Intelligent monitoring system for health of garden plants based on image recognition

Abstract

The present application relates to the field of garden plant health monitoring, in particular to a garden plant health intelligent monitoring system based on image recognition, comprising a multispectral image acquisition module, a differential geometry feature extraction module, a multiscale topological analysis module, a geodesic line evolution prediction module, a health state evaluation module and a warning and decision support module, the system models the plant leaf surface as a two-dimensional Riemann manifold, accurately extracts the curvature distribution, shape features and boundary complexity of the disease spot; through the multiscale topological analysis method, the connectivity and complexity of the internal structure of the disease spot are captured; based on the geodesic distance, a disease spot diffusion dynamics model is constructed to predict the disease spot propagation path and rate; a multi-level health evaluation system from leaf to garden is constructed to realize the quantitative evaluation of the plant health state. The present application significantly improves the disease identification accuracy, especially the identification ability of early and similar diseases; the prediction time window is extended from 12 hours of the traditional method to 72 hours.

Description

Technical Field

[0001] This invention relates to the field of garden plant health monitoring, and in particular to an intelligent garden plant health monitoring system based on image recognition. This system utilizes multispectral image acquisition technology, differential geometry theory, and deep learning algorithms to achieve accurate monitoring, analysis, and early warning of the health status of garden plants. Background Technology

[0002] Garden plants are an important component of urban ecosystems, and their health directly impacts the quality of the urban ecological environment. Traditional garden plant health monitoring mainly relies on manual inspections, which suffers from drawbacks such as high workload, low efficiency, and strong subjectivity, making it difficult to meet the needs of large-scale garden plant management. With the development of computer vision and artificial intelligence technologies, image recognition-based plant health monitoring methods are gradually emerging, but existing technologies still have the following problems:

[0003] On the one hand, existing image recognition methods are mainly based on simple color and texture features, resulting in limited accuracy in identifying plant diseases, especially in the early stages of disease and in complex environments. On the other hand, existing technologies typically focus only on disease identification and lack the ability to predict disease development trends, leading to delayed control measures and missed opportunities for optimal intervention. Furthermore, existing systems often employ isolated assessment methods, lacking a multi-level health assessment system from leaf to overall plant health, making it difficult to support refined decision-making in garden plant management.

[0004] Therefore, there is an urgent need for an intelligent monitoring system that can accurately identify plant diseases, predict their development trends, and provide multi-level health assessments to improve the scientific nature and effectiveness of garden plant health management. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent monitoring system for the health of garden plants based on image recognition. This system innovatively combines multispectral imaging technology with differential geometry theory to construct a multi-level plant health monitoring and assessment system from micro to macro. It realizes the accurate identification of plant diseases, the scientific prediction of development trends, and the quantitative assessment of health status, providing strong technical support for the health management of garden plants.

[0006] An image recognition-based intelligent monitoring system for the health of garden plants includes:

[0007] A multispectral image acquisition module is used to acquire multispectral image data of garden plant leaves;

[0008] The differential geometric feature extraction module is communicatively connected to the multispectral image acquisition module. It is used to receive the multispectral image data, model the plant leaf surface as a two-dimensional Riemannian manifold, and extract the curvature distribution, shape features, and boundary complexity of the lesion area.

[0009] The multi-scale topology analysis module is communicatively connected to the differential geometric feature extraction module. It is used to receive the geometric features of the lesion region, analyze the topological structure features of the lesion at multiple scales, and generate a topological feature vector that reflects the connectivity and complexity of the internal structure of the lesion.

[0010] The geodesic evolution prediction module is communicatively connected to the multi-scale topology analysis module and the differential geometric feature extraction module. It is used to construct a lesion diffusion dynamics model based on the geodesic distance field on the leaf surface and predict the propagation path and rate of lesions on the leaf surface.

[0011] The health status assessment module is communicatively connected to the differential geometric feature extraction module, the multi-scale topology analysis module, and the geodesic evolution prediction module. It is used to comprehensively assess the plant health status by integrating the geometric features of the lesions, the topological feature vectors, and the lesion propagation prediction results.

[0012] The early warning and decision support module is communicatively connected to the health status assessment module and is used to generate prevention and control suggestions and early warning information based on the plant health status.

[0013] Preferably, the multispectral image acquisition module includes:

[0014] Image acquisition unit, used to acquire multi-band images within the wavelength range of 400-1000nm;

[0015] The environmental parameter monitoring unit is used to collect environmental parameters such as temperature, humidity, and light intensity.

[0016] An image preprocessing unit is used to perform spatial registration and spectral correction on the multi-band image, and generate a standardized multispectral image and a leaf region mask.

[0017] The data management unit is used to establish the correlation between the multispectral image and the environmental parameters, and to store historical monitoring data.

[0018] Preferably, the differential geometric feature extraction module includes:

[0019] The blade surface parameterization unit is used to convert the blade region in the multispectral image into a triangular mesh representation and establish a mapping relationship from three-dimensional space to two-dimensional parameter domain.

[0020] The lesion region identification unit is used to identify lesion regions by segmenting RGB channel differences and thresholds, and to accurately determine lesion boundaries;

[0021] Curvature calculation unit is used to estimate principal curvature values ​​at the vertices of the lesion region mesh and to calculate the average curvature and Gaussian curvature distribution characteristics.

[0022] The feature vector construction unit is used to integrate curvature features, shape features, and boundary features to generate differential geometric feature vectors of lesions.

[0023] Preferably, the multi-scale topology analysis module includes:

[0024] Scale-space construction unit is used to construct a height function based on lesion pixel intensity and generate lesion representations at multiple scales;

[0025] Topological feature calculation unit, used to calculate the number of connected components, void structure and Betti number of lesion region at each scale;

[0026] The persistence analysis unit is used to record the scale of the appearance and disappearance of topological features and to identify significant topological features with high persistence.

[0027] The feature vector generation unit is used to calculate the statistical characteristics of the persistent distribution and combine the topological information of each dimension to form a topological feature vector.

[0028] Preferably, the geodesic evolution prediction module includes:

[0029] The leaf vein extraction unit is used to identify the leaf vein structure and perform hierarchical division based on the thickness and connectivity of the leaf veins.

[0030] The geodesic distance calculation unit is used to incorporate leaf vein structure information into a distance function to calculate the geodesic distance between any two points on the leaf surface.

[0031] The diffusion model construction unit is used to construct the lesion diffusion equation based on the geodesic distance field and set boundary conditions that reflect the leaf structure and environmental conditions.

[0032] The time series prediction unit is used to generate lesion development prediction results for 24 hours, 48 ​​hours, and 72 hours.

[0033] Preferably, the health status assessment module includes:

[0034] The leaf health calculation unit is used to calculate the health status of a single leaf based on the geometric features, topological features, and location information of lesions.

[0035] The whole plant health calculation unit is used to comprehensively consider the health status of all visible leaves and the location of lesions in different leaves to calculate the overall health status of the plant.

[0036] The park health calculation unit is used to assess the overall health status of the park based on the health level and spatial distribution of each plant.

[0037] The growth assessment unit is used to evaluate the growth status of plants by analyzing the RGB channel differences of leaves.

[0038] Preferably, the early warning and decision support module includes:

[0039] The disease identification unit is used to identify disease types based on differential geometric features and topological features;

[0040] The prevention and control suggestion generation unit is used to generate targeted prevention and control measures suggestions based on the disease type and predicted development trend;

[0041] The early warning information push unit is used to send early warning information to managers when the plant health status reaches the early warning threshold;

[0042] The effectiveness evaluation unit is used to track changes in plant health status after the implementation of control measures and to evaluate the control effect.

[0043] Preferably, the data storage module is also included, the data storage module comprising:

[0044] A time-series database is used to store time-series data of monitoring images and environmental parameters;

[0045] Feature library, used to store extracted differential geometric features and topological features;

[0046] A knowledge base is used to store expert knowledge about disease types, symptom characteristics, and control methods.

[0047] A model library is used to store trained recognition and prediction models.

[0048] The data storage module is communicatively connected to various functional modules of the system for data storage, retrieval, and sharing.

[0049] Preferably, it also includes a user interaction module, the user interaction module comprising:

[0050] The mobile terminal application unit is used to display monitoring results and early warning information on smartphones or tablets;

[0051] The web-based management unit provides a browser-based interface for system management and data analysis.

[0052] The on-site display unit is used to provide real-time monitoring information and prevention and control guidance in the park.

[0053] The user interaction module is communicatively connected to the early warning and decision support module to realize human-computer interaction and information display.

[0054] Preferably, the system collects environmental data through a wireless sensor network, performs preliminary data processing through edge computing nodes, conducts in-depth analysis and storage through a cloud platform, and supports simultaneous access and management by multiple users, thereby enabling remote monitoring, real-time early warning and intelligent management of the health status of garden plants.

[0055] The present invention has the following beneficial effects:

[0056] 1. By introducing differential geometry theory to analyze the morphological characteristics of lesions, the surface of plant leaves is modeled as a two-dimensional Riemannian manifold, which accurately quantifies the curvature distribution, shape characteristics and boundary complexity of lesions, greatly improving the accuracy of disease identification, especially the ability to distinguish between early and similar diseases.

[0057] 2. By employing a multi-scale topological analysis method, the connectivity and complexity of the internal structure of lesions are captured at different scales, forming a unique topological fingerprint that effectively distinguishes different types and stages of disease development, especially significantly improving the accuracy of identifying complex lesions.

[0058] 3. A lesion spread dynamics model was constructed based on geodesic distance, taking into account the influence of leaf surface geometry and vein distribution on disease spread. This model enabled accurate prediction of lesion development trends, extending the prediction time window from 12 hours in traditional methods to 72 hours, providing ample time for timely intervention.

[0059] 4. A multi-level health assessment system was constructed, from leaves to the entire park. By scientifically quantifying health indicators at each level, a comprehensive assessment of the health status of garden plants was achieved, providing a scientific basis for refined management decisions.

[0060] 5. The system integrates functional modules such as image acquisition, feature extraction, health assessment, trend prediction, and decision support, forming a complete technical solution that significantly improves the efficiency and accuracy of garden plant health management and reduces plant loss and management costs. Attached Figure Description

[0061] Figure 1 This is a schematic diagram of the overall architecture of the intelligent monitoring system for garden plant health based on image recognition according to the present invention;

[0062] Figure 2 This is a schematic diagram of the structure of the multispectral image acquisition module of the present invention;

[0063] Figure 3 This is a schematic diagram of the differential geometric feature extraction module of the present invention;

[0064] Figure 4 This is a schematic diagram of the multi-scale topology analysis module of the present invention;

[0065] Figure 5This is a schematic diagram of the geodesic evolution prediction module of the present invention;

[0066] Figure 6 This is a schematic diagram of the health status assessment module of the present invention;

[0067] Figure 7 This is a schematic diagram of the early warning and decision support module of the present invention;

[0068] Figure 8 This is a schematic diagram illustrating the data flow between the modules of the system of the present invention;

[0069] Figure 9 This is a schematic diagram of the differential geometric feature extraction process for lesions in this invention;

[0070] Figure 10 This is a schematic diagram of the multi-scale topological analysis process for lesions in this invention;

[0071] Figure 11 This is a schematic diagram of the lesion evolution prediction process based on geodesic distance according to the present invention.

[0072] Figure 12 This is a schematic diagram of the multi-level health assessment process of the present invention. Detailed Implementation

[0073] Please refer to Figures 1-12 The present invention will now be described in further detail with reference to the accompanying drawings and embodiments.

[0074] like Figure 1 As shown, the intelligent health monitoring system for garden plants based on image recognition of the present invention includes a multispectral image acquisition module 1, a differential geometric feature extraction module 2, a multi-scale topological analysis module 3, a geodesic evolution prediction module 4, a health status assessment module 5, an early warning and decision support module 6, a data storage module 7, and a user interaction module 8.

[0075] Multispectral image acquisition module 1 is used to acquire multispectral image data of garden plant leaves. Differential geometric feature extraction module 2 communicates with multispectral image acquisition module 1, receiving multispectral image data, modeling the plant leaf surface as a two-dimensional Riemannian manifold, and extracting the curvature distribution, shape features, and boundary complexity of the lesion region. Multiscale topology analysis module 3 communicates with differential geometric feature extraction module 2, receiving the geometric features of the lesion region, analyzing the topological structure features of the lesion at multiple scales, and generating topological feature vectors reflecting the connectivity and complexity of the lesion's internal structure. Geodesic evolution prediction module 4 communicates with multiscale topology analysis module 3 and differential geometric feature extraction module 2, constructing a lesion diffusion dynamics model based on the geodesic distance field of the leaf surface, and predicting the propagation path and rate of lesions on the leaf surface. Health status assessment module 5 communicates with differential geometric feature extraction module 2, multiscale topology analysis module 3, and geodesic evolution prediction module 4, integrating lesion geometric features, topological feature vectors, and lesion propagation prediction results to assess the plant's health status. The early warning and decision support module 6 communicates with the health status assessment module 5 to generate prevention and control suggestions and early warning information based on the plant's health status.

[0076] The modules exchange data and collaborate through standardized interfaces, forming a complete closed loop from data collection, feature extraction, analysis and prediction to application services, enabling comprehensive monitoring and intelligent management of the health status of garden plants.

[0077] like Figure 2 As shown, the multispectral image acquisition module 1 includes an image acquisition unit 11, an environmental parameter monitoring unit 12, an image preprocessing unit 13, and a data management unit 14.

[0078] The image acquisition unit 11 is used to acquire multi-band images within the wavelength range of 400-1000nm. Preferably, this unit employs a high-resolution multispectral camera with a spatial resolution of not less than 0.1mm / pixel, including at least five spectral bands covering the visible and near-infrared regions. In one embodiment of the present invention, five bands—400-500nm, 500-600nm, 600-700nm, 700-800nm, and 800-1000nm—are selected, corresponding to blue light, green light, red light, and two near-infrared bands, respectively, which can effectively capture the reflectivity and health status changes of plant leaves.

[0079] The environmental parameter monitoring unit 12 is used to collect environmental parameters such as temperature, humidity, and light intensity. This unit acquires key parameters of the plant growth environment through temperature and humidity sensors and light intensity sensors. These parameters will serve as important inputs for predicting disease development trends. In practical applications, the temperature sensor has a measurement range of -40℃ to 85℃ and an accuracy of ±0.5℃; the humidity sensor has a measurement range of 0% to 100%RH and an accuracy of ±3%RH; and the light intensity sensor has a measurement range of 0-100,000 lux, which can adapt to the monitoring needs under different light conditions.

[0080] The image preprocessing unit 13 is used to perform spatial registration and spectral correction on the multi-band images, and generate a standardized multispectral image and a leaf region mask. This unit first applies Gaussian filtering for noise suppression, with the filter kernel size adaptively adjusted according to the image resolution, typically choosing a size of 3×3 or 5×5; then, image enhancement is performed, including contrast adjustment and histogram equalization; next, the leaf region is extracted using threshold segmentation and edge detection methods to generate a leaf mask; finally, spatial registration and radiometric correction are performed on the multi-band images to eliminate differences caused by inconsistent imaging conditions.

[0081] The data management unit 14 is used to establish the correlation between multispectral images and environmental parameters, and to store historical monitoring data. This unit uses timestamps and spatial coordinates to index the data, establishing a structured data organization method that supports efficient data retrieval and analysis. For long-term monitoring, the system automatically compresses and archives the data while retaining key feature information, balancing storage efficiency and data integrity.

[0082] like Figure 3 As shown, the differential geometric feature extraction module 2 includes a leaf surface parameterization unit 21, a lesion area identification unit 22, a curvature calculation unit 23, and a feature vector construction unit 24.

[0083] The blade surface parameterization unit 21 is used to convert the blade region in the multispectral image into a triangular mesh representation and establish a mapping relationship from three-dimensional space to the two-dimensional parameter domain. This unit first generates point cloud data based on the blade mask, then constructs a triangular mesh using the Delaunay triangulation algorithm. The mesh density is adaptively adjusted according to the blade size and detail requirements, typically no less than 100 triangles per square centimeter. Next, a least-squares conformal mapping algorithm is used to establish the mapping from the three-dimensional surface to the two-dimensional parameter domain, preserving the geometric characteristics of the blade surface.

[0084] The lesion region identification unit 22 is used to identify lesion regions through RGB channel differences and threshold segmentation, and to accurately determine the lesion boundaries. This unit first calculates the difference features of the RGB three channels, particularly the difference between the green channel and the red channel, which shows a clear distinction between healthy and diseased tissue. Preferably, when the difference is less than a threshold T1 (in practice, T1 is usually set to 15-25, depending on the plant species and light conditions), it is initially identified as a lesion region. Then, a region growth algorithm is applied to refine the lesion boundaries. The seed point is selected from the central region of the initial segmentation result, and the growth conditions include color similarity and gradient change. Finally, morphological operations (such as opening and closing operations) are applied to optimize the shape of the lesion region, eliminating noise and small holes.

[0085] Curvature calculation unit 23 is used to estimate the principal curvature values ​​at the vertices of the mesh in the lesion region and to calculate the average curvature and Gaussian curvature distribution characteristics. This unit is based on discrete differential geometry theory and uses the cotangent-weighted Laplace-Beltrammian operator to calculate the principal curvature of the surface. For each vertex v on the mesh, the curvature calculation formula is:

[0086] ,

[0087] in, and As vertex The principal curvature at that point, The shape operator is calculated as follows:

[0088] ,

[0089] in, As vertex The area of ​​the Voronoi region, in square millimeters. As vertex The set of adjacent vertices, and To be with the edge The opposite angles of two adjacent triangles, in radians. The coordinates of the current vertex. The coordinates of adjacent vertices are all three-dimensional vectors. and Vertices and The unit normal vectors at each location are three-dimensional unit vectors, with superscripts indicating their positions. This represents the transpose of a vector.

[0090] Calculate the mean curvature based on the principal curvature. and Gaussian curvature :

[0091] ,

[0092] ,

[0093] in, As vertex The average curvature at that point, in millimeters. As vertex Gaussian curvature at that point, in millimeters (mm⁻²). and Vertices The maximum and minimum principal curvatures at the point are both in millimeters.

[0094] By statistically analyzing the curvature distribution within the lesion area, features such as the mean curvature, variance, skewness, and kurtosis are extracted to form a curvature distribution feature vector.

[0095] Feature vector construction unit 24 integrates curvature features, shape features, and boundary features to generate a differential geometric feature vector of the lesion. This unit first calculates shape factors (such as roundness, ellipticity, elongation, etc.) to quantify the overall shape of the lesion. Roundness C is defined as:

[0096] ,

[0097] in, Roundness, dimensionless, ranging from 0 to 1. The area of ​​the lesion is expressed in square millimeters. The perimeter of the lesion is measured in millimeters. Pi is the mathematical constant for circumference. The closer the roundness value is to 1, the closer the shape is to a circle.

[0098] Boundary complexity Calculated using fractal dimension:

[0099] ,

[0100] in, Boundary complexity, dimensionless. The perimeter of the lesion is measured in millimeters. The area of ​​the lesion is expressed in square millimeters. It represents a logarithm with base 10. The larger the value, the more complex the boundary.

[0101] Finally, the curvature features, shape features, and boundary features are normalized and then weighted and fused according to the weights determined in the experiment to generate a comprehensive feature vector. :

[0102] ,

[0103] in, The comprehensive feature vector is a multi-dimensional vector. For curvature eigenvectors, For shape feature vectors, For boundary feature vectors, , and These are the corresponding weighting coefficients, dimensionless, and are typically set to... , and It can be adjusted according to different plants and disease types.

[0104] like Figure 4 As shown, the multi-scale topology analysis module 3 includes a scale space construction unit 31, a topology feature calculation unit 32, a persistence analysis unit 33, and a feature vector generation unit 34.

[0105] The scale-space construction unit 31 is used to construct a height function based on lesion pixel intensity and generate lesion representations at multiple scales. This unit first uses the grayscale image of the lesion region as the height function, and then generates a series of smoothed versions at different scales through Gaussian filtering. Regarding the scale parameter... The Gaussian filter kernel is defined as:

[0106] ,

[0107] in, For Gaussian filter kernel, These are pixel coordinates, in pixels. This is a scale parameter, in pixels. Pi is the base of the natural logarithm.

[0108] Images at different scales are generated through convolution operations. :

[0109] ,

[0110] in, For the scale Images, The image is the original image; * indicates a convolution operation.

[0111] In practical applications, scale parameters It usually takes a series of values, such as Pixels cover multiple scale levels, from detail to the whole.

[0112] The topological feature calculation unit 32 is used to calculate the number of connected components, void structure, and Betti number of lesion regions at each scale. This unit sets a series of thresholds t for the image at each scale to generate a hyperlevel set.

[0113] ,

[0114] in, Representing an image At the threshold The superlevel set below, The scale is represented as Images, The threshold value is the range of grayscale values ​​in the image (usually 0-255). To meet the conditions The set of pixel coordinates.

[0115] For each hyperlevel set, compute its 0-dimensional and 1-dimensional Betti numbers: Indicates the number of connected components. This indicates the number of holes. These topological features, varying with threshold and scale, constitute the topological feature spectrum of the lesion.

[0116] The persistence analysis unit 33 is used to record the scales at which topological features appear and disappear, and to identify salient topological features with high persistence. This unit constructs a persistence graph, recording the birth and death scales of each topological feature:

[0117] ,

[0118] in, For persistence graphs, Let be the birth scale of the i-th topological feature. Let be the death scale of the i-th topological feature, where all values ​​are positive real numbers, and the units are related to the scale parameter. Same, usually in pixels. For the total number of features, For feature indexing.

[0119] The persistence value is defined as the difference between the death and birth scales:

[0120] ,

[0121] in, Let be the persistence value of the i-th topological feature, with units and scale parameters. The same, usually in pixels. The higher the persistence value, the more significant the topological feature, and the better it reflects the essential characteristics of the lesion.

[0122] The feature vector generation unit 34 is used to calculate the statistical characteristics of the persistent distribution and combine the topological information of each dimension to form a topological feature vector. This unit calculates the statistical characteristics of the persistent distribution, such as mean, variance, maximum value, and persistent histogram, and combines these features into a topological feature vector. :

[0123] ,

[0124] in, The topological feature vector is a multi-dimensional vector. For a 0-dimensional topological feature, a persistent feature sub-vector is formed. For a 1-dimensional topological feature, a persistent feature sub-vector is provided. These are statistical feature vectors.

[0125] Finally, the feature vectors are reduced in dimensionality and optimized to remove redundant information and retain the most discriminative features. In practice, principal component analysis (PCA) is typically used to reduce the feature dimension to 20-30 dimensions, retaining approximately 95% of the information.

[0126] like Figure 5 As shown, the geodesic evolution prediction module 4 includes a leaf vein extraction unit 41, a geodesic distance calculation unit 42, a diffusion model construction unit 43, and a time series prediction unit 44.

[0127] The leaf vein extraction unit 41 is used to identify the leaf vein structure and perform hierarchical classification based on vein thickness and connectivity. This unit first enhances the near-infrared image, as leaf veins exhibit significant reflectivity in the near-infrared band, particularly in the 800-900 nm range. Then, a multi-scale linear structure enhancement filter (such as a Frangi filter) is applied to strengthen the leaf vein structure.

[0128] ,

[0129] in, The enhanced leaf vein image uses values ​​ranging from 0 to 1. and Let be the eigenvalues ​​of the Hessian matrix, satisfying The unit is related to the image intensity gradient. for and The ratio, dimensionless. The Euclidean norm of the eigenvalues, and This is a control parameter, dimensionless, and is usually set to... , .

[0130] Next, adaptive threshold segmentation is applied to extract the leaf vein network, and morphological refinement is used to obtain the leaf vein skeleton. Finally, based on the vein thickness and connectivity, a hierarchical division is performed, typically into three levels: main veins, secondary veins, and reticulate veins, laying the foundation for the subsequent establishment of the geodesic distance field.

[0131] The geodesic distance calculation unit 42 incorporates leaf vein structure information into a distance function to calculate the geodesic distance between any two points on the leaf surface. This unit first constructs a Riemannian metric that considers leaf vein structure, defined as:

[0132] ,

[0133] in, For position The metric tensor component at that location, The Kronecker delta function (1 when i=j, 0 otherwise). For position The leaf vein intensity at a given location ranges from 0 to 1. This is a weighting parameter, dimensionless, with a value range of 0-1. It controls the degree of influence of leaf veins on geodesic distance, and is usually set to 0.3-0.7, which can be adjusted according to the leaf vein characteristics of different plants.

[0134] Based on this metric, the geodesic distance from the center of the lesion to various points on the leaf is calculated using the FastMarching Method:

[0135] ,

[0136] in, Let x be the geodesic distance function from the center of the lesion to location x, in millimeters. Let the gradient of the distance function be . For measuring tensors The reverse, and The distance function is respectively in and Partial derivatives in the direction.

[0137] Diffusion model building unit 43 is used to construct the lesion diffusion equation based on the geodesic distance field and to set boundary conditions that reflect leaf structure and environmental conditions. This unit uses the geodesic thermal diffusion equation to model the lesion diffusion process:

[0138] ,

[0139] in, This indicates the severity of the disease at location x at time t, with a value ranging from 0 to 1, where t is time in hours. Denotes the divergence operator, This is the diffusion coefficient field, measured in square millimeters per hour, and is related to the geodesic distance field.

[0140] ,

[0141] in, The basic diffusion coefficient is expressed in square millimeters per hour, typically ranging from 0.01 to 0.1 mm² / h. The distance from the center of the lesion to location x is the geodesic distance in millimeters. This is the distance influence factor, measured in millimeters, typically 5–15 mm. This is an environmental factor adjustment term, dimensionless, considering the impact of environmental factors such as temperature and humidity on disease spread.

[0142] ,

[0143] in, The temperature influence factor is dimensionless. Humidity is a dimensionless factor.

[0144] Temperature Influence Factors Defined as:

[0145] ,

[0146] in, This is the actual temperature, in degrees Celsius. The optimal temperature, measured in degrees Celsius, is typically 22-28°C. This is a temperature sensitivity parameter, measured in degrees Celsius, typically between 5-10°C.

[0147] Humidity influencing factors Defined as:

[0148] ,

[0149] in, This is the actual relative humidity, expressed as a percentage. The optimal humidity is expressed as a percentage, typically 80% to 95%.

[0150] Time series prediction unit 44 generates lesion development prediction results for 24 hours, 48 ​​hours, and 72 hours. This unit uses numerical methods (such as the finite difference method) to solve the diffusion equation and predict the disease distribution at different future time points. To improve computational efficiency, an adaptive time step and spatial grid are used, with finer grids used in key areas (such as lesion boundaries). The prediction results include changes in lesion area growth rate, diffusion direction, and severity, providing a scientific basis for prevention and control decisions.

[0151] In practical applications, the system continuously tracks the difference between the actual development of lesions and the predicted results, and constantly adjusts the model parameters through machine learning methods (such as Bayesian optimization) to improve prediction accuracy. Preferably, the system's prediction accuracy can reach over 85%, providing reliable early warning information for the prevention and control of diseases in garden plants.

[0152] like Figure 6 As shown, the health status assessment module 5 includes a leaf health calculation unit 51, a whole plant health calculation unit 52, a park health calculation unit 53, and a growth evaluation unit 54.

[0153] The leaf health calculation unit 51 is used to calculate the health status of a single leaf based on the geometric, topological, and location information of lesions. This unit comprehensively considers factors such as lesion area, location, shape, and color to calculate the leaf health status HL.

[0154] ,

[0155] in, The value represents leaf health, is dimensionless, and ranges from 0 to 1. This indicates the area of ​​the lesion, expressed in square millimeters. This represents the total area of ​​the blades, expressed in square millimeters. The damage index represents the impact of the location of the lesion on the health of the leaf. It is dimensionless and ranges from 0 to 1 (lesions closer to the veins and petioles have a higher damage index). This represents the effect of lesion color contrast, is dimensionless, and ranges from 0 to 1 (higher color contrast indicates a more severe condition). and These are weighting coefficients, dimensionless, and are typically set to... and . The value ranges from 0 to 1, with a larger value indicating healthier leaves.

[0156] The overall plant health calculation unit 52 comprehensively considers the health status of all visible leaves and the location of disease spots on different leaves to calculate the overall plant health status. This unit calculates the overall plant health status. :

[0157] ,

[0158] in, The overall health status is dimensionless and ranges from 0 to 1. Indicates that the whole tree has A leaf, For leaf index, Indicates the first The health status of a leaf blade is dimensionless. Indicates the lesion is in the first The damage index of a leaf location on the tree's health is dimensionless and ranges from 0 to 1 (the damage index is higher for lesions located on new leaves and key growth sites). This is a weighting coefficient, dimensionless, and is typically set to 0.4. Equal to the total number of blades . The value ranges from 0 to 1, with a larger value indicating a healthier plant.

[0159] The park health calculation unit 53 is used to assess the overall health status of the park based on the health level of each plant and its spatial distribution. This unit calculates the park's health score. :

[0160] ,

[0161] in, The health status of the park is dimensionless and ranges from 0 to 1. It indicates that there are in the park Plants For plant indexing, Indicates the first The health status of each plant is dimensionless. The damage index represents the position of the lesion in the kth plant and its impact on the overall health of the park. It is dimensionless and ranges from 0 to 1 (the damage index is higher for diseases located in key landscape locations and rare plants). This is a weighting coefficient, dimensionless, and is usually set to 0.3. It equals the total number of plants, m. The HP value ranges from 0 to 1, with a higher value indicating a healthier environment.

[0162] The growth assessment unit 54 is used to evaluate the plant's growth status by analyzing the RGB channel differences of the leaves. This unit assesses plant growth by analyzing the greenness of the leaves (the difference between the G channel and the R channel) and calculates the plant's average growth index VL.

[0163] ,

[0164] in, The average growth vigor index of plants is dimensionless and ranges from 0 to 1. N represents the number of all visible leaves in the tree, and t is the leaf index. Let t represent the growth index of the t-th plant leaf. It is dimensionless and calculated as follows:

[0165] ,

[0166] in, Let be the average value of the green channel of the t-th leaf, ranging from 0 to 255. is the average value of the red channel on the t-th leaf, ranging from 0 to 255. 30 is an empirical threshold, indicating that the green-red difference of a healthy leaf is usually greater than this value. The VL value ranges from 0 to 1, with a larger value indicating better plant growth.

[0167] like Figure 7 As shown, the early warning and decision support module 6 includes a disease identification unit 61, a prevention and control suggestion generation unit 62, an early warning information push unit 63, and an effect evaluation unit 64.

[0168] The disease identification unit 61 is used to identify disease types based on differential geometric features and topological features. This unit employs a multi-class support vector machine (SVM) or random forest algorithm to identify common garden plant diseases based on extracted feature vectors. In embodiments of this invention, the disease types that the system can identify include, but are not limited to: anthracnose, powdery mildew, rust, leaf spot, and wilt. The identification results include disease type and confidence level, providing a basis for subsequent prevention and control decisions.

[0169] The prevention and control suggestion generation unit 62 is used to generate targeted prevention and control measures based on disease type and predicted development trend. This unit, based on an expert knowledge base, combines disease type, severity, development trend, and environmental conditions to generate personalized prevention and control suggestions. The suggestions include specific information such as timing of treatment, pesticide selection, application method, and dosage, while also considering environmental requirements and cost factors, prioritizing eco-friendly prevention and control solutions.

[0170] The early warning information push unit 63 is used to send early warning information to management personnel when the plant health status reaches the early warning threshold. This unit sets multiple early warning thresholds: a leaf health HL below 0.8 indicates a mild early warning, below 0.6 indicates a moderate early warning, and below 0.4 indicates a severe early warning. Early warning information is pushed to relevant management personnel via mobile terminal app, SMS, or email to ensure timely detection and handling of plant health problems.

[0171] The effectiveness evaluation unit 64 is used to track changes in plant health status after the implementation of control measures and to evaluate the control effect. This unit compares plant health indicators before and after control, calculates the improvement rate and recovery speed, and evaluates the effectiveness of the control measures. Simultaneously, it records the effectiveness data of different control schemes, providing a reference for control decisions in similar situations in the future, and realizing the accumulation and optimization of control knowledge.

[0172] Data storage module 7 includes a time-series database, a feature database, a knowledge base, and a model database.

[0173] The time-series database stores time-series data of monitoring images and environmental parameters. This database employs a distributed time-series architecture, supporting efficient data storage and retrieval. Data is indexed by timestamp and spatial location, facilitating historical data retrieval and analysis. For long-term monitoring data, the system uses a multi-level storage strategy: recent data is stored on high-speed storage media, while historical data is compressed and archived.

[0174] The feature library stores extracted differential geometric and topological features. It employs a key-value pair storage structure, using plant IDs and timestamps as keys and feature vectors as values, supporting fast feature retrieval and comparison. Furthermore, the feature library supports incremental updates, allowing newly extracted features to be efficiently added to existing datasets.

[0175] The knowledge base stores expert knowledge on disease types, symptom characteristics, and control methods. It employs an ontology model to organize knowledge, establishing semantic relationships between diseases, symptoms, environmental factors, and control methods, supporting complex knowledge reasoning and queries. The knowledge base content originates from the experience of plant pathology experts and scientific literature, and is continuously enriched and optimized through practical data collected during system operation.

[0176] The model library stores trained recognition and prediction models. It supports model version management and performance evaluation, recording training data, parameter settings, and performance metrics to facilitate model selection and updates. Furthermore, the library provides a model deployment interface, supporting online updates and replacements to ensure continuous system optimization.

[0177] User interaction module 8 includes a mobile terminal application unit, a web-based management unit, and a live display unit.

[0178] The mobile application unit displays monitoring results and early warning information on smartphones or tablets. This unit provides an intuitive user interface, showcasing plant health status, disease identification results, and control recommendations. Users can view garden plant health monitoring data, receive early warning information, and perform simple system operations anytime via their mobile devices. The application supports offline mode, still providing basic functionality even with unstable network connections.

[0179] The web-based management unit provides a browser-based interface for system management and data analysis. Designed for landscape management and technical personnel, it offers richer system management and data analysis functions, including historical data querying, statistical analysis, system configuration, and user management. The web interface features a responsive design, adapting to different screen sizes and resolutions to ensure a good user experience.

[0180] The on-site display unit provides real-time monitoring information and prevention guidance within the park. This unit includes a waterproof display terminal installed within the park, showing plant health status and brief prevention guidelines, allowing on-site staff to directly access relevant information. The displayed content is concise and clear, focusing on the most pressing issues and avoiding information overload.

[0181] The image recognition-based intelligent monitoring system for the health of garden plants of this invention collects environmental data through a wireless sensor network, performs preliminary data processing through edge computing nodes, conducts in-depth analysis and storage through a cloud platform, and supports simultaneous access and management by multiple users, thereby realizing remote monitoring, real-time early warning and intelligent management of the health status of garden plants.

[0182] In actual deployment, the system first conducts environmental assessment and planning to determine monitoring points and equipment configuration; then, it builds network infrastructure and installs servers and storage devices; next, it deploys and debugs the monitoring equipment; finally, it configures the software system and optimizes its parameters. After the system is running, regular equipment maintenance and software updates are performed to ensure stable operation and continuous optimization.

[0183] This invention is applicable to various scenarios, including urban parks, roadside greenbelts, botanical gardens, nurseries, and ecological landscape areas. In urban parks, the system enables comprehensive health monitoring of plants, timely detection of diseases and provision of prevention and control suggestions, reducing plant losses and improving the park's landscape quality. In roadside greenbelts, the system can remotely monitor the health status of street trees, provide early warnings of potential safety hazards, and reduce management costs and safety risks. In botanical gardens and nurseries, the system provides refined plant health management services, offering technical support for the protection of rare plants and the cultivation of high-quality seedlings.

[0184] The embodiments described above are merely illustrative of specific implementations of the present invention, and while the descriptions are detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. An intelligent monitoring system for the health of garden plants based on image recognition, characterized in that, include: A multispectral image acquisition module is used to acquire multispectral image data of garden plant leaves; The differential geometric feature extraction module is communicatively connected to the multispectral image acquisition module. It is used to receive the multispectral image data, model the plant leaf surface as a two-dimensional Riemannian manifold, and extract the curvature distribution, shape features, and boundary complexity of the lesion area. The multi-scale topology analysis module is communicatively connected to the differential geometric feature extraction module. It is used to receive the geometric features of the lesion region, analyze the topological structure features of the lesion at multiple scales, and generate a topological feature vector that reflects the connectivity and complexity of the internal structure of the lesion. The multi-scale topology analysis module includes: a scale space construction unit, used to construct a height function based on lesion pixel intensity and generate lesion representations at multiple scales; a topology feature calculation unit, used to calculate the number of connected components, hole structure, and Betti number of lesion regions at each scale; a persistence analysis unit, used to record the scales at which topology features appear and disappear, and to identify significant topology features with high persistence; and a feature vector generation unit, used to calculate the statistical characteristics of persistence distribution and combine topology information from various dimensions to form a topology feature vector. The geodesic evolution prediction module, communicatively connected to the multi-scale topology analysis module and the differential geometric feature extraction module, is used to construct a lesion diffusion dynamics model based on the geodesic distance field of the leaf surface, and predict the propagation path and rate of lesions on the leaf surface. The geodesic evolution prediction module includes: a vein extraction unit for identifying the leaf vein structure and performing hierarchical division based on vein thickness and connectivity; a geodesic distance calculation unit for incorporating vein structure information into a distance function to calculate the geodesic distance between any two points on the leaf surface; a diffusion model construction unit for constructing a lesion diffusion equation based on the geodesic distance field and setting boundary conditions reflecting leaf structure and environmental conditions; and a time series prediction unit for generating lesion development prediction results for 24 hours, 48 ​​hours, and 72 hours. The health status assessment module is communicatively connected to the differential geometric feature extraction module, the multi-scale topology analysis module, and the geodesic evolution prediction module. It is used to comprehensively assess plant health status by integrating the geometric features of lesions, the topological feature vectors, and the lesion propagation prediction results. The health status assessment module includes: a leaf health calculation unit, used to calculate the health level of a single leaf based on the geometric features, topological features, and location information of lesions; a whole plant health calculation unit, used to comprehensively consider the health level of all visible leaves and the location of lesions in different leaves to calculate the health level of the entire plant; a park health calculation unit, used to assess the health status of the entire park based on the health level and spatial distribution of each plant; and a growth evaluation unit, used to assess plant growth status by analyzing the RGB channel differences of leaves. The early warning and decision support module is communicatively connected to the health status assessment module and is used to generate prevention and control suggestions and early warning information based on the plant health status.

2. The system according to claim 1, characterized in that, The multispectral image acquisition module includes: Image acquisition unit, used to acquire multi-band images within the wavelength range of 400-1000nm; The environmental parameter monitoring unit is used to collect environmental parameters such as temperature, humidity, and light intensity. An image preprocessing unit is used to perform spatial registration and spectral correction on the multi-band image, and generate a standardized multispectral image and a leaf region mask. The data management unit is used to establish the correlation between the multispectral image and the environmental parameters, and to store historical monitoring data.

3. The system according to claim 1, characterized in that, The differential geometric feature extraction module includes: The blade surface parameterization unit is used to convert the blade region in the multispectral image into a triangular mesh representation and establish a mapping relationship from three-dimensional space to two-dimensional parameter domain. The lesion region identification unit is used to identify lesion regions by segmenting RGB channel differences and thresholds, and to accurately determine lesion boundaries; Curvature calculation unit is used to estimate principal curvature values ​​at the vertices of the lesion region mesh and to calculate the average curvature and Gaussian curvature distribution characteristics. The feature vector construction unit is used to integrate curvature features, shape features, and boundary features to generate differential geometric feature vectors of lesions.

4. The system according to claim 1, characterized in that, The early warning and decision support module includes: The disease identification unit is used to identify disease types based on differential geometric features and topological features; The prevention and control suggestion generation unit is used to generate targeted prevention and control measures suggestions based on the disease type and predicted development trend; The early warning information push unit is used to send early warning information to managers when the plant health status reaches the early warning threshold; The effectiveness evaluation unit is used to track changes in plant health status after the implementation of control measures and to evaluate the control effect.

5. The system according to claim 1, characterized in that, It also includes a data storage module, which comprises: A time-series database is used to store time-series data of monitoring images and environmental parameters; Feature library, used to store extracted differential geometric features and topological features; A knowledge base is used to store expert knowledge about disease types, symptom characteristics, and control methods. A model library is used to store trained recognition and prediction models. The data storage module is communicatively connected to various functional modules of the system for data storage, retrieval, and sharing.

6. The system according to claim 1, characterized in that, It also includes a user interaction module, which includes: The mobile terminal application unit is used to display monitoring results and early warning information on smartphones or tablets; The web-based management unit provides a browser-based interface for system management and data analysis. The on-site display unit is used to provide real-time monitoring information and prevention and control guidance in the park. The user interaction module is communicatively connected to the early warning and decision support module to realize human-computer interaction and information display.

7. The system according to claim 1, characterized in that, The system collects environmental data through a wireless sensor network, performs preliminary data processing through edge computing nodes, conducts in-depth analysis and storage through a cloud platform, and supports simultaneous access and management by multiple users, enabling remote monitoring, real-time early warning and intelligent management of the health status of garden plants.

Images