stresses using only visual features. At the same time, it uses the prior distribution of cultivation density to remove scattered background pixels that deviate from the standard planting grid, thereby improving the purity and reliability of the initial candidate area.

[0034] Step S2.12 delineates a three-dimensional region of interest that fits the actual spatial distribution of the plant by using plant coordinates and depth point cloud data based on a set of spatial geometric parameters. This achieves precise focusing from the global image to the local plant. Based on this, a two-dimensional joint distribution of hue saturation is constructed and the cluster centers and boundary features of diseased leaves are extracted. This generates hue saturation diseased leaf feature parameters that match the subsequent spatial context constraints, effectively solving the background interference problem caused by traditional global feature extraction. This provides a highly relevant feature data foundation for subsequent texture analysis and semantic optimization.

[0035] Compared to traditional detection methods that rely solely on color threshold segmentation, this step effectively overcomes the limitations of purely visual criteria in complex forest lighting environments by introducing a mechanism that integrates ambient light normalization correction and physiological electrical parameters. This enables accurate differentiation between disease-related yellowing and physiological dehydration, reducing the false positive rate. Simultaneously, by constructing a three-dimensional region of interest based on depth point clouds and extracting feature parameters adapted to the spatial context, a hierarchical feature extraction system from global monitoring to local analysis is established. This solves the problem of disconnect between feature extraction and subsequent spatial reasoning in existing technologies, comprehensively improving the accuracy of early identification of yellowing leaves and the continuity of feature data, laying a data foundation for building a high-confidence disease identification model.

[0036] Example 4: Please refer to Figure 1 The specific method for step S2.2 is as follows: S2.21. Combine the chromaticity data of the green-red axis and the blue-yellow axis with the capacitance value distribution of the physiological verification parameter set, construct a dynamic probability model to distinguish between disease-induced yellowing and physiological water loss, optimize its parameter estimation, calculate the posterior probability of a pixel belonging to the diseased leaf category, and combine the prior of the cultivation density of the spatial geometric parameter set to remove isolated pixels that deviate from the plant distribution pattern, and generate a physiologically verified high-confidence yellowing diseased leaf identification region. S2.22. Construct a two-dimensional joint distribution of hue saturation to extract the cluster center and boundary features of diseased leaves. Integrate the optimized yellowed leaf regions, probability distribution parameters and cluster features. Dynamically adjust the feature weights based on the environmental light parameter set. Combine the ground object spectral feature parameter set to construct a multi-dimensional joint descriptor that is compatible with subsequent texture analysis and spatial context constraints, and generate a disease-sensitive feature parameter set.

[0037] In this embodiment: Step S2.21 constructs a dynamic probability model that integrates physiological criteria by combining the chromaticity data of the green-red axis and the blue-yellow axis with the capacitance value distribution of the physiological verification parameter set, and optimizes the parameter estimation. It effectively calculates the posterior probability of each pixel belonging to the diseased leaf category. Combined with the prior of the cultivation density of the spatial geometric parameter set, it removes isolated background pixels that deviate from the plant distribution pattern, and generates a physiologically verified high-confidence yellowing diseased leaf identification area. This solves the technical problem that it is difficult to distinguish ground objects with similar spectra but different physiological characteristics by simple visual features, and improves the reliability of the identification results.

[0038] Step S2.22 extracts the cluster centers and boundary features of diseased leaves by constructing a two-dimensional joint distribution of hue saturation. It integrates the optimized yellowing leaf regions, probability distribution parameters, and cluster features, dynamically adjusts the feature weights based on the environmental light parameter set, and constructs a multi-dimensional joint descriptor that is compatible with subsequent texture analysis and spatial context constraints by combining the ground object spectral feature parameter set. This generates a disease-sensitive feature parameter set, realizes the effective fusion of multi-source heterogeneous data and feature dimensionality reduction, and solves the problem of one-sided recognition caused by insufficient information in a single feature dimension.

[0039] Compared to traditional detection methods that rely solely on color thresholds or simple probability statistics, this step effectively overcomes the bottleneck in identifying spectrally mixed ground features by constructing a dynamic probability model that integrates physiological and electrical characteristics, achieving accurate differentiation between disease-related yellowing and physiological dehydration. Simultaneously, by constructing a multi-dimensional joint feature descriptor and establishing an adaptation mechanism with subsequent processing steps, a technical closed loop is formed from coarse screening to fine screening and then to feature fusion. This solves the problem of the disconnect between feature extraction and subsequent analysis in existing technologies, improving the accuracy of yellowing leaf identification and the continuity of feature data, and providing high-quality data support for building a highly robust disease identification model.

[0040] Example 5: Please refer to Figure 1 In step S3, the specific method for texture morphology optimization is as follows; S3.1. Targeting the unique parallel distribution of leaf veins and heterogeneous texture pattern of necrotic tissue in the brown spot lesions of the stem-clasping strip leaves of Fritillaria cirrhosa, a pixel-level gray-scale spatial relationship matrix is ​​constructed by calling the near-infrared channel data of the multispectral image parameter set and the spatial distribution of capacitance values ​​of the physiological verification parameter set. The surface roughness and anisotropy index, which characterize the degree of necrosis of lesion tissue and the difference in physiological water loss, are extracted. A color and texture joint feature descriptor is constructed by integrating the disease sensitive feature parameter set. Based on the ground spectral feature parameter set, brown spot lesions with similar hues but heterogeneous texture structure are identified from the background, and physiologically verified texture-constrained lesion candidate masks are generated. S3.2. Based on the surface normal vector of the depth point cloud of the spatial geometric parameter set, estimate the surface geometry of the stem-holding leaf, generate superpixel units that fit the bending shape of the leaf, and establish a spatial adjacency graph. Statistically measure the spatial distribution consistency of the joint features of color and texture within the unit. Adaptively construct a multi-scale morphological opening and closing operation structure element sequence based on the scale features of forest moss and dead leaves and the abnormal distribution of capacitance values ​​in the ground cover spectral feature parameter set. Cascade filter out small background interference and restore the continuity of brown spot disease edge. Construct a texture morphology optimization parameter set that is compatible with subsequent spatial context constraints.

[0041] In this embodiment: Step S3.1 Targeting the unique leaf vein parallelism and heterogeneous texture pattern of necrotic tissue in the brown spot disease of Fritillaria cirrhosa leaves, a gray-scale spatial relationship matrix is ​​constructed by combining near-infrared data and capacitance value distribution. The surface roughness and anisotropy index, which characterize the degree of necrosis and physiological water loss of the disease spots, are extracted. A joint descriptor is constructed by fusing color features, which effectively distinguishes brown spot disease spots with similar hues but heterogeneous texture structure from bark and soil background. Physiologically validated texture-constrained disease spot candidate masks are generated, reducing the risk of misjudgment caused by color-similar ground features.

[0042] Step S3.2: Based on the surface normal vectors of the deep point cloud, the geometric morphology of the leaf surface clasping the stem is analyzed to generate superpixel units that fit the bending characteristics of the leaf and establish spatial adjacency relationships. The spatial distribution consistency of color and texture features within the unit is statistically analyzed. Based on the scale characteristics of forest moss and dead branches and fallen leaves and the abnormal distribution of capacitance values, a multi-scale morphological structural element sequence is adaptively constructed. This effectively filters out small background interference and restores the continuity of brown spot lesion edges. A texture morphology optimization parameter set adapted to subsequent spatial semantic analysis is constructed, which improves the precision and topological integrity of lesion region segmentation.

[0043] Compared to traditional detection methods that rely solely on color thresholds or simple texture statistics, this step effectively overcomes the technical bottleneck of color space overlap between brown spot lesions and bark / soil by constructing a gray-scale spatial relationship matrix that integrates physiological and electrical properties. This enables accurate identification based on joint verification of texture heterogeneity and physiological activity. Simultaneously, by introducing surface geometric constraints from deep point cloud analysis to guide superpixel segmentation and adaptive morphological filtering, the challenges of traditional planar segmentation failing to conform to leaf curvature and fixed-scale filtering easily damaging lesion edges are addressed. This improves the accuracy and boundary preservation of brown spot lesion identification, providing a refined morphological data foundation for building a high-confidence disease identification model.

[0044] Example 6: Please refer to Figure 1 The specific method for step S3.1 is as follows: S3.11. Call the near-infrared channel data of the multispectral image parameter set, combine it with the spatial distribution of capacitance values ​​of the physiological verification parameter set to construct a pixel-level gray-scale spatial relationship matrix, correct the texture directionality index according to the capacitance value gradient, extract the surface roughness and adaptive directionality index that characterize the difference between necrotic lesion tissue and physiological water loss, and establish a texture feature mapping that integrates physiological criteria. S3.12. Map the color features and texture index of the disease sensitive feature parameter set to the joint feature space. Based on the difference in texture benchmark between bark and soil in the ground object spectral feature parameter set, identify brown spots with similar hues but heterogeneous texture structures from the background, and generate texture-constrained spot candidate masks that are physiologically verified and adapted to subsequent spatial context constraints.

[0045] In this embodiment: Step S3.11 constructs a pixel-level grayscale spatial relationship matrix by calling near-infrared channel data and combining it with the spatial distribution of capacitance values ​​in the physiological verification parameter set. Based on the capacitance value gradient, the texture directionality index is corrected, effectively extracting the surface roughness and adaptive directionality index that characterize the difference between lesion tissue necrosis and physiological dehydration. A texture feature mapping that integrates physiological criteria is established, solving the technical problem that simple texture analysis is difficult to distinguish between disease necrosis and physiological stress, and improving the accuracy of texture features in characterizing the true physiological state of lesions.

[0046] Step S3.12 maps the color features and texture index of the disease-sensitive feature parameter set to the joint feature space. Based on the difference in texture benchmarks between bark and soil in the ground object spectral feature parameter set, it effectively distinguishes brown spot lesions with similar hues but heterogeneous texture structures from background ground objects. It generates a texture-constrained lesion candidate mask that has been physiologically verified and is compatible with subsequent spatial context constraints. This achieves a leap from single color segmentation to color-texture fusion discrimination and reduces the false alarm rate caused by traditional color threshold segmentation.

[0047] Compared to traditional detection methods that rely solely on color thresholds or simple texture statistics, this step effectively overcomes the challenge of overlapping lesion tissue necrosis and physiological dehydration in the texture feature space by introducing capacitance gradient correction of the texture directionality index, achieving accurate mapping of texture features based on physiological electrical properties. Simultaneously, by constructing a joint color-texture feature space and establishing an adaptation mechanism with subsequent spatial context constraints, the problem of the disconnect between feature extraction and spatial semantic analysis in existing technologies is solved, improving the distinguishability of brown spot lesions from the bark and soil background, and providing a refined feature data foundation for building a high-confidence disease identification model.

[0048] Example 7: Please refer to Figure 1 The specific method for step S3.2 is as follows: S3.21. Based on the spatial geometric parameter set, the differential geometry of the stem-holding leaf surface is analyzed by deep point cloud analysis. Adaptive superpixel units that fit the curvature change of the leaf are constructed and topological connections are established. The spatial distribution consistency of color and texture features within the unit is statistically analyzed to form a surface feature descriptor that is compatible with subsequent conditional random fields. S3.22. Based on the joint constraints of the small-scale features of the ground object spectral feature parameter set and the abnormal distribution of capacitance values ​​in the physiological verification parameter set, an adaptive multi-scale morphological structure element sequence is constructed to restore the continuity of brown spot lesion edges and filter out background interference, generating a texture morphology optimization parameter set coupled with spatial context constraints.

[0049] In this embodiment: Step S3.21 uses deep point cloud analysis to analyze the differential geometric features of the curved surface of the stem-clasping leaf, constructs an adaptive superpixel unit that fits the actual curvature change of the leaf, and establishes topological connections. This effectively solves the problem of poor edge fit caused by the curved shape of the leaf, which is difficult for traditional planar segmentation algorithms to handle. By statistically analyzing the spatial distribution consistency of color and texture features within the unit, a surface feature descriptor adapted to the subsequent conditional random field is formed, improving the matching degree between the feature description and the plant geometry, and providing a high-precision surface data foundation for spatial semantic optimization.

[0050] Step S3.22 constructs joint constraints based on the scale characteristics and abnormal capacitance distribution of forest moss and fallen leaves, adaptively building a multi-scale morphological structure element sequence. This effectively restores the continuity of brown spot lesion edges and filters out minor background interference, generating a texture morphology optimization parameter set coupled with spatial context constraints. This technical solution overcomes the limitations of fixed-scale morphological filtering, which easily damages lesion edges or cannot effectively filter out multi-scale background interference. It improves the precision and topological integrity of lesion region segmentation, providing high-quality region labeling data for subsequent accurate disease identification.

[0051] Compared to traditional techniques that rely solely on planar image segmentation and fixed-scale morphological filtering, this step effectively overcomes the technical bottleneck of two-dimensional planar analysis's inability to accurately capture the true curvature of clasping leaves by introducing surface differential geometric constraints from deep point cloud analysis. This achieves adaptive superpixel segmentation based on plant surface geometry, improving the accuracy of segmentation unit alignment with leaf boundaries. Simultaneously, by constructing a multi-scale morphological filtering mechanism that integrates ground-scale features and physiological and electrical properties, the contradiction between maintaining lesion edge integrity and filtering out multi-scale background interference in traditional single-scale filtering is resolved, achieving fine-grained lesion edge restoration and effective suppression of background interference. This technical solution comprehensively improves the accuracy and edge preservation of brown spot lesion region segmentation, providing a refined morphological data foundation adapted to plant geometry and physiological state for building high-confidence disease identification models.

[0052] Example 8: Please refer to Figure 1 In step S4, the specific method for spatial semantic optimization is as follows: S4.1. Based on the posterior probability of the disease-sensitive feature parameter set and the region labeling of the texture morphology optimization parameter set, an initial configuration is constructed. A fully connected conditional random field model is established to define a univariate potential energy term for the fusion of color and texture confidence. Based on the cultivation density prior of the spatial geometric parameter set, a position potential energy function is constructed to penalize the spatial inconsistency of deviation from the norm grid. Based on the strip geometry of the stem-clamping leaf analyzed by the depth point cloud, a shape potential energy function is constructed to constrain the directional continuity of adjacent nodes, forming a joint potential energy model that integrates cultivation norms and morphological priors. S4.2. The message passing algorithm is used to iteratively optimize the node label allocation. The capacitance value of the physiological verification parameter set is introduced to construct the physiological activity constraint term. The non-living area labels with capacitance values ​​lower than the physiological activity threshold are forcibly corrected to suppress background misclassification. The accurate disease area markers are generated through iterative convergence. The severity level is calculated based on the proportion of diseased leaf pixels and the proportion of lesion area based on the optimized node labels. The spatial geometric parameter set of plant coordinates is integrated to locate the spatial distribution of the disease, forming a disease identification parameter set that includes disease category identifiers, severity quantification values ​​and spatial coordinates.

[0053] In this embodiment: Step S4.1 constructs an initial configuration by fusing the posterior probability of disease-sensitive features with texture morphology markers, establishes a fully connected conditional random field model and defines a univariate potential energy term that fuses color and texture confidence, constructs a position potential energy function based on the cultivation density prior to penalize spatial distribution anomalies that deviate from the standard planting grid, and constructs a shape potential energy function based on the strip geometric features of stem-hugging leaves obtained from deep point cloud analysis to constrain the directional continuity of adjacent nodes, forming a joint potential energy model that fuses cultivation standards and morphological priors. This effectively suppresses interference from ground features with similar spectral features but scattered spatial distributions, and elevates discrete pixel-level classification to object-level understanding that conforms to cultivation semantics.

[0054] Step S4.2 employs a message passing algorithm to iteratively optimize node labels. It introduces capacitance value data from the physiological verification parameter set to construct physiological activity constraints, forcibly correcting labels of non-living regions with capacitance values ​​below the physiological activity threshold to suppress background misclassification. After iterative convergence, it generates accurate disease region markers. Based on the optimized node labels, it calculates the percentage of diseased leaf pixels and the percentage of lesion area and determines the severity level. It integrates plant coordinates to locate the spatial distribution of diseases, ultimately forming a disease identification parameter set that includes disease category identifiers, severity quantification values, and accurate spatial coordinates. This achieves quantitative assessment and precise spatial positioning of disease severity.

[0055] Compared to traditional detection methods that rely solely on spectral features or simple threshold segmentation, this step effectively overcomes the bottleneck of distinguishing spectrally overlapping ground features in complex forest backgrounds by constructing a conditional random field model that integrates prior knowledge of cultivation density and leaf geometry. This achieves a leap from pixel-level classification to plant-level semantic understanding. By introducing physiological and electrical properties as an activity constraint to correct visual segmentation results, the problem of misclassification of non-living regions caused by relying solely on optical features is solved, improving the accuracy and reliability of disease area labeling.

[0056] This technical solution effectively suppresses interference from background features such as scattered fallen leaves and patches of moss by using spatial context constraints and multi-source evidence fusion. It enables precise location and quantitative assessment of the severity of yellowing and brown spot diseased leaves, improves the accuracy of early disease identification and spatial location precision, and provides high-confidence data support for subsequent variable control decisions.

[0057] Example 9: Please refer to Figure 1 The specific method for step S4.1 is as follows: S4.11. By fusing the surface adaptive superpixel unit in the texture morphology optimization parameter set with the depth point cloud surface normal vector data in the spatial geometry parameter set, a geodesic node set that fits the bending morphology of the stem-clasping leaf is constructed. The initial weight configuration of the nodes is dynamically adjusted according to the ground object spectral feature parameter set and the physiological verification parameter set to form a graph node topology structure that is adapted to the plant surface geometry. S4.12. Analyze the spatial geometric parameter set of cultivation density prior and depth point cloud surface differential geometric features, construct a position potential energy function to penalize deviations from the standard planting grid spatial distribution, extract the direction of the stem-holding leaf strip geometric principal axis to construct an anisotropic shape potential energy function to constrain the continuity of the surface orientation, and couple them to form a joint potential energy model that integrates cultivation norms and morphological priors.

[0058] In this embodiment: Step S4.11 constructs a geodesic node set that fits the real bending morphology of the stem-clasping leaf by fusing the surface adaptive superpixel unit and the depth point cloud surface normal vector data. The initial weight configuration of the nodes is dynamically adjusted according to the spectral characteristics of the ground object and the physiological verification parameters, forming a graph node topology structure that is adapted to the geometric height of the plant surface. This effectively solves the problem of the disconnect between the node connection relationship and the real plant geometry in traditional planar image segmentation, and improves the accuracy of the graph model in representing the orientation of the leaf surface.

[0059] Step S4.12 constructs a positional potential function that penalizes deviations from the standard planting grid spatial distribution by analyzing the prior of cultivation density and the differential geometric features of the depth point cloud surface. It also extracts the geometric principal axis direction of the stem-holding leaves to construct an anisotropic shape potential function that constrains the continuity of the surface orientation. This coupling forms a joint potential model that integrates cultivation norms and morphological priors, effectively suppressing interference from ground features such as scattered fallen leaves and patches of moss that are spectrally similar but spatially abnormal. This achieves an overall object-level understanding from discrete pixels to a model that conforms to the semantics of cultivation.

[0060] Compared to traditional probabilistic graphical models that rely solely on planar image features or simple spatial proximity relationships, this sub-step effectively overcomes the technical bottleneck of two-dimensional planar models failing to accurately reflect the curved morphology of clasping leaves by introducing surface differential geometry derived from deep point cloud analysis to construct geodesic node sets and anisotropic shape potential energy. This achieves graph topology construction that is compatible with the actual geometric morphology of the plant. Furthermore, by dynamically adjusting node weights by integrating prior cultivation density and physiological verification parameters, the problem of traditional fixed-weight configurations being ill-suited to the complex and variable environment under the forest canopy is solved. This improves the accuracy and spatial consistency of conditional random field models in identifying diseased areas, providing a high-confidence probabilistic inference foundation for precise disease localization.

[0061] Example 10: Please refer to Figure 1 The specific method for step S4.2 is as follows: S4.21. Based on the mean field approximation, the message passing iterative optimization of node label configuration is implemented. The capacitance value of the physiological verification parameter set is integrated to construct a complementary evidence item of physiological activity and mapped to an energy correction function. According to the distribution characteristics of capacitance value below the physiological activity threshold, the labels of non-living areas are forcibly corrected to remove the diseased leaf category. The misclassification of background ground features with similar spectral characteristics but abnormal capacitance value is suppressed. The high-confidence accurate disease area markers are generated iteratively and converged. S4.22. Based on the optimized and converged node labels, the percentage of diseased leaf pixels and the percentage of diseased area are statistically analyzed and mapped to the plant-level diseased leaf rate and severity level index. The spatial geometric parameter set of plant coordinates and cultivation density priors are integrated to construct the disease spatial distribution topology, forming a disease identification parameter set containing disease category identifiers, severity quantification values ​​and precise spatial coordinates to support subsequent variable control decisions.

[0062] In this embodiment: Step S4.21 iteratively optimizes node labels using a message passing algorithm based on the mean field approximation, and constructs complementary evidence terms of physiological activity by fusing capacitance value data from the physiological verification parameter set and mapping them to an energy correction function. This effectively solves the technical bottleneck of difficulty in distinguishing between living diseased leaves and inanimate background features by simply relying on spectral features. By forcibly correcting the labels of inanimate regions based on the distribution characteristics of capacitance values ​​below the physiological activity threshold, the misclassification of background features with similar spectral features but abnormal capacitance values ​​is suppressed. The high-confidence, accurate disease area markers generated after iterative convergence improve the reliability and biological rationality of the identification results.

[0063] Step S4.22 calculates the percentage of diseased leaf pixels and the percentage of lesion area based on the optimized and converged node labels, and maps them to the plant-level diseased leaf rate and severity level index, effectively achieving a quantitative assessment of disease severity. By integrating plant coordinates and prior cultivation density from the spatial geometric parameter set to construct a disease spatial distribution topology, a disease identification parameter set containing disease category identifiers, severity quantification values, and precise spatial coordinates is formed. This provides high-precision spatial positioning support for subsequent targeted pesticide application and precision irrigation, improving the accuracy of control operations and resource utilization efficiency.

[0064] Compared to traditional detection methods that rely solely on spectral features or simple threshold segmentation, this sub-step effectively overcomes the bottleneck of identifying spectrally similar but physiologically diverse ground features by constructing an energy correction function that integrates physiological and electrical properties. This achieves a forced label correction mechanism based on live physiological activity, reducing the false alarm rate of inactive background features. Simultaneously, by constructing a disease spatial distribution topology that integrates plant coordinates and prior cultivation density, it achieves a leap from discrete pixel-level classification to continuous plant-level semantic understanding, enabling synergistic optimization of quantitative disease severity assessment and precise spatial localization. This technical solution comprehensively improves the accuracy and reliability of disease area labeling, providing a high-confidence data foundation and precise spatial guidance for subsequent variable-based control decisions, and enhancing the refinement of forest-grown Fritillaria cirrhosa disease management.

[0065] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or as software functional units. The above are merely embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made based on the description and drawings of this application, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

[0066] The specific embodiments of the invention have been described in detail above, but they are only examples, and this application is not limited to the specific embodiments described above. For those skilled in the art, any equivalent modifications or substitutions to the invention are also within the scope of this application. Therefore, all equivalent changes, modifications, and improvements made without departing from the spirit and principles of this application should be covered within the scope of this application.

Claims

1. A method for detecting the growth status of Fritillaria cirrhosa under forest cultivation, characterized in that: The specific steps are as follows: S1. Regional monitoring deployment: Divide the cultivation area into n detection areas, deploy sensor groups to collect multi-source data, and generate multispectral image parameter set, ambient light parameter set, physiological verification parameter set, spatial geometric parameter set and ground object spectral characteristic parameter set; The multispectral image parameter set includes: visible light image data, grayscale data after illumination normalization, and near-infrared channel data; The ambient light parameter set includes: spectral power distribution data and photosynthetically active radiation data; The physiological verification parameter set includes: capacitance data and conductivity data; The spatial geometric parameter set includes: depth point cloud data, plant spatial coordinate data, and prior data on cultivation density; The set of spectral feature parameters of ground features includes: spectral response feature data of diseased leaves, texture benchmark difference data of bark and soil, and scale feature data of moss and dead branches and fallen leaves in the forest. S2. Diseased leaf feature extraction: Convert the visible light image to the Lab color space, extract the yellowed diseased leaf area based on the blue-yellow axis chromaticity adaptive threshold, construct a hybrid probability model to optimize recognition, extract hue saturation features, and generate a set of disease-sensitive feature parameters; S3. Texture morphology optimization: Based on near-infrared data, a gray-scale spatial relationship matrix is ​​constructed to extract texture statistics. Color features are fused to distinguish brown spots from the background. Superpixel units are generated based on depth point clouds. The edges of the spots are optimized through morphological filtering to generate a texture morphology optimization parameter set. S4. Spatial semantic optimization: Based on the disease-sensitive feature parameter set and texture morphology optimization parameter set, an initial segmentation mask is constructed. A conditional random field model is established to integrate cultivation density to construct position potential energy and leaf geometry to construct shape potential energy. Capacitance value constraints are introduced to optimize labels and generate an accurate disease identification parameter set. S5. Prevention and Control Decision: Calculate the severity of diseases based on the accurate disease identification parameter set, generate targeted pesticide application and irrigation suggestions, and transmit them to the operating equipment.

2. The method for detecting the growth status of Fritillaria cirrhosa under forest cultivation according to claim 1, characterized in that, In step S2, the specific method for extracting diseased leaf features is as follows: S2.

1. Convert the visible light data in the multispectral image parameter set to the Lab color space, extract the blue-yellow axis chromaticity components and calculate their mean and standard deviation to construct an adaptive threshold, extract pixels with blue-yellow axis chromaticity values ​​greater than the threshold as the initial candidate regions for yellowing diseased leaves, and simultaneously extract the diseased leaf clustering features of hue saturation joint distribution. S2.2 Construct a mixed probability model of two-dimensional chromaticity of green-red axis and blue-yellow axis, estimate model parameters through iterative algorithm and calculate the posterior probability of each pixel belonging to yellow leaf disease, remove low confidence pixels in the initial candidate area to optimize the recognition results, and integrate hue saturation clustering features and probability model parameters to generate a set of disease-sensitive feature parameters.

3. The method for detecting the growth status of Fritillaria cirrhosa under forest cultivation according to claim 2, characterized in that, The specific method for step S2.1 is as follows: S2.

11. Based on the spectral power distribution and photosynthetically active radiation data of the ambient light parameter set, the multispectral image parameter set is normalized and corrected and mapped to the Lab color space. The blue-yellow axis and green-red axis chromaticity components are extracted. The capacitance and conductivity data of the physiological verification parameter set are combined to distinguish between disease chlorosis and physiological water deficiency chlorosis. An adaptive chromaticity threshold model that integrates physiological criteria is constructed. Pixels that meet the threshold and conform to the prior distribution of cultivation density in the spatial geometric parameter set are extracted to form the initial candidate region for chlorotic leaves. S2.

12. Based on the plant coordinates and depth point cloud data of the spatial geometric parameter set, a three-dimensional region of interest is delineated. The multispectral image parameter set is mapped to the hue saturation color space to construct a two-dimensional joint distribution. Combining the spectral response characteristics of diseased leaves from the ground object spectral feature parameter set, diseased leaf cluster centers and boundary features that conform to the cultivation space pattern are extracted within the three-dimensional region of interest. Hue saturation diseased leaf feature parameters that match the subsequent spatial context constraints are generated.

4. The method for detecting the growth status of Fritillaria cirrhosa under forest cultivation according to claim 3, characterized in that, The specific method for step S2.2 is as follows: S2.

21. Combine the chromaticity data of the green-red axis and the blue-yellow axis with the capacitance value distribution of the physiological verification parameter set, construct a dynamic probability model to distinguish between disease-induced yellowing and physiological water loss, optimize its parameter estimation, calculate the posterior probability of a pixel belonging to the diseased leaf category, and combine the prior of the cultivation density of the spatial geometric parameter set to remove isolated pixels that deviate from the plant distribution pattern, and generate a physiologically verified high-confidence yellowing diseased leaf identification region. S2.

22. Construct a two-dimensional joint distribution of hue saturation to extract the cluster center and boundary features of diseased leaves. Integrate the optimized yellowed leaf regions, probability distribution parameters and cluster features. Dynamically adjust the feature weights based on the environmental light parameter set. Combine the ground object spectral feature parameter set to construct a multi-dimensional joint descriptor that is compatible with subsequent texture analysis and spatial context constraints, and generate a disease-sensitive feature parameter set.

5. The method for detecting the growth status of Fritillaria cirrhosa under forest cultivation according to claim 4, characterized in that, In step S3, the specific methods for texture morphology optimization are as follows; S3.

1. Targeting the unique parallel distribution of leaf veins and heterogeneous texture pattern of necrotic tissue in the brown spot lesions of the stem-clasping strip leaves of Fritillaria cirrhosa, a pixel-level gray-scale spatial relationship matrix is ​​constructed by calling the near-infrared channel data of the multispectral image parameter set and the spatial distribution of capacitance values ​​of the physiological verification parameter set. The surface roughness and anisotropy index, which characterize the degree of necrosis of lesion tissue and the difference in physiological water loss, are extracted. A color and texture joint feature descriptor is constructed by integrating the disease sensitive feature parameter set. Based on the ground spectral feature parameter set, brown spot lesions with similar hues but heterogeneous texture structure are identified from the background, and physiologically verified texture-constrained lesion candidate masks are generated. S3.

2. Based on the surface normal vector of the depth point cloud of the spatial geometric parameter set, estimate the surface geometry of the stem-holding leaf, generate superpixel units that fit the bending shape of the leaf, and establish a spatial adjacency graph. Statistically measure the spatial distribution consistency of the joint features of color and texture within the unit. Adaptively construct a multi-scale morphological opening and closing operation structure element sequence based on the scale features of forest moss and dead leaves and the abnormal distribution of capacitance values ​​in the ground cover spectral feature parameter set. Cascade filter out small background interference and restore the continuity of brown spot disease edge. Construct a texture morphology optimization parameter set that is compatible with subsequent spatial context constraints.

6. The method for detecting the growth status of Fritillaria cirrhosa under forest cultivation according to claim 5, characterized in that, The specific method for step S3.1 is as follows: S3.

11. Call the near-infrared channel data of the multispectral image parameter set, combine it with the spatial distribution of capacitance values ​​of the physiological verification parameter set to construct a pixel-level gray-scale spatial relationship matrix, correct the texture directionality index according to the capacitance value gradient, extract the surface roughness and adaptive directionality index that characterize the difference between necrotic lesion tissue and physiological water loss, and establish a texture feature mapping that integrates physiological criteria. S3.

12. Map the color features and texture index of the disease sensitive feature parameter set to the joint feature space. Based on the difference in texture benchmark between bark and soil in the ground object spectral feature parameter set, identify brown spots with similar hues but heterogeneous texture structures from the background, and generate texture-constrained spot candidate masks that are physiologically verified and adapted to subsequent spatial context constraints.

7. The method for detecting the growth status of Fritillaria cirrhosa under forest cultivation according to claim 6, characterized in that, The specific method for step S3.2 is as follows: S3.

21. Based on the spatial geometric parameter set, the differential geometry of the stem-holding leaf surface is analyzed by deep point cloud analysis. Adaptive superpixel units that fit the curvature change of the leaf are constructed and topological connections are established. The spatial distribution consistency of color and texture features within the unit is statistically analyzed to form a surface feature descriptor that is compatible with subsequent conditional random fields. S3.

22. Based on the joint constraints of the small-scale features of the ground object spectral feature parameter set and the abnormal distribution of capacitance values ​​in the physiological verification parameter set, an adaptive multi-scale morphological structure element sequence is constructed to restore the continuity of brown spot lesion edges and filter out background interference, generating a texture morphology optimization parameter set coupled with spatial context constraints.

8. The method for detecting the growth status of Fritillaria cirrhosa under forest cultivation according to claim 7, characterized in that, In step S4, the specific method for spatial semantic optimization is as follows: S4.

1. Based on the posterior probability of the disease-sensitive feature parameter set and the region labeling of the texture morphology optimization parameter set, an initial configuration is constructed. A fully connected conditional random field model is established to define a univariate potential energy term for the fusion of color and texture confidence. Based on the cultivation density prior of the spatial geometric parameter set, a position potential energy function is constructed to penalize the spatial inconsistency of deviation from the norm grid. Based on the strip geometry of the stem-clamping leaf analyzed by the depth point cloud, a shape potential energy function is constructed to constrain the directional continuity of adjacent nodes, forming a joint potential energy model that integrates cultivation norms and morphological priors. S4.

2. The message passing algorithm is used to iteratively optimize the node label allocation. The capacitance value of the physiological verification parameter set is introduced to construct the physiological activity constraint term. The non-living area labels with capacitance values ​​lower than the physiological activity threshold are forcibly corrected to suppress background misclassification. The accurate disease area markers are generated through iterative convergence. The severity level is calculated based on the proportion of diseased leaf pixels and the proportion of lesion area based on the optimized node labels. The spatial geometric parameter set of plant coordinates is integrated to locate the spatial distribution of the disease, forming a disease identification parameter set that includes disease category identifiers, severity quantification values ​​and spatial coordinates.

9. The method for detecting the growth status of Fritillaria cirrhosa under forest cultivation according to claim 7, characterized in that, The specific method for step S4.1 is as follows: S4.

11. By fusing the surface adaptive superpixel unit in the texture morphology optimization parameter set with the depth point cloud surface normal vector data in the spatial geometry parameter set, a geodesic node set that fits the bending morphology of the stem-clasping leaf is constructed. The initial weight configuration of the nodes is dynamically adjusted according to the ground object spectral feature parameter set and the physiological verification parameter set to form a graph node topology structure that is adapted to the plant surface geometry. S4.

12. Analyze the spatial geometric parameter set of cultivation density prior and depth point cloud surface differential geometric features, construct a position potential energy function to penalize deviations from the standard planting grid spatial distribution, extract the direction of the stem-holding leaf strip geometric principal axis to construct an anisotropic shape potential energy function to constrain the continuity of the surface orientation, and couple them to form a joint potential energy model that integrates cultivation norms and morphological priors.

10. A method for detecting the growth status of Fritillaria cirrhosa under forest cultivation according to claim 7, characterized in that, The specific method for step S4.2 is as follows: S4.

21. Based on the mean field approximation, the message passing iterative optimization of node label configuration is implemented. The capacitance value of the physiological verification parameter set is integrated to construct a complementary evidence item of physiological activity and mapped to an energy correction function. According to the distribution characteristics of capacitance value below the physiological activity threshold, the labels of non-living areas are forcibly corrected to remove the diseased leaf category. The misclassification of background ground features with similar spectral characteristics but abnormal capacitance value is suppressed. The high-confidence accurate disease area markers are generated iteratively and converged. S4.

22. Based on the optimized and converged node labels, the percentage of diseased leaf pixels and the percentage of diseased area are statistically analyzed and mapped to the plant-level diseased leaf rate and severity level index. The spatial geometric parameter set of plant coordinates and cultivation density priors are integrated to construct the disease spatial distribution topology, forming a disease identification parameter set containing disease category identifiers, severity quantification values ​​and precise spatial coordinates to support subsequent variable control decisions.

Images