Walnut tree disease and pest monitoring method based on unmanned aerial vehicle remote sensing data analysis

Abstract

The application discloses a walnut tree disease and pest monitoring method based on unmanned aerial vehicle remote sensing data analysis, and relates to the technical field of data processing. The method acquires a multi-time-phase tree crown image sequence of a single walnut tree, extracts brightness information and texture features, constructs a tree crown region segmentation result, and obtains the area and spatial distribution features of a tree crown pore region and a tree crown patch region based on connected domain analysis. Further, the method combines a tree interconnection relationship network, acquires an abnormal starting time of the single walnut tree, calculates the propagation score between abnormal single walnut trees according to the combination relationship of time difference and propagation weight, and constructs an abnormal propagation chain relationship. The abnormal diffusion relationship representation and spatial correlation analysis are realized through the propagation score, so as to depict the propagation mode and evolution law of the disease and pests in the tree crown and between the trees. The application improves the fine degree of tree crown damage region recognition and enhances the structural expression capability of the propagation relationship modeling.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method for monitoring walnut tree diseases and pests based on UAV remote sensing data analysis. Background Technology

[0002] In recent years, with the development of unmanned aerial vehicle (UAV) platforms and imaging sensors such as multispectral and hyperspectral sensors, agricultural and forestry vegetation monitoring based on UAV remote sensing data has received widespread attention. UAV remote sensing has the advantages of flexible acquisition cycles, high spatial resolution, and strong operational mobility, and has been gradually applied to crop growth monitoring, leaf nutrient status inversion, and pest and disease identification, providing a new data foundation and technical approach for the spatial distribution and dynamic monitoring of pests and diseases.

[0003] Currently, Chinese invention patent application CN115841621A discloses a tea tree disease identification system based on UAV remote sensing data. This system combines UAV remote sensing technology with tea tree disease control, utilizing UAVs to collect hyperspectral image data of the tea tree canopy. The collected image data is then corrected and processed. Based on the reflectance data measured by hyperspectral remote sensing, the chlorophyll content and leaf area index of the tea tree canopy are calculated. By comparing and analyzing the spectral characteristics of healthy and diseased tea leaves, the system can effectively predict the status of pests and diseases in tea gardens in advance. This can meet the need for high-frequency and regular monitoring of tea tree pest and disease information and improve the efficiency of tea tree pest and disease surveys. However, the relevant technologies lack a method for identifying pest and disease anomalies at the individual tree scale, specifically for economic forest trees like walnut trees, which constructs multi-temporal features and performs joint analysis of temporal and spatial deviations. They also fail to combine the abnormal information of a single walnut tree with its spatial relationship within the orchard, construct an inter-tree adjacency network, and calculate inter-tree propagation weights to analyze the transmission chain and spread risk of pests and diseases. Furthermore, they do not utilize canopy structure indicators and canopy texture indicators such as canopy porosity, canopy patches, and grayscale texture at the individual tree level to finely distinguish pest and disease damage patterns. Summary of the Invention

[0004] The technical problem solved by this invention is that existing technologies are difficult to identify abnormal pests and diseases using multi-temporal remote sensing data from UAVs at the scale of a single walnut tree in a timely and accurate manner. They are also difficult to simultaneously consider the impact of temporal changes and spatial neighborhood differences on the occurrence process of pests and diseases, difficult to distinguish the canopy damage patterns caused by different types of pests and diseases, and lack a systematic method for reconstructing the spread path of pests and diseases in walnut orchards and conducting spread risk assessment based on abnormal information of a single tree.

[0005] To address the aforementioned technical problems, this invention provides the following technical solution: a method for monitoring walnut tree diseases and pests based on UAV remote sensing data analysis, comprising the following steps: Step S1: Obtain multi-temporal UAV remote sensing image sequences, perform unified processing on the multi-temporal UAV remote sensing image sequences, and construct a list of individual walnut trees; Step S2: Extract the features of each monitoring time based on the list of individual walnut trees and construct the feature time series. Establish a health reference, calculate the time deviation index and spatial deviation index and combine them to obtain the time series of abnormal index of individual walnut trees, and determine the set of abnormal individual walnut trees. Step S3: Extract the canopy structure indicators and canopy texture indicators at each monitoring time based on the set of abnormal single walnut trees, construct the corresponding time series, perform joint analysis, and determine the damage mode of each abnormal single walnut tree. Step S4: Establish an inter-tree adjacency network based on the list of individual walnut trees, calculate the inter-tree propagation weights and construct propagation chains, iteratively update the risk value of individual walnut trees, obtain the final risk value of individual walnut trees, and form a risk distribution.

[0006] As a preferred embodiment of the walnut tree pest and disease monitoring method based on UAV remote sensing data analysis described in this invention, step S1 specifically includes: Step S11: Obtain a multi-temporal UAV remote sensing image sequence covering the walnut orchard area. Perform radiometric and geometric corrections on each UAV remote sensing image in the multi-temporal UAV remote sensing image sequence. Register each UAV remote sensing image to a unified coordinate reference system to obtain a standardized remote sensing image sequence. Step S12: Use altitude and spectral information to segment the UAV remote sensing images at each monitoring time in the standardized remote sensing image sequence to obtain the set of canopy regions of a single walnut tree at each monitoring time.

[0007] As a preferred embodiment of the walnut tree pest and disease monitoring method based on UAV remote sensing data analysis described in this invention, step S1 further includes: Step S13: Based on the spatial location of the canopy area of ​​each walnut tree in the set of canopy areas of each walnut tree at each monitoring time, perform spatial location matching between different monitoring times, associate the canopy areas of the walnut trees corresponding to the spatial locations at different monitoring times as the same walnut tree, assign a unique walnut tree identifier to each walnut tree, and construct a list of walnut trees.

[0008] As a preferred embodiment of the walnut tree pest and disease monitoring method based on UAV remote sensing data analysis described in this invention, step S2 specifically includes: Step S21: For each single walnut tree in the list of single walnut trees, at each monitoring time of the standardized remote sensing image sequence, extract vegetation index features and texture features from the canopy area of ​​the single walnut tree corresponding to that monitoring time. Arrange the vegetation index features and texture features of the same single walnut tree at all monitoring times in chronological order to obtain the feature time series of the single walnut tree. Step S22: In the list of individual walnut trees, select a portion of individual walnut trees that meet the preset health judgment logic as a set of healthy candidate individual walnut trees based on the value range and time stability of the characteristic time series of each individual walnut tree. For the characteristic time series of each walnut tree in the set of healthy candidate walnut trees, each characteristic is statistically analyzed at each monitoring time to obtain the mean and standard deviation of the characteristic at that monitoring time. The mean at each monitoring time is used to form the healthy reference growth curve of the characteristic, and the standard deviation at each monitoring time is used to form the healthy reference discrete curve of the characteristic. Step S23: For each walnut tree in the list of individual walnut trees, at each monitoring time and for each feature, based on the mean and standard deviation of the feature value of that individual walnut tree at that monitoring time and the corresponding healthy reference growth curve, calculate the time deviation index. The calculation formula is as follows: ; in, A single walnut tree During monitoring time Features Time deviation indicator, A single walnut tree During monitoring time Features Feature values, Features During monitoring time The mean of the healthy reference growth curve, Features During monitoring time The standard deviation of the health reference discrete curve.

[0009] As a preferred embodiment of the walnut tree pest and disease monitoring method based on UAV remote sensing data analysis described in this invention, step S2 further includes: Step S24: At each monitoring time of the standardized remote sensing image sequence, based on the spatial location information of each walnut tree in the list of individual walnut trees, for each individual walnut tree, a spatial neighborhood set is formed by M spatially adjacent individual walnut trees according to a preset spatial adjacency logic; for each individual walnut tree, at each monitoring time and for each feature, a spatial deviation index is calculated based on the feature value of that feature of that individual walnut tree at that monitoring time and the mean and standard deviation of that feature of each individual walnut tree in its spatial neighborhood set. The calculation formula is as follows: ; in, A single walnut tree During monitoring time Features Spatial deviation index Features of individual walnut trees in a spatial neighborhood set During monitoring time The mean, Characteristics of individual walnut trees in a spatial neighborhood set During monitoring time Standard deviation; Step S25: For each single walnut tree in the list of single walnut trees, at each monitoring time, the time deviation index and spatial deviation index of the single walnut tree at that monitoring time are combined according to a preset combination logic to obtain the single walnut tree abnormality index at that monitoring time. The single walnut tree abnormality indexes at each monitoring time are arranged in chronological order to form a time series of single walnut tree abnormality indexes. Step S26: According to the preset anomaly judgment logic, analyze the time series of the anomaly index of a single walnut tree. For a single walnut tree that meets the preset anomaly judgment logic, determine it as an abnormal single walnut tree. The monitoring time when the preset anomaly judgment logic is first met is determined as the anomaly start time of the abnormal single walnut tree. The anomaly index of the single walnut tree after the anomaly start time in the time series of the single walnut tree anomaly index is calculated according to the preset aggregation logic to obtain the anomaly intensity index. All abnormal single walnut trees are formed into an abnormal single walnut tree set.

[0010] As a preferred embodiment of the walnut tree pest and disease monitoring method based on UAV remote sensing data analysis described in this invention, step S3 specifically includes: step S31, for each abnormal walnut tree in the abnormal single tree set, the single tree crown area corresponding to each monitoring time is obtained for the abnormal single tree within the entire monitoring time range, and the single tree crown area corresponding to each monitoring time is spatially divided to obtain an image sub-region set; Among them, the time of the abnormal start time is used as the time dividing point, and each monitoring time is divided into the monitoring time before the abnormal start time and the monitoring time after the abnormal start time. Step S32: For each abnormal single walnut tree, at each monitoring time, perform morphological operations on each image sub-region in the image sub-region set corresponding to the abnormal single walnut tree at that monitoring time, and separate the brightness structure in the image sub-region, and extract the canopy structure index that characterizes the canopy porosity feature and canopy patch feature. The canopy structure indicators include the ratio of the area of ​​the canopy pore region to the area of ​​the image sub-region, the statistical value of the canopy patch region area, and the spatial concentration of the canopy patch region. The ratio of the area of ​​the canopy pore region to the area of ​​the image sub-region, the statistical value of the area of ​​the canopy patch region, and the spatial concentration of the canopy patch region are used as the canopy structure indicators of the image sub-region at the monitoring time. The crown structure indices of all image sub-regions of the same abnormal single walnut tree at each monitoring time are summarized to obtain the time series of crown structure indices of the abnormal single walnut tree.

[0011] As a preferred embodiment of the walnut tree pest and disease monitoring method based on UAV remote sensing data analysis described in this invention, the processing logic for morphological operations based on image brightness information includes: The brightness of the image sub-regions is normalized to obtain a normalized brightness image; Perform opening and closing operations on the normalized brightness image to separate the bright and dark regions. The low-brightness areas are binarized and connected component analysis is performed to identify the connected components belonging to the canopy pore region. The area of ​​each connected component is calculated, and the areas of all canopy pore regions are summed to obtain the total area of ​​the canopy pore region. The highlighted areas are binarized and connected component analysis is performed to identify the connected components belonging to the canopy patch areas. The area and spatial location of each canopy patch area are statistically analyzed, and the area statistics and spatial concentration of the canopy patch areas are calculated respectively.

[0012] As a preferred embodiment of the walnut tree pest and disease monitoring method based on UAV remote sensing data analysis described in this invention, step S3 further includes: Step S33: For each abnormal single walnut tree, at each monitoring time, the gray-level co-occurrence matrix features are calculated using the gray-level co-occurrence matrix and the local binary pattern features are calculated using the local binary pattern. The gray-level co-occurrence matrix features and the local binary pattern features are used as the crown texture index of the abnormal single walnut tree at that monitoring time. The crown texture indexes at each monitoring time are arranged in chronological order to obtain the crown texture index time series of the abnormal single walnut tree. Step S34: For each abnormal single walnut tree, the time series of the single walnut tree abnormality index, the time series of the crown structure index, and the time series of the crown texture index of the abnormal single walnut tree are jointly analyzed. Based on the time of the abnormality start time, the time series are compared in segments to obtain the changes in the single walnut tree abnormality index, the ratio of the area of ​​the crown pore region to the area of ​​the image sub-region, the spatial concentration of the crown patch region, and the crown texture index before and after the time of the abnormality start time. Step S35: Based on the preset damage mode determination logic, each abnormal single walnut tree is classified into a corresponding damage mode according to the abnormal index of a single walnut tree, the ratio of the area of ​​the canopy pore area to the area of ​​the image sub-region, the spatial concentration of the canopy patch area, and the change characteristics of the canopy texture index before and after the abnormality start time. Each abnormal single walnut tree is then assigned a damage mode identifier. The damage patterns include insect damage patterns and disease damage patterns.

[0013] As a preferred embodiment of the walnut tree pest and disease monitoring method based on UAV remote sensing data analysis described in this invention, step S4 specifically includes: Step S41: Based on the spatial location information of each walnut tree in the list of individual walnut trees, for each individual walnut tree, determine C spatially adjacent individual walnut trees according to the preset spatial adjacency logic, establish the inter-tree adjacency relationship between each individual walnut tree, and form an inter-tree adjacency relationship network. Step S42: For any two single walnut trees with inter-tree adjacency in the inter-tree adjacency relationship network, calculate the corresponding inter-tree propagation weight based on the spatial location information of the two single walnut trees and in combination with the preset propagation weight calculation rules, and assign the inter-tree propagation weight to each pair of single walnut trees with inter-tree adjacency in the inter-tree adjacency relationship network. Step S43: Based on the anomalous start time and anomalous intensity index of each anomalous walnut tree in the set of anomalous individual walnut trees, under the constraint of inter-tree propagation weights, for any pair of anomalous individual walnut trees with inter-tree adjacency relationships... and abnormal single walnut trees When an abnormal single walnut tree The anomaly started later than the abnormal single walnut tree. When the abnormal start time is determined, the abnormal individual walnut tree is calculated based on the combination of the time difference of the abnormal start time and the inter-tree propagation weight. To an abnormal single walnut tree The spread score; Step S44: For each abnormal walnut tree in the set of abnormal single walnut trees, select the abnormal single walnut tree whose propagation score conforms to the preset selection logic as the candidate propagation source single walnut tree. Associate the candidate propagation source single walnut tree with the abnormal single walnut tree through directed connection. Starting from the abnormal single walnut tree with the earliest abnormal start time, connect each abnormal single walnut tree in sequence along the direction of propagation score from large to small according to the directed connection relationship to form a propagation chain composed of abnormal single walnut trees.

[0014] As a preferred embodiment of the walnut tree pest and disease monitoring method based on UAV remote sensing data analysis described in this invention, step S4 further includes: Step S45: For each single walnut tree in the list of single walnut trees, map the abnormal intensity index of the single walnut tree to the initial risk value of the single walnut tree. For single walnut trees that do not belong to the set of abnormal single walnut trees, set their initial risk value of the single walnut tree to a preset non-abnormal risk value. Discrete iterative calculations are performed on the inter-tree adjacency network. In each iteration, for each individual walnut tree, the risk value of the individual walnut tree in the current iteration step is allocated to the individual walnut trees with inter-tree adjacency according to the inter-tree propagation weight. The risk value of the individual walnut tree in the previous iteration step is combined with the risk value of the individual walnut tree received from the adjacent individual walnut trees in the current iteration step according to the preset combination logic. The risk value of the individual walnut tree is updated. The above discrete iterative calculations are repeated until the preset stopping condition is met, and the final risk value of each individual walnut tree is obtained. Step S46: Associate the spatial location of each individual walnut tree with the final risk value of that individual walnut tree, and construct a risk distribution based on the spatial location information of the individual walnut trees, with the spatial location of the individual walnut trees as discrete points and the final risk value of the individual walnut trees as attribute values.

[0015] The beneficial effects of this invention are as follows: This application innovatively constructs a multi-temporal health reference growth curve at the scale of a single walnut tree, couples temporal deviation indicators with spatial deviation indicators to form a time series of anomaly indices for single walnut trees, and further combines the spatial location information of single walnut trees and the onset time of anomalies to construct a network of inter-tree propagation relationships, realizing an integrated monitoring process from early anomaly identification of single trees to orchard-level propagation risk assessment. By using the anomaly index and propagation score of single walnut trees as core quantification quantities, the temporal deviation of single trees, spatial neighborhood comparison, and inter-tree propagation chains are unified into the same analytical framework. This allows the method to adapt to different walnut orchard layouts and multi-temporal monitoring scenarios without relying on complex model training. It can accurately locate abnormal single trees and characterize the propagation paths and high-risk areas of pests and diseases in walnut orchards, providing direct support for precise pest and disease monitoring and regional control. Attached Figure Description

[0016] Figure 1 The flowchart illustrates the steps of a method for monitoring walnut tree diseases and pests based on UAV remote sensing data analysis, as provided in one embodiment of the present invention. Detailed Implementation

[0017] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0018] Example, refer to Figure 1 This paper provides a method for monitoring walnut tree diseases and pests based on UAV remote sensing data analysis, including the following steps: Step S1: Obtain multi-temporal UAV remote sensing image sequences, perform unified processing on the multi-temporal UAV remote sensing image sequences, and construct a list of individual walnut trees; Step S2: Extract the features of each monitoring time based on the list of individual walnut trees and construct the feature time series. Establish a health reference, calculate the time deviation index and spatial deviation index and combine them to obtain the time series of abnormal index of individual walnut trees, and determine the set of abnormal individual walnut trees. Step S3: Extract the canopy structure indicators and canopy texture indicators at each monitoring time based on the set of abnormal single walnut trees, construct the corresponding time series, perform joint analysis, and determine the damage mode of each abnormal single walnut tree. Step S4: Establish an inter-tree adjacency network based on the list of individual walnut trees, calculate the inter-tree propagation weights and construct propagation chains, iteratively update the risk value of individual walnut trees, obtain the final risk value of individual walnut trees, and form a risk distribution.

[0019] In specific implementation, step S1 includes: Step S11: Obtain a multi-temporal UAV remote sensing image sequence covering the walnut orchard area. Perform radiometric and geometric corrections on each UAV remote sensing image in the multi-temporal UAV remote sensing image sequence. Register each UAV remote sensing image to a unified coordinate reference system to obtain a standardized remote sensing image sequence. Step S12: Use altitude and spectral information to segment the UAV remote sensing images at each monitoring time in the standardized remote sensing image sequence to obtain the set of canopy regions of a single walnut tree at each monitoring time.

[0020] Specifically, acquiring multi-temporal UAV remote sensing image sequences covering a walnut orchard area typically involves planning a flight path for the UAV above the orchard, equipping it with a multispectral or hyperspectral sensor, and repeating the flight at fixed time intervals (e.g., weekly or bi-weekly). This records images at each monitoring moment, along with the corresponding GPS coordinates, attitude angles, and flight altitude, thus forming a time-ordered image sequence. To eliminate inconsistencies between images from different monitoring moments caused by variations in illumination, atmospheric conditions, and UAV attitude fluctuations, each image requires radiometric and geometric correction. Radiometric correction involves converting the grayscale values ​​of the original image to reflectance and using a calibration plate or atmospheric correction model (such as 6S or FLAASH) to eliminate the effects of atmospheric scattering and absorption, obtaining the true surface reflectance. Geometric correction utilizes the UAV's built-in GPS / IMU data and ground control points, employing polynomial transformations or collinearity equation models to correct lens distortion and geometric deformation caused by terrain undulations. After these corrections are completed, all images from the monitoring times need to be registered to a unified coordinate reference system (e.g., WGS84 UTM partition). This is typically achieved using feature matching algorithms (such as SIFT) combined with RANSAC and affine transformations to ensure strict spatial alignment of each image, resulting in a standardized remote sensing image sequence. When performing region segmentation on this basis, both height and spectral information are utilized. Height information is derived from LiDAR onboard the UAV or a normalized digital surface model generated through stereo matching. A height threshold (e.g., greater than 1.5 meters) can be set to initially filter out ground and low vegetation. Spectral information is obtained by calculating the normalized vegetation index, using multispectral bands and the height channel as input, and outputting a binary mask of the canopy and background. Noise is removed through morphological operations, and then connected component analysis is used to label the canopy region of each individual walnut tree, thus obtaining a set of canopy regions for each walnut tree at each monitoring time.

[0021] Traditional methods often neglect radiometric and geometric corrections, leading to inconsistent brightness and misalignment between images across multiple time periods, making subsequent feature extraction unreliable for temporal comparisons. Using only spectral information, tree canopies are easily confused with shadows and weeds, resulting in low segmentation accuracy, especially in young walnut trees or sparsely planted orchards. This application ensures the consistency of multi-time period image sequences through mandatory radiometric and geometric corrections and unified registration, laying a data foundation for subsequent time series calculations of individual tree anomaly indices. Simultaneously, fusing height and spectral information for region segmentation significantly improves the accuracy of canopy identification, effectively excluding ground weeds and enabling automated, batch extraction of individual tree canopies. Thus, step S12 provides a high-quality, repeatable data preprocessing workflow for the entire monitoring method, enabling subsequent steps (such as temporal and spatial deviation analysis and propagation chain construction) to operate stably at the individual tree scale, improving the reliability of early pest and disease identification and propagation risk assessment.

[0022] In specific implementation, step S1 also includes: Step S13: Based on the spatial location of the canopy area of ​​each walnut tree in the set of canopy areas of each walnut tree at each monitoring time, perform spatial location matching between different monitoring times, associate the canopy areas of the walnut trees corresponding to the spatial locations at different monitoring times as the same walnut tree, assign a unique walnut tree identifier to each walnut tree, and construct a list of walnut trees.

[0023] Specifically, the spatial location of the canopy region of each individual walnut tree refers to the geometric center (centroid) coordinates or the geographic coordinates of the bounding box of each independent canopy region in the image at each monitoring time after segmentation in step S12. These coordinates represent the true spatial location in a unified coordinate reference system (e.g., WGS84 UTM), and are usually calculated from the latitude and longitude or the average of the planar coordinates of all pixels in the canopy region. Alternatively, the outline or minimum bounding rectangle of the canopy region can be used to characterize its spatial occupancy. Since the UAV remote sensing images at different monitoring times have undergone radiometric correction, geometric correction, and registration, and are all aligned to the same coordinate reference system, the spatial locations of the canopy regions of the same walnut tree in images at different monitoring times should theoretically be very close (only slight swaying due to tree growth or wind is possible). Spatial location matching between different monitoring times is essentially a multi-temporal data association problem: For each canopy region detected in the image of a later monitoring time, its spatial location is compared with the spatial locations of all canopy regions in the image of the previous monitoring time (or a baseline monitoring time). A nearest neighbor matching strategy is typically used, where a reasonable distance threshold is set (e.g., 1-2 meters, depending on the canopy width and positioning error of the walnut tree). When the centroid distance between two canopy regions is less than this threshold, they are considered to belong to the same walnut tree. If canopy regions from multiple monitoring times are correlated, a "trajectory" is established, and a unique single-tree identifier (e.g., Tree_001, Tree_002, and Tree_003) is assigned to this trajectory. For newly appearing canopy regions that cannot be matched with any existing trajectory, a new identifier is created; for canopy regions missing at a certain monitoring time (e.g., occluded or segmented), the identifier is retained but the data is not filled in. Finally, all the identifiers and the corresponding canopy area geometry and feature values ​​for each identifier at each monitoring time are compiled into a table or structured data object, called the "List of Individual Walnut Trees". This list can be understood as a dictionary or data frame, where each row represents a walnut tree, each column represents a monitoring time, and each cell stores the canopy area outline, centroid coordinates, vegetation index, and texture features of the tree at that monitoring time; if a tree is not observed at a certain monitoring time, it is marked as empty. For example, assuming that three walnut trees are identified through matching at three monitoring times (T1, T2, T3), the constructed list of individual walnut trees is shown in Table 1; Table 1:

[0024] In this table, each row contains a unique walnut tree ID, and each column corresponds to a monitoring time. The cells store the centroid coordinates and key feature values ​​(NDVI for example) of the tree's canopy region at that monitoring time. In practical applications, cells can also store more complex data structures such as the complete outline point set of the canopy region, texture feature vectors, and vegetation index arrays. For canopy regions that were missed due to occlusion or segmentation at a given monitoring time, the corresponding cells are marked as "empty" or "NaN". This organized data structure allows subsequent steps to extract the feature time series of each individual tree along the time axis using the "walnut tree ID" as an index (e.g., the NDVI sequence [0.78, 0.76, 0.55] for Tree_001). Simultaneously, the spatial location in each cell is used to calculate the adjacency relationship between trees (e.g., calculating the distance between two trees using the centroid coordinates), thus providing direct data support for the time deviation index and spatial deviation index in step S2, and the construction of the propagation chain in step S4.

[0025] In specific implementation, step S2 includes: Step S21: For each single walnut tree in the list of single walnut trees, at each monitoring time of the standardized remote sensing image sequence, extract vegetation index features and texture features from the canopy area of ​​the single walnut tree corresponding to that monitoring time. Arrange the vegetation index features and texture features of the same single walnut tree at all monitoring times in chronological order to obtain the feature time series of the single walnut tree. Step S22: In the list of individual walnut trees, select a portion of individual walnut trees that meet the preset health judgment logic as a set of healthy candidate individual walnut trees based on the value range and time stability of the characteristic time series of each individual walnut tree. For the characteristic time series of each walnut tree in the set of healthy candidate walnut trees, each characteristic is statistically analyzed at each monitoring time to obtain the mean and standard deviation of the characteristic at that monitoring time. The mean at each monitoring time is used to form the healthy reference growth curve of the characteristic, and the standard deviation at each monitoring time is used to form the healthy reference discrete curve of the characteristic. Step S23: For each walnut tree in the list of individual walnut trees, at each monitoring time and for each feature, based on the mean and standard deviation of the feature value of that individual walnut tree at that monitoring time and the corresponding healthy reference growth curve, calculate the time deviation index. The calculation formula is as follows: ; in, A single walnut tree During monitoring time Features Time deviation indicator, A single walnut tree During monitoring time Features Feature values, Features During monitoring time The mean of the healthy reference growth curve, Features During monitoring time The standard deviation of the health reference discrete curve.

[0026] Specifically, vegetation index features are numerical indicators obtained from multispectral or hyperspectral reflectance data of walnut tree canopy areas through band calculations. Common ones include the Normalized Difference Vegetation Index (NDVI, calculated using near-infrared and red bands to reflect chlorophyll content and photosynthetic intensity), the Enhanced Vegetation Index (EVI, corrected for atmospheric and soil background effects), and the Ratio Vegetation Index (RVI). These indices can sensitively capture the tree's growth status and stress level. Texture features describe the spatial distribution pattern of pixel brightness within the canopy area. They are usually calculated using the Gray-Level Co-occurrence Matrix (GLCM) to obtain statistics on contrast, correlation, energy, and homogeneity, or extracted using Local Binary Pattern (LBP). Texture features can reflect structural changes such as sparse leaves, spots, and withered areas, which are often closely related to canopy damage caused by pests and diseases.

[0027] The range of values ​​and the stability of changes over time are two core criteria for screening healthy candidate walnut trees. The range of values ​​refers to the interval between the minimum and maximum values ​​of a feature (e.g., NDVI) at all monitoring times throughout the entire monitoring period. The NDVI of healthy walnut trees is typically stable within a relatively high range (e.g., 0.7–0.9), while the values ​​of abnormal trees may be lower or fluctuate drastically. The stability of changes over time measures the magnitude of fluctuations in the feature time series of the same tree between different monitoring times. It can be quantified using variance, standard deviation, coefficient of variation, or the sum of the absolute values ​​of the differences between adjacent monitoring times. A highly stable series means that the tree did not experience sudden increases or decreases throughout the monitoring period, consistent with the expectation of healthy growth.

[0028] The pre-defined health assessment logic combines these two dimensions to select the most likely healthy individuals as a reference. First, the mean, variance, and range of each feature for each tree in the list of individual walnut trees are calculated over the entire time series. Then, thresholds are set, such as requiring a mean NDVI greater than 0.7, a variance less than 0.01, and a lower limit of NDVI values ​​not lower than 0.6. Only trees that simultaneously meet these conditions are included in the healthy candidate set. The health reference curve must originate from truly healthy trees. If individuals already affected by pests and diseases are included, it will cause the reference curve to shift, reducing the sensitivity of subsequent anomaly detection. By selecting individuals whose feature values ​​are stable and at a high level over a long period, the representativeness and reliability of the reference curve can be maximized.

[0029] After identifying the set of healthy candidate walnut trees, statistics were compiled for each tree in the set at each monitoring time and for each characteristic: For a fixed monitoring time and a fixed characteristic, the characteristic values ​​of all trees in the healthy candidate set at that monitoring time were collected to form a numerical list, and then the arithmetic mean and sample standard deviation of this list were calculated. Since the trees in the healthy candidate set are considered to be in a healthy state throughout the monitoring period, these statistics can reflect "the average level of characteristics that a normal walnut tree should have at that monitoring time and the natural fluctuation range between individuals." Next, the arithmetic means calculated from all monitoring times were connected in chronological order to obtain the healthy reference growth curve of the characteristics, which describes the average trend of the healthy population from the first monitoring time to the last monitoring time (e.g., NDVI increases in spring, stabilizes in summer, and decreases in autumn). Similarly, the sample standard deviations of all monitoring times were connected in chronological order to obtain the healthy reference dispersion curve, which reflects the degree of dispersion within the healthy population at different times (e.g., large individual differences in the early growth stage, and tending to be consistent during the vigorous growth stage).

[0030] The time deviation metric was calculated using a formula in the Z-score normalized form. Instead of using absolute difference, relative error, or percentiles, the Z-score eliminates differences in dimensions and absolute value ranges between different features, allowing the deviations of NDVI (typically 0.2–0.9) and texture features (potentially tens to hundreds) to be compared on the same scale. Secondly, it considers both the difference between an individual and the healthy mean, as well as the sample standard deviation of the natural fluctuation range of the healthy population. When the sample standard deviation is large (i.e., there are inherently large differences among healthy trees), even a large absolute difference may not be considered abnormal. Conversely, when the sample standard deviation is small, small deviations will be amplified, which aligns with the principle in actual biological monitoring that "deviations in stable populations are more alarming." Thirdly, the formula preserves the direction of deviation (positive values ​​indicate features above the healthy mean, negative values ​​indicate below the healthy mean). Deviations in different directions may correspond to different types of stress (e.g., diseases often lead to a decrease in NDVI, while certain insect pests may cause abnormal increases in specific indices due to reddening of the canopy in the early stages).

[0031] Traditional methods often set a fixed NDVI threshold (e.g., below 0.5 is considered abnormal), but ignore the characteristic changes of healthy trees at different growth stages and in different seasons. For example, the NDVI is naturally lower when walnut leaves are tender in spring, and using a fixed threshold at this time will produce a large number of false alarms. This application constructs a health reference growth curve that changes with the monitoring time, dynamically adapting to the phenological stage of walnut trees, so that the definition of "abnormal" is linked to the health status of the population at the current monitoring time, which greatly improves the sensitivity of early weak anomaly identification. At the same time, the sensitivity of deviation index is adjusted by using the health reference discrete curve, which enhances the detection capability during periods of strong population consistency and avoids false alarms during periods of large natural population differences. In addition, the time deviation index is used as the basis for subsequent anomaly index combinations, providing a standardized input for multi-feature joint judgment, so that the entire monitoring method does not depend on specific features or specific thresholds, and has strong generalization and robustness. Compared with traditional methods that only use spatial neighborhood comparison or only use single phase detection, this application has taken the lead in realizing "temporal dynamic anomaly detection" of walnut tree diseases and pests by using health reference modeling in the time dimension. It can capture two modes: slow deterioration and sudden change, laying a quantitative foundation for determining the anomaly start time and anomaly intensity index in step S2.

[0032] In specific implementation, step S2 also includes: Step S24: At each monitoring time of the standardized remote sensing image sequence, based on the spatial location information of each walnut tree in the list of individual walnut trees, for each individual walnut tree, according to the preset spatial adjacency logic, determine the M individual walnut trees that are spatially adjacent to the individual walnut tree to form a spatial neighborhood set.

[0033] For each individual walnut tree, at each monitoring time and for each feature, the spatial deviation index is calculated based on the feature value of that individual walnut tree at that monitoring time and the mean and standard deviation of that feature of all individual walnut trees in its spatial neighborhood set. The calculation formula is as follows: ; in, A single walnut tree During monitoring time Features Spatial deviation index Features of individual walnut trees in a spatial neighborhood set During monitoring time The mean, Characteristics of individual walnut trees in a spatial neighborhood set During monitoring time Standard deviation; Step S25: For each single walnut tree in the list of single walnut trees, at each monitoring time, the time deviation index and spatial deviation index of the single walnut tree at that monitoring time are combined according to a preset combination logic to obtain the single walnut tree abnormality index at that monitoring time. The single walnut tree abnormality indexes at each monitoring time are arranged in chronological order to form a time series of single walnut tree abnormality indexes. Step S26: According to the preset anomaly judgment logic, analyze the time series of the anomaly index of a single walnut tree. For a single walnut tree that meets the preset anomaly judgment logic, determine it as an abnormal single walnut tree. The monitoring time when the preset anomaly judgment logic is first met is determined as the anomaly start time of the abnormal single walnut tree. The anomaly index of the single walnut tree after the anomaly start time in the time series of the single walnut tree anomaly index is calculated according to the preset aggregation logic to obtain the anomaly intensity index. All abnormal single walnut trees are formed into an abnormal single walnut tree set.

[0034] Specifically, the spatial location information refers to the centroid coordinates or bounding box geographic coordinates of the canopy area of ​​each walnut tree after segmentation in step S12. These coordinates are located in a unified coordinate reference system (such as WGS84 UTM) and can accurately express the true distribution location of each tree in the orchard.

[0035] The pre-defined spatial adjacency logic is used to determine the M trees that are spatially adjacent to the target walnut tree, forming its spatial neighborhood set. Commonly used spatial adjacency logics include distance-based K-nearest neighbor or fixed-radius neighborhood methods. For example, a distance threshold is set (usually 1.5 times the average crown width of the walnut tree, approximately 3 to 5 meters), and all trees whose centroid distance from the target tree is less than this threshold are included in the neighborhood. Alternatively, the M nearest trees (M is generally 4 to 8) can be directly selected. The spread of pests and diseases in orchards often exhibits spatial continuity. The probability of neighboring trees influencing each other through root contact, insect migration, and wind and rain is significantly higher than that of distant trees. Therefore, spatial neighborhood comparison can eliminate regional environmental differences (such as uneven soil fertility, local irrigation differences, or light shading), thereby highlighting the abnormal condition of individual trees. If spatial adjacency is not used and the average value of the entire orchard is used directly as a reference, a large number of misjudgments may occur due to the inherent spatial heterogeneity of the orchard.

[0036] The pre-defined combination logic refers to the rule of merging the temporal deviation index and spatial deviation index of each walnut tree at the same monitoring time into a single-tree anomaly index. The combination logic includes taking the maximum value of the two, a weighted sum, or the Euclidean norm. For example, the anomaly index can be defined as the square root of the sum of the squares of the absolute values ​​of the temporal and spatial deviation indices, or simply taking the larger of the two absolute values. Alternatively, weighting coefficients (e.g., 0.6 and 0.4) can be assigned to temporal and spatial deviations based on prior knowledge before summing. The temporal deviation index reflects the tree's temporal difference relative to the overall healthy growth trend. It may be insensitive to stresses occurring synchronously throughout the orchard (such as drought, frost damage, or seasonal changes) because the health reference curve itself will also decline. The spatial deviation index, on the other hand, reflects the tree's immediate difference relative to its neighbors. It may be insensitive to large-scale, evenly distributed diseases because neighbors are also in similar abnormal states. By combining the two, the system can simultaneously identify global anomalies (prominent temporal deviation) and localized point-like anomalies (prominent spatial deviation), compensating for the shortcomings of a single indicator and thus improving the robustness of monitoring. In practice, the absolute values ​​of time deviation and spatial deviation can be taken or the signs can be retained. Then, the abnormal index of each monitoring moment can be calculated according to the preset formula, and then arranged into a time series in chronological order.

[0037] The preset anomaly detection logic is used to determine whether a walnut tree is an abnormal individual and to identify the onset time of its anomaly. Commonly used detection logics include threshold methods, duration methods, or statistical control chart methods. For example, it can be set that if the anomaly index of a single walnut tree exceeds 2.0 (corresponding to the significance level of the Z-score) for two consecutive monitoring times, or exceeds the threshold three times cumulatively within a period of time, it is considered an anomaly, and the monitoring time that first meets this condition is recorded as the anomaly onset time. If the anomaly index of a single walnut tree never meets the conditions of the preset anomaly detection logic, then the single walnut tree is considered a healthy individual. This design is because a single, accidental measurement fluctuation (such as changes in drone attitude, light flicker, registration error) may cause the anomaly index to temporarily exceed the standard. If anomalies are determined based on a single observation, a large number of false alarms will be generated. Therefore, a duration or multiple triggering conditions are needed to filter out noise. At the same time, accurately determining the anomaly onset time is crucial for the propagation chain analysis in the subsequent step S4, because only by knowing the earliest time when anomalies appeared on each tree can the direction and speed of pest and disease propagation be inferred.

[0038] The pre-defined aggregation logic is used to calculate a comprehensive indicator representing the severity of the anomaly, i.e., the anomaly intensity index, from the anomaly index time series after the anomaly's onset. The aggregation logic can take the average, maximum, or area under the integral (i.e., the time integral under the anomaly index curve) of all anomaly indices after the anomaly's onset. For example, the anomaly intensity index can be set as the arithmetic mean of the anomaly indices at each time point from the anomaly's onset to the last monitoring time, multiplied by the anomaly's duration. The purpose of this is to condense the anomaly's temporal dynamics into a single value, facilitating the calculation of the propagation score and initial risk value in subsequent step S4. Using only the single-point anomaly index at the onset may underestimate the subsequent continuous deterioration; using only the maximum value may ignore anomalies that have been at a low level for a long time but have a significant cumulative impact. Therefore, a reasonable aggregation logic needs to balance the instantaneous intensity of the peak and the duration of the anomaly, and is usually pre-defined based on the actual data distribution and business needs.

[0039] This application introduces a synergistic mechanism of spatial deviation index and temporal deviation index in the monitoring of walnut tree diseases and pests for the first time, and constructs a time series of abnormal index for a single walnut tree through combinatorial logic, thereby realizing the automatic determination of the start time of the abnormality and the quantification of the abnormality intensity. Compared with traditional methods that rely solely on time series analysis or spatial neighborhood comparison, this approach has advantages in three main aspects: First, spatial deviation indicators can eliminate false positives caused by local environmental differences within the orchard (such as uneven terrain, soil, and irrigation), making the monitoring results more focused on the true health status of individual trees. Temporal deviation indicators, on the other hand, eliminate the overall drift caused by seasonal and phenological changes. The two complement each other, significantly reducing the false alarm rate. Second, the dual reference system of time and space enables the method to adapt to stresses at different scales (orchard-wide disasters and localized outbreaks of pests and diseases), eliminating the need to retrain complex models for different scenarios and greatly improving the generalization and transferability of the method. Third, the precise location of the anomaly initiation time and the aggregated calculation of the anomaly intensity indicators provide reliable basic data for the subsequent construction of the transmission chain and risk diffusion analysis in step S4, forming a closed loop from individual tree anomaly identification to orchard-level transmission path reconstruction. Compared to traditional manual visual interpretation or fixed threshold remote sensing monitoring methods, this method significantly improves the automation, spatiotemporal sensitivity, and anti-interference capabilities of anomaly detection, and is a core technical component for achieving the monitoring objective of "early detection, accurate location, and intensity determination."

[0040] In specific implementation, step S3 includes: Step S31: For each abnormal single walnut tree in the set of abnormal single walnut trees, obtain the single walnut tree canopy area corresponding to each monitoring time within the entire monitoring time range, and spatially divide the single walnut tree canopy area corresponding to each monitoring time to obtain a set of image sub-regions. Among them, the time of the abnormal start time is used as the time dividing point, and each monitoring time is divided into the monitoring time before the abnormal start time and the monitoring time after the abnormal start time. Step S32: For each abnormal single walnut tree, at each monitoring time, perform morphological operations on each image sub-region in the image sub-region set corresponding to the abnormal single walnut tree at that monitoring time, and separate the brightness structure in the image sub-region, and extract the canopy structure index that characterizes the canopy porosity feature and canopy patch feature. The canopy structure indicators include the ratio of the area of ​​the canopy pore region to the area of ​​the image sub-region, the statistical value of the canopy patch region area, and the spatial concentration of the canopy patch region. The ratio of the area of ​​the canopy pore region to the area of ​​the image sub-region, the statistical value of the area of ​​the canopy patch region, and the spatial concentration of the canopy patch region are used as the canopy structure indicators of the image sub-region at the monitoring time. The crown structure indices of all image sub-regions of the same abnormal single walnut tree at each monitoring time are summarized to obtain the time series of crown structure indices of the abnormal single walnut tree.

[0041] Specifically, spatial division of the canopy area of ​​a single walnut tree at each monitoring time is necessary because the distribution of pests and diseases on the canopy is often uneven; some areas may be severely damaged while others are relatively healthy. Analyzing the entire canopy as a whole would lose detailed information about localized damage. The spatial division involves dividing the canopy area into multiple sub-regions according to certain rules. Common methods include: dividing the canopy into concentric rings (inner, middle, and outer rings) based on the radius from the centroid; dividing it into several sectors according to azimuth (e.g., one sector every 45 degrees); or uniformly dividing the canopy into grid-like sub-regions. After division, a set of image sub-regions for the abnormal tree at that monitoring time is obtained, and each sub-region will be analyzed independently. The monitoring time refers to the time series sampling time corresponding to the multi-temporal data acquisition process for a single walnut tree.

[0042] Image brightness information refers to the grayscale value or brightness channel value of each pixel in a standardized remote sensing image. In multispectral images, the near-infrared band or the green band is usually selected because these bands are more sensitive to the physiological state and structural changes of leaves. Canopy pore areas appear as dark areas with low brightness in the image because these areas lack leaf cover and may directly show ground shadows or lower dead branches, resulting in low reflectivity. On the other hand, canopy patch areas appear as bright areas with high brightness, usually corresponding to lesions, discolored leaves due to insect bites, yellowed tissue, or areas of mold growth. These tissues show abnormally high or low reflectivity at specific wavelengths. The specific process of morphological operations based on image brightness information is as follows: First, the brightness of the image sub-regions is normalized, stretching the brightness values ​​to the range of 0 to 255 to obtain a normalized brightness image. Then, opening and closing operations are performed on the image sequentially. The opening operation (erosion followed by dilation) can remove isolated noise points and small spots, while the closing operation (dilation followed by erosion) can fill small gaps and smooth the edges of bright areas. By adjusting the size of the structuring element (e.g., using a circular structuring element with a radius of 3 to 5 pixels), the bright and low-bright areas can be separated. After completing the morphological operations, the separated bright and low-bright areas are binary segmented, i.e., a brightness threshold is set, with pixels above the threshold marked as 1 (bright area) and pixels below the threshold marked as 0 (low-bright area). Next, connected component analysis is performed on the binarized image, using an eight-neighbor or four-neighbor search algorithm to find all connected pixel blocks, and the area and spatial location of each connected component are calculated. In low-brightness areas, larger connected regions typically correspond to genuine canopy pores (i.e., cavities caused by missing leaves), while isolated dark spots with excessively small areas may be noise. These can be filtered using a preset area threshold (e.g., connected regions smaller than 0.5% of the total canopy area are discarded). The final identified effective low-brightness connected regions are the canopy pore regions. In high-brightness areas, the identified connected regions are the canopy patch regions, with each patch representing a local abnormal tissue. Next, three canopy structure indicators are calculated: the area of ​​the pore region and the ratio of the canopy pore region area to the image sub-region area, i.e., porosity, reflecting the degree of leaf loss within that sub-region; canopy patch region area statistics, such as the total area, average area, or area variance of the patches, used to quantify the severity and uniformity of the patch distribution; and the spatial concentration of the canopy patch region, typically measured by the reciprocal of the average distance between patches and the ratio of the minimum circumscribed circle radius of the patch distribution to the total patch area, to determine whether the patches are clustered together or dispersed.Finally, the canopy structure indices of all image sub-regions of the same anomalous walnut tree at each monitoring time are summarized. The summarization method involves statistically aggregating the porosity, patch area statistics, and spatial concentration of all sub-regions of the tree at each monitoring time. For example, the average, maximum, or weighted sum of porosity across all sub-regions is calculated, and the patch indices are processed similarly. The aggregated values ​​at each time point are then arranged chronologically to form a time series of canopy structure indices for the anomalous tree. In this way, each anomalous tree has porosity curves, patch area curves, and patch concentration curves that change over time from the onset of the anomaly. These curves reflect the spread and deterioration of pests and diseases within the canopy, providing crucial structural evidence for distinguishing between pest damage patterns and disease damage patterns in subsequent steps S34 and S35. This anomaly analysis delves from the overall characteristics of the entire canopy to the structural details within the canopy. Through morphological identification of porosity and patches, it can quantitatively characterize the types of physical damage caused by pests and diseases—a depth that traditional methods using only vegetation indices or texture features cannot achieve, laying a solid foundation for refined differentiation of damage patterns.

[0043] In specific implementation, the processing logic for morphological operations based on image brightness information includes: The brightness of the image sub-regions is normalized to obtain a normalized brightness image; Perform opening and closing operations on the normalized brightness image to separate the bright and dark regions. The low-brightness areas are binarized and connected component analysis is performed to identify the connected components belonging to the canopy pore region. The area of ​​each connected component is calculated, and the areas of all canopy pore regions are summed to obtain the total area of ​​the canopy pore region. The highlighted areas are binarized and connected component analysis is performed to identify the connected components belonging to the canopy patch areas. The area and spatial location of each canopy patch area are statistically analyzed, and the area statistics and spatial concentration of the canopy patch areas are calculated respectively.

[0044] Specifically, the processing logic for morphological operations based on image brightness information includes opening and closing operations, binarization, and connected component analysis as core operations. When performing opening and closing operations on a normalized brightness image, a structuring element needs to be defined first. Typically, a disk-shaped or square convolution kernel is chosen, with its radius or side length determined based on the image resolution and the expected scale of the canopy openings. For example, in an image with a spatial resolution of 5 cm, a circular structuring element with a radius of 3 to 5 pixels can be used. The opening operation involves first performing an erosion operation on the image, sliding the structuring element across the image. Only when all pixels in the area completely covered by the structuring element are highlighted is the center pixel retained; otherwise, it is set to a low value. This step eliminates isolated tiny bright spots and fine bright lines. After erosion, a dilation operation is performed on the image. When the structuring element slides across the area, as long as there is one highlighted pixel, the center pixel is highlighted, thus restoring the edges that were eroded but originally belonged to larger bright areas. The opening operation removes noise, smooths bright area boundaries, and separates adhered bright spots. The closing operation, on the other hand, first dilates and then erodes. Dilation fills the tiny dark holes inside the bright areas and connects adjacent bright spots, while erosion restores the expanded boundaries after dilation, ultimately filling the pores inside the bright areas and smoothing the edges. By combining opening and closing operations, the complex brightness patterns in tree canopy images caused by leaf shading, light and shadow, and lesions can be decomposed into relatively complete bright and dark areas.

[0045] It should be noted that canopy pores (where leaves are missing) are generally considered to appear as dark areas in images (due to shadows or background soil), while canopy patches (disease spots or dead leaves) may appear as bright areas or areas of abnormal reflection under specific wavelength conditions. Therefore, this application explicitly adopts a unified brightness mapping rule: low brightness areas correspond to canopy pore areas, and high brightness areas correspond to canopy patch areas.

[0046] When binarizing low-brightness areas, a brightness threshold needs to be set. This threshold is typically determined automatically using Otsu's maximum inter-class variance method, or a fixed threshold can be set empirically (e.g., pixels with brightness values ​​below a certain threshold are considered foreground). After binarization, pixels within the low-brightness area are marked as 1, and the background as 0. Connected component analysis is then performed, using an eight-neighbor search algorithm to find all connected pixel blocks, with each connected component considered a candidate target. For connected components belonging to the canopy pore region, filtering is required based on their area, shape, and location. For example, noisy connected components with areas too small (less than 0.1% of the total canopy area) and pseudo-pores that are too long and narrow (aspect ratio greater than 5) and may be caused by tree branch shadows are excluded. The remaining connected components are considered true canopy pores. The area of ​​each pore connected component is then calculated (i.e., the number of pixels multiplied by the actual area of ​​a single pixel), and all pore areas are summed to obtain the total area of ​​the canopy pore region.

[0047] The highlighted areas are also binarized and connected component analyzed, but the target for identification is the canopy patch region. Similar to porosity identification, an area threshold needs to be set to filter out noise, and morphological features (such as roundness and compactness) are used for further selection. For each identified patch connected component, not only its area is calculated, but its spatial location is also recorded, such as the centroid coordinates, minimum bounding rectangle, or convex hull of the patch. Based on this area and location information, canopy patch area statistics can be calculated, such as total patch area, average area, median area, or area variance, reflecting the severity of the patches; at the same time, the spatial concentration of canopy patch regions is calculated. Common calculation methods include: the ratio of the minimum convex hull area of ​​the patch distribution to the total canopy area (the smaller the ratio, the more concentrated the patches), the average distance from the centroid of all patches to the centroid of the canopy (the smaller the distance, the more concentrated the patches), or the nearest neighbor index (the ratio of the average nearest neighbor distance to the expected distance of the random distribution, less than 1 indicates a clustered distribution). These indicators comprehensively reflect the spatial distribution patterns of pests and diseases within the tree canopy, providing a quantitative basis for subsequently distinguishing between pests (which typically cause scattered small patches) and diseases (which often form clustered large patches).

[0048] This approach integrates morphological operations, binarization segmentation, and connected component analysis into a complete processing chain. It abandons the crude methods of traditional approaches that rely on subjective visual interpretation or simple thresholding. Opening and closing operations effectively suppress the effects of image noise and uneven illumination, making the extraction of pores and patches more robust. Simultaneously, connected component analysis allows for the statistical analysis of individual attributes of each pore and patch, leading to the calculation of physically meaningful area statistics and spatial concentration. These quantitative indicators accurately characterize the structural damage features caused by different types of pests and diseases on the tree canopy, providing direct evidence for damage pattern determination in steps S34 and S35. Without such refined processing, it would be difficult to distinguish between the two drastically different pathological manifestations of "uniform leaf yellowing" and "localized leaf loss and spot aggregation" based solely on overall vegetation indices or texture features.

[0049] In specific implementation, step S3 also includes: Step S33: For each abnormal single walnut tree, at each monitoring time, the gray-level co-occurrence matrix features are calculated using the gray-level co-occurrence matrix and the local binary pattern features are calculated using the local binary pattern. The gray-level co-occurrence matrix features and the local binary pattern features are used as the crown texture index of the abnormal single walnut tree at that monitoring time. The crown texture indexes at each monitoring time are arranged in chronological order to obtain the crown texture index time series of the abnormal single walnut tree. Step S34: For each abnormal single walnut tree, the time series of the single walnut tree abnormality index, the time series of the crown structure index, and the time series of the crown texture index of the abnormal single walnut tree are jointly analyzed. Based on the time of the abnormality start time, the time series are compared in segments to obtain the changes in the single walnut tree abnormality index, the ratio of the area of ​​the crown pore region to the area of ​​the image sub-region, the spatial concentration of the crown patch region, and the crown texture index before and after the time of the abnormality start time. Step S35: Based on the preset damage mode determination logic, each abnormal walnut tree is classified into a corresponding damage mode according to the abnormal index of a single walnut tree, the ratio of the area of ​​the canopy pore area to the area of ​​the image sub-region, the spatial concentration of the canopy patch area, and the change characteristics of the canopy texture index before and after the abnormality. Each abnormal walnut tree is then assigned a damage mode identifier. The damage modes include insect damage mode and disease damage mode.

[0050] Specifically, the texture features extracted in step S21 are global texture statistics calculated at the scale of the entire canopy region, used in the anomaly detection stage to determine whether the overall health status of the tree deviates from the normal range; while the canopy texture index extracted in step S33 are local texture features calculated at the scale of the image sub-regions divided in step S31, capable of capturing texture differences and spatial distribution patterns in different areas within the canopy, used for refined differentiation of damage patterns. The calculation objects and analysis purposes of the two are different.

[0051] By introducing canopy texture indices and conducting joint analysis, a fine distinction between insect damage patterns and disease damage patterns can be achieved. The gray-level co-occurrence matrix (GLCM) is a statistical method based on the spatial correlation of image pixel gray-level values. It constructs a square matrix by calculating the frequency of gray-level value pairs between two pixels in an image with specific spatial relationships (e.g., horizontally adjacent, vertically adjacent, or diagonally adjacent). The number of rows and columns of the matrix equals the number of gray levels in the image (usually the original image is quantized to 16 or 32 levels to reduce computation). Using the GLCM, a series of statistical quantities reflecting image texture characteristics can be calculated. Commonly used quantities include contrast (reflecting the clarity and depth of grooves; higher contrast indicates coarser texture), energy (also known as the second angular moment, reflecting the uniformity of texture; higher energy indicates more regular texture), homogeneity (measuring the uniformity of local gray-level distribution; higher homogeneity indicates smoother texture), and correlation (measuring the linear dependence of gray levels). For the canopy region of an abnormal single walnut tree, the canopy region image is first converted to a grayscale image, then its gray-level co-occurrence matrix (GLCM) is calculated, and the aforementioned feature values ​​are extracted from it to obtain the GLCM features at that monitoring time. Local binary pattern recognition (LBR) is an operator that describes the local texture patterns of an image. For each pixel, it uses the grayscale value of its center pixel as a threshold, compares the grayscale values ​​of surrounding pixels (usually a 3×3 window, i.e., 8 neighboring pixels) with the center value. Values ​​greater than or equal to the center value are marked as 1, and values ​​less than are marked as 0. These 8 binary numbers are then combined into a binary string in a clockwise or counterclockwise direction and converted to a decimal number. This value is the LBR code for that pixel. After calculating the LBR codes for all pixels in the entire canopy region image, the frequency of each code is statistically analyzed and normalized to form a LBR histogram, which serves as the LBR feature. Compared with gray-level co-occurrence matrix, local binary mode is more robust to changes in illumination and has a faster computation speed. It can effectively capture subtle structural changes on the canopy surface, such as the roughness of lesion edges and abrupt texture changes around leaf holes.

[0052] This study combines time series analysis of the abnormality index, canopy structure index, and canopy texture index of individual walnut trees to compare the change patterns of these indicators before and after the onset of the anomaly. Specifically, each time series is divided into two phases: a normal period before the onset of the anomaly and a development period after. The mean, variance, and trend slope of each indicator are calculated during the normal period, and the same statistics are calculated during the development period, focusing on whether there are abrupt changes near the onset of the anomaly. For example, the combined analysis examines whether the ratio of canopy porosity area to the total area of ​​the image sub-region significantly increases in a short period near the onset of the anomaly. If the porosity suddenly jumps from 5% in the normal period to 20%, it suggests the presence of pests causing rapid leaf drop. Similarly, the combined analysis examines whether the spatial concentration of canopy patches increases in a short period near the onset of the anomaly. If the patch concentration changes from a scattered state to a highly clustered state, it suggests the presence of locally spreading diseases. For canopy texture indices, co-analysis detects significant changes in the contrast, energy, homogeneity, and local binary pattern histogram of the gray-level co-occurrence matrix near the anomaly initiation time. For example, a sudden increase in contrast indicates that the canopy surface has become rough, possibly caused by insect-eaten holes; a sudden decrease in homogeneity indicates that the texture uniformity has been disrupted, possibly caused by scattered lesions. Co-analysis does not examine a single index in isolation, but rather integrates the consistency of the direction, magnitude, and timing of changes in multiple indices to form a multi-dimensional anomaly feature vector. Temporal changes include detecting whether the ratio of the area of ​​canopy pores to the area of ​​the image sub-region increases rapidly near the anomaly initiation time, whether the spatial concentration of canopy patch areas increases rapidly near the anomaly initiation time, and whether canopy texture indices change rapidly near the anomaly initiation time.

[0053] The preset damage mode determination logic is based on the multi-dimensional feature vector obtained from the above joint analysis, classifying each abnormal single walnut tree into either insect damage mode or disease damage mode. The determination logic can be designed as a set of rules. For example, if after the onset of the anomaly, the ratio of the area of ​​the canopy pore region to the area of ​​the image sub-region increases rapidly and continuously, while the spatial concentration of the canopy patch region does not change significantly or even decreases, and the texture features show a significant increase in contrast and an increase in the high-frequency components of the local binary pattern histogram, then it is determined to be an insect damage mode. This is because insects feeding on leaves will cause a large number of discrete perforations or notches, forming pores but not easily forming clustered patches, and the rough edges of the insect holes lead to an increase in texture contrast. Conversely, if after the onset of the anomaly, the spatial concentration of the canopy patch area significantly increases, the pore area ratio changes little or increases slowly, and the texture features show decreased homogeneity and specific peaks in the local binary pattern histogram (corresponding to larger, uniform patches), then it is determined to be a disease damage pattern. This is because pathogen infection often starts from local spots and gradually expands to form aggregated necrotic patches. The internal texture of the patches is relatively uniform, but the edges form clear boundaries with healthy tissue. Besides these rule-based judgment logics, machine learning methods can also be used. A batch of known pests and diseases can be pre-collected, the above indicators can be extracted as features, and a classifier (such as support vector machine, random forest, or lightweight gradient booster) can be trained to automatically classify new abnormal trees. After the judgment is completed, a damage pattern identifier is assigned to each abnormal walnut tree, such as "Pest" for pest damage and "Pathogen" for disease damage. This identifier will be recorded in the list of individual walnut trees for subsequent transmission chain analysis and risk assessment.

[0054] This application advances anomaly identification from simply determining "whether pests or diseases have occurred" to a deeper level of "what type of pests or diseases have occurred." Traditional remote sensing monitoring methods can only determine whether trees are healthy, but cannot distinguish the specific causes of unhealthiness. This application, however, introduces gray-level co-occurrence matrix and local binary pattern analysis methods, combined with structural indicators of canopy porosity and patches, to construct a multimodal feature system capable of distinguishing between pests and diseases. Pests typically cause discrete, randomly distributed small holes and notches, with a rapid increase in porosity and low patch concentration, resulting in a coarse texture. Diseases, on the other hand, often form continuous, clustered necrotic patches with high patch concentration and a relatively gradual increase in porosity, exhibiting a well-defined blocky pattern. This distinction is crucial for the selection of subsequent control measures—pests require insecticides, while diseases require fungicides. Confusion can lead to control failure. This not only enhances the diagnostic capabilities of monitoring methods but also provides direct technical support for differentiated control in precision agriculture.

[0055] In specific implementation, step S4 includes:

[0056] Step S41: Based on the spatial location information of each walnut tree in the list of individual walnut trees, for each individual walnut tree, determine C spatially adjacent individual walnut trees according to the preset spatial adjacency logic, establish the inter-tree adjacency relationship between each individual walnut tree, and form an inter-tree adjacency relationship network.

[0057] Step S42: For any two single walnut trees with inter-tree adjacency in the inter-tree adjacency relationship network, calculate the corresponding inter-tree propagation weight based on the spatial location information of the two single walnut trees and in combination with the preset propagation weight calculation rules, and assign the inter-tree propagation weight to each pair of single walnut trees with inter-tree adjacency in the inter-tree adjacency relationship network.

[0058] Step S43: Based on the anomalous start time and anomalous intensity index of each anomalous walnut tree in the set of anomalous individual walnut trees, under the constraint of inter-tree propagation weights, for any pair of anomalous individual walnut trees with inter-tree adjacency relationships... and abnormal single walnut trees When an abnormal single walnut tree The anomaly started later than the abnormal single walnut tree. When the abnormal start time is determined, the abnormal individual walnut tree is calculated based on the combination of the time difference of the abnormal start time and the inter-tree propagation weight. To an abnormal single walnut tree The spread score;

[0059] Step S44: For each abnormal walnut tree in the set of abnormal single walnut trees, select the abnormal single walnut tree whose propagation score conforms to the preset selection logic as the candidate propagation source single walnut tree. Associate the candidate propagation source single walnut tree with the abnormal single walnut tree through directed connection. Starting from the abnormal single walnut tree with the earliest abnormal start time, connect each abnormal single walnut tree in sequence along the direction of propagation score from large to small according to the directed connection relationship to form a propagation chain composed of abnormal single walnut trees.

[0060] Specifically, the pre-defined spatial adjacency logic is essentially the same as the logic for determining the spatial neighborhood set in step S24. However, the purpose of step S41 is to construct a tree adjacency relationship network covering the entire orchard, rather than simply calculating a local neighborhood for each tree. This logic typically employs the distance-based K-nearest neighbor method or the fixed-radius neighborhood method. For example, a distance threshold is set (generally 1.5 to 2 times the average crown width of walnut trees, approximately 3 to 6 meters). When the Euclidean distance between the centroids of the crowns of two walnut trees is less than this threshold, they are considered to have an adjacency relationship. Alternatively, for each walnut tree, the C nearest neighbors are selected, where C is typically 4 to 8. This logic is adopted because the spread of pests and diseases in the orchard exhibits spatial continuity. The probability of neighboring trees influencing each other through insect migration, root contact, rain splash, or agricultural operations is significantly higher than that of distant trees. Therefore, only by establishing a spatial adjacency relationship network can the topological structure of the spread be accurately depicted. The specific method for establishing an inter-tree adjacency network is as follows: Treat each walnut tree in the list of individual walnut trees as a node in the network. Then, for each tree, find its C adjacent trees according to a predefined spatial adjacency logic. Add an undirected edge between each pair of adjacent nodes. This way, all nodes and edges form an undirected graph, i.e., an inter-tree adjacency network. This network can be stored in the form of an adjacency matrix or an adjacency list, where the adjacency matrix... Line 1 A column element of 1 indicates a tree. Kazuki Adjacent, 0 indicates not adjacent.

[0061] Preset propagation weights quantify the ease with which pests and diseases spread from one tree to another, reflecting the impact of spatial distance, environmental barriers, or the distribution of propagation media on propagation efficiency. Rules for calculating inter-tree propagation weights are typically based on the spatial location information of the two trees. The most common rule defines the propagation weight as the reciprocal or negative exponential function of the Euclidean distance between the two trees; that is, the closer the distance, the greater the weight, because the probability of propagation between neighbors is higher. More refined rules can incorporate factors such as wind direction, terrain, and canopy size. For example, the propagation weight can be set equal to the distance attenuation factor multiplied by the wind direction influence coefficient. If the angle between the line connecting the two trees and the prevailing wind direction is less than 30 degrees, the weight increases; otherwise, it decreases. In practice, this can be simplified to propagation weight = 1 / (distance + constant), where the constant is to avoid infinity when the distance is zero. After calculating the propagation weights, they are assigned to each pair of adjacent nodes in the inter-tree adjacency network. The original undirected graph then becomes a weighted undirected graph, with each edge having a weight value representing the propagation intensity. The "constraint on inter-tree propagation weights" mentioned in step S43 refers to the requirement that the propagation score must be positively correlated with the propagation weights when calculating the propagation score. That is, the higher the propagation weight (the closer the two trees are or the more favorable the propagation conditions), the higher the propagation score from the source tree to the target tree should be. A situation where the propagation weight is very small but the propagation score is very large cannot occur. This is a fundamental constraint to ensure the physical rationality of the propagation chain. The propagation score is a combination function of the inter-tree propagation weights, the time difference function of the anomaly initiation time, and the anomaly intensity index function.

[0062] The combined function for calculating the propagation score requires combining the inter-tree propagation weights, the time difference between anomaly initiation times, and the anomaly intensity index. A typical combined function form is: Propagation Score = Propagation Weight × exp(-attenuation coefficient × time difference) × anomaly intensity index. This formula means that if the tree... The anomaly started later than the tree. So, the tree As a tree The possible propagation sources have a propagation score that is directly proportional to the propagation weight between the two trees (the closer the distance, the higher the score), and has a negative exponential relationship with the time difference (the smaller the time difference, the higher the score, because faster propagation is more reasonable), and is related to the tree... The severity of the anomaly is directly proportional to the source tree's anomaly (the more severe the anomaly, the stronger its ability to spread pests and diseases). If such a combination function is not used, and the propagation relationship is determined solely by time sequence or distance, two spatially adjacent trees with significant time differences may be incorrectly associated as a propagation chain, or the strong influence of severely anomalous trees on surrounding trees may be ignored. In this embodiment, when two anomalous single walnut trees are not directly adjacent in the inter-tree adjacency network, a candidate path between them can be determined in the inter-tree adjacency network. Under the condition that the path hop count does not exceed a preset maximum hop count threshold of 3 hops, the propagation weights between adjacent nodes on the path are cumulatively calculated to obtain an equivalent propagation weight. This equivalent propagation weight is then substituted into the propagation score combination function for calculation, thereby characterizing the indirect propagation process of pests and diseases through intermediate transition nodes.

[0063] The pre-defined selection logic is used to determine the most likely source of propagation for each anomalous walnut tree. A commonly used selection logic is: for a given anomalous tree... In all anomalies, the start time is earlier than that of its neighboring anomaly trees. In the process, the tree with the highest propagation score is selected as the candidate propagation source. If multiple candidates have similar scores but are significantly higher than others, multiple candidate sources can be retained. A more stringent approach is to only accept candidate sources with propagation scores exceeding a preset threshold; if no candidate exceeds the threshold, that tree is considered the initial propagation source (i.e., the first tree to show symptoms). The selection logic must consider the possibility of noise and mismatches. For example, long observation intervals or location errors may lead to inaccurate propagation score calculations; therefore, a minimum score threshold is usually required. Associating candidate propagation sources with anomalous trees through directed connections means starting from the candidate source tree... Point to target tree A directed edge represents the direction of pest and disease spread. Specifically, based on the tree adjacency network, nodes selected as source-target pairs are retained and assigned a direction. After completing the directed association of all abnormal trees, starting from the abnormal tree with the earliest anomaly initiation time (the tree without candidate sources), the path is traversed sequentially along the direction of the directed edge, connecting one or more directed paths. Each path is a transmission chain composed of an abnormal single walnut tree. These transmission chains not only demonstrate the spatial path of pest and disease spread in the orchard but also allow for the calculation of the transmission speed based on the anomaly initiation time of each node in the chain, providing a basis for precise control and blocking. By organically combining the results of single-tree anomaly identification with the spatial network model, a leap from "which trees are diseased" to "where the disease comes from and where it goes" has been achieved for the first time in walnut tree pest and disease monitoring. This provides direct spatial decision support for targeted pesticide application, isolation and removal, and transmission blocking in orchards—a function that traditional remote sensing monitoring methods cannot achieve.

[0064] In specific implementation, step S4 also includes:

[0065] Step S45: For each single walnut tree in the list of single walnut trees, map the abnormal intensity index of the single walnut tree to the initial risk value of the single walnut tree. For single walnut trees that do not belong to the set of abnormal single walnut trees, set their initial risk value of the single walnut tree to a preset non-abnormal risk value.

[0066] Discrete iterative calculations are performed on the inter-tree adjacency network. In each iteration, for each individual walnut tree, the risk value of the individual walnut tree in the current iteration step is allocated to the individual walnut trees with inter-tree adjacency according to the inter-tree propagation weight. The risk value of the individual walnut tree in the previous iteration step is combined with the risk value of the individual walnut tree received from the adjacent individual walnut trees in the current iteration step according to the preset combination logic. The risk value of the individual walnut tree is updated. The above discrete iterative calculations are repeated until the preset stopping condition is met, and the final risk value of each individual walnut tree is obtained.

[0067] Step S46: Associate the spatial location of each individual walnut tree with the final risk value of that individual walnut tree, and construct a risk distribution based on the spatial location information of the individual walnut trees, with the spatial location of the individual walnut trees as discrete points and the final risk value of the individual walnut trees as attribute values.

[0068] Specifically, the anomalous information and propagation relationships identified in the preceding steps are further transformed into quantifiable risk values, and risk diffusion simulation calculations are performed on a spatial scale across the entire orchard, ultimately generating a visualized risk distribution map. Mapping the anomaly intensity index to an initial risk value for a single walnut tree essentially converts the severity of the anomaly into a value suitable for iterative propagation calculations. The anomaly intensity index is typically a non-negative real number; for example, the average of the anomaly index sequence after the anomaly's inception might fall between 0 and 10. The mapping function can be linear, such as directly using the anomaly intensity index as the initial risk value, or compressing it to the interval between 0 and 1 using a sigmoid function. Alternatively, a piecewise function can be used, such as mapping an anomaly intensity index less than 2 to 0.2, between 2 and 5 to 0.5, and greater than 5 to 0.9. Different mapping methods will affect the sensitivity and convergence speed of subsequent iterative propagation, but the core idea is that trees with more severe anomalies have higher initial risk values. For trees that do not belong to the set of anomalous individual walnut trees, their initial risk value is set to a preset non-nominal risk value, which is usually 0 or a very small positive number such as 0.01. Setting it to 0 has the advantage that these trees themselves have no source of anomalousness, and their future risk value depends entirely on the propagation contribution received from anomalous neighbors. This clearly demonstrates the process of pests and diseases spreading from anomalous trees to the surrounding areas. Assigning a non-zero base risk value to non-nominal trees would introduce noise, blurring the starting point of the propagation chain and making it difficult to accurately locate the initial risk source.

[0069] Discrete iterative computation on an inter-tree adjacency network simulates the gradual diffusion of risk in space. Specifically, the inter-tree adjacency network is treated as a graph, where each node represents a walnut tree, and the weight of each edge is the inter-tree propagation weight calculated in step S42. At the start of each iteration, each node has an initial risk value (the mapped value for abnormal trees, and the preset non-abnormal risk value for non-abnormal trees). In each iteration, for each walnut tree, its current iteration risk value is first distributed to its neighboring nodes according to the propagation weights of the edges connected to that tree. For example, if tree A is adjacent to trees B and C, with propagation weights of 0.8 and 0.2 respectively, and tree A's current risk value is 10, then the risk increment it transmits to tree B is 10 × 0.8 / (0.8 + 0.2) = 8, and to tree C it is 2. Then, each tree collects the sum of risk increments transmitted from all its neighbors, merges it with its own risk value from the previous iteration according to a preset combination logic, and updates the tree's new risk value for the current iteration. The pre-defined combinational logic can be a simple addition, where the new risk value equals the risk value of the previous iteration plus the sum of risk increments received from neighbors. Alternatively, a decay factor can be introduced, such as new risk value = 0.9 × old value + received increments, to prevent the risk value from growing indefinitely. Repeating this process, the risk gradually spreads along the adjacency network from the anomaly tree outwards, simulating the spread of pests and diseases in a real orchard. Pre-defined stopping conditions are rules for terminating iterative calculations. Common stopping conditions include: the maximum rate of change of the risk value of all nodes between two consecutive iterations being less than a threshold (e.g., 0.001), indicating that the risk distribution has converged to a stable state, or the number of iterations reaching the preset maximum number (e.g., 100), preventing infinite loops due to non-convergence. This ensures that the iterative calculation yields a stable risk distribution while avoiding unnecessary waste of computational resources. Especially in large orchards with thousands of walnut trees, stopping too early may result in the risk not spreading far enough, while stopping too late incurs excessive computational overhead. A reasonable stopping condition strikes a balance between computational accuracy and efficiency.

[0070] After iterative convergence, the final risk value for each walnut tree is obtained. The next step is to correlate these risk values ​​with spatial locations to construct a risk distribution. Corresponding the spatial location of each walnut tree to the final risk value essentially involves adding a "Final Risk Value" column to the list of individual walnut trees. Each row records the centroid coordinates of the corresponding walnut tree and the final calculated risk value. Since the list of individual walnut trees already records the centroid coordinates of each tree at each monitoring time (from the spatial location matching in step S13), it is only necessary to extract the coordinates of any one monitoring time as the representative spatial location of that tree and then correlate them one-to-one with the final risk value. Based on these spatial locations and risk values, risk distribution is constructed using two common methods: one is discrete point mapping, which directly marks each walnut tree as a point on the orchard map, using the color or size of the point to indicate the level of risk, for example, high-risk trees are represented by large red dots, and low-risk trees by small green dots. This method is intuitive but cannot show the risk trend of continuous areas; the other is spatial interpolation, which uses Kriging interpolation or inverse distance weighted interpolation, taking the risk value of discrete points as input, to estimate the risk value of each location in the continuous space of the entire orchard, generating a risk contour map or heat map. This method can identify high-incidence areas of risk clustering and the gradient direction of risk diffusion. Regardless of the method used, the resulting risk distribution map can serve as a direct basis for orchard managers to carry out precise prevention and control: high-risk areas need to be prioritized for pesticide application or isolation, key nodes in the transmission chain can be cut off, and low-risk areas can be monitored routinely.

[0071] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0072] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the protection scope of the present invention.

Claims

1. A method for monitoring walnut tree diseases and pests based on UAV remote sensing data analysis, characterized in that, Includes the following steps: Step S1: Obtain multi-temporal UAV remote sensing image sequences, perform unified processing on the multi-temporal UAV remote sensing image sequences, and construct a list of individual walnut trees; Step S2: Extract the features of each monitoring time based on the list of individual walnut trees and construct the feature time series. Establish a health reference, calculate the time deviation index and spatial deviation index and combine them to obtain the time series of abnormal index of individual walnut trees, and determine the set of abnormal individual walnut trees. The formula for calculating the time deviation index is: ； in, A single walnut tree During monitoring time Features Time deviation indicator, A single walnut tree During monitoring time Features Feature values, Features During monitoring time The mean of the healthy reference growth curve, Features During monitoring time The standard deviation of the health reference dispersion curve; The formula for calculating the spatial deviation index is: ； in, A single walnut tree During monitoring time Features Spatial deviation index Features of individual walnut trees in a spatial neighborhood set During monitoring time The mean, Features of individual walnut trees in a spatial neighborhood set During monitoring time Standard deviation; Step S3: Extract the canopy structure indicators and canopy texture indicators at each monitoring time based on the set of abnormal single walnut trees, construct the corresponding time series, perform joint analysis, and determine the damage mode of each abnormal single walnut tree. The canopy structure indicators include the ratio of the area of ​​the canopy pore region to the area of ​​the image sub-region, the statistical value of the canopy patch region area, and the spatial concentration of the canopy patch region. The gray-level co-occurrence matrix features were calculated using the gray-level co-occurrence matrix, and the local binary pattern features were calculated using the local binary pattern. The gray-level co-occurrence matrix features and the local binary pattern features were used as the canopy texture indicators of the abnormal single walnut tree at the monitoring time. Step S4: Based on the list of individual walnut trees, establish an inter-tree adjacency relationship network, calculate the inter-tree propagation weight and construct the propagation chain, iteratively update the risk value of individual walnut trees, obtain the final risk value of individual walnut trees and form a risk distribution. Specifically, for each walnut tree in the list of individual walnut trees, the abnormal intensity index of that individual walnut tree is mapped to the initial risk value of the individual walnut tree. For individual walnut trees that do not belong to the set of abnormal individual walnut trees, their initial risk value of the individual walnut tree is set to a preset non-abnormal risk value. Discrete iterative calculations are performed on the inter-tree adjacency network. In each iteration, for each individual walnut tree, the risk value of the individual walnut tree in the current iteration step is distributed to the individual walnut trees with which it has an inter-tree adjacency relationship according to the inter-tree propagation weight. The risk value of the individual walnut tree in the previous iteration step is combined with the risk value of the individual walnut tree received from the adjacent individual walnut trees in the current iteration step according to a preset combination logic, and the risk value of the individual walnut tree is updated. The above discrete iterative calculations are repeated until the preset stopping condition is met, and the final risk value of each individual walnut tree is obtained.

2. The method for monitoring walnut tree diseases and pests based on UAV remote sensing data analysis as described in claim 1, characterized in that, Step S1 specifically includes: Step S11: Obtain a multi-temporal UAV remote sensing image sequence covering the walnut orchard area. Perform radiometric and geometric corrections on each UAV remote sensing image in the multi-temporal UAV remote sensing image sequence. Register each UAV remote sensing image to a unified coordinate reference system to obtain a standardized remote sensing image sequence. Step S12: Use altitude and spectral information to segment the UAV remote sensing images at each monitoring time in the standardized remote sensing image sequence to obtain the set of canopy regions of a single walnut tree at each monitoring time.

3. The method for monitoring walnut tree diseases and pests based on UAV remote sensing data analysis as described in claim 2, characterized in that, Step S1 also includes: Step S13: Based on the spatial location of the canopy area of ​​each walnut tree in the set of canopy areas of each walnut tree at each monitoring time, perform spatial location matching between different monitoring times, associate the canopy areas of the walnut trees corresponding to the spatial locations at different monitoring times as the same walnut tree, assign a unique walnut tree identifier to each walnut tree, and construct a list of walnut trees.

4. The method for monitoring walnut tree diseases and pests based on UAV remote sensing data analysis as described in claim 3, characterized in that, Step S2 specifically includes: Step S21: For each single walnut tree in the list of single walnut trees, at each monitoring time of the standardized remote sensing image sequence, extract vegetation index features and texture features from the canopy area of ​​the single walnut tree corresponding to that monitoring time. Arrange the vegetation index features and texture features of the same single walnut tree at all monitoring times in chronological order to obtain the feature time series of the single walnut tree. Step S22: In the list of individual walnut trees, select a portion of individual walnut trees that meet the preset health judgment logic as a set of healthy candidate individual walnut trees based on the value range and time stability of the characteristic time series of each individual walnut tree. For the characteristic time series of each walnut tree in the set of healthy candidate walnut trees, each characteristic is statistically analyzed at each monitoring time to obtain the mean and standard deviation of the characteristic at that monitoring time. The mean at each monitoring time is used to form the healthy reference growth curve of the characteristic, and the standard deviation at each monitoring time is used to form the healthy reference discrete curve of the characteristic. Step S23: For each walnut tree in the list of individual walnut trees, at each monitoring time and for each feature, based on the mean and standard deviation of the feature value of that individual walnut tree at that monitoring time and the corresponding healthy reference growth curve, calculate the time deviation index. The calculation formula is as follows: ； in, A single walnut tree During monitoring time Features Time deviation indicator, A single walnut tree During monitoring time Features Feature values, Features During monitoring time The mean of the healthy reference growth curve, Features During monitoring time The standard deviation of the health reference discrete curve.

5. The method for monitoring walnut tree diseases and pests based on UAV remote sensing data analysis as described in claim 4, characterized in that, Step S2 also includes: Step S24: At each monitoring time of the standardized remote sensing image sequence, based on the spatial location information of each walnut tree in the list of individual walnut trees, for each individual walnut tree, according to the preset spatial adjacency logic, determine the M individual walnut trees that are spatially adjacent to the individual walnut tree to form a spatial neighborhood set. For each individual walnut tree, at each monitoring time and for each feature, the spatial deviation index is calculated based on the feature value of that individual walnut tree at that monitoring time and the mean and standard deviation of that feature of all individual walnut trees in its spatial neighborhood set. The calculation formula is as follows: ； in, A single walnut tree During monitoring time Features Spatial deviation index Features of individual walnut trees in a spatial neighborhood set During monitoring time The mean, Features of individual walnut trees in a spatial neighborhood set During monitoring time Standard deviation; Step S25: For each single walnut tree in the list of single walnut trees, at each monitoring time, the time deviation index and spatial deviation index of the single walnut tree at that monitoring time are combined according to a preset combination logic to obtain the single walnut tree abnormality index at that monitoring time. The single walnut tree abnormality indexes at each monitoring time are arranged in chronological order to form a time series of single walnut tree abnormality indexes. Step S26: According to the preset anomaly judgment logic, analyze the time series of the anomaly index of a single walnut tree. For a single walnut tree that meets the preset anomaly judgment logic, determine it as an abnormal single walnut tree. The monitoring time when the preset anomaly judgment logic is first met is determined as the anomaly start time of the abnormal single walnut tree. The anomaly index of the single walnut tree after the anomaly start time in the time series of the single walnut tree anomaly index is calculated according to the preset aggregation logic to obtain the anomaly intensity index. All abnormal single walnut trees are formed into an abnormal single walnut tree set.

6. The method for monitoring walnut tree diseases and pests based on UAV remote sensing data analysis as described in claim 5, characterized in that, Step S3 specifically includes: Step S31: For each abnormal single walnut tree in the set of abnormal single walnut trees, obtain the single walnut tree canopy area corresponding to each monitoring time within the entire monitoring time range, and spatially divide the single walnut tree canopy area corresponding to each monitoring time to obtain a set of image sub-regions. Among them, the time of the abnormal start time is used as the time dividing point, and each monitoring time is divided into the monitoring time before the abnormal start time and the monitoring time after the abnormal start time. Step S32: For each abnormal single walnut tree, at each monitoring time, perform morphological operations on each image sub-region in the image sub-region set corresponding to the abnormal single walnut tree at that monitoring time, and separate the brightness structure in the image sub-region, and extract the canopy structure index that characterizes the canopy porosity feature and canopy patch feature. The canopy structure indicators include the ratio of the area of ​​the canopy pore region to the area of ​​the image sub-region, the statistical value of the canopy patch region area, and the spatial concentration of the canopy patch region. The ratio of the area of ​​the canopy pore region to the area of ​​the image sub-region, the statistical value of the area of ​​the canopy patch region, and the spatial concentration of the canopy patch region are used as the canopy structure indicators of the image sub-region at the monitoring time. The crown structure indices of all image sub-regions of the same abnormal single walnut tree at each monitoring time are summarized to obtain the time series of crown structure indices of the abnormal single walnut tree.

7. The method for monitoring walnut tree diseases and pests based on UAV remote sensing data analysis as described in claim 6, characterized in that, The processing logic for morphological operations based on image brightness information includes: The brightness of the image sub-regions is normalized to obtain a normalized brightness image; Perform opening and closing operations on the normalized brightness image to separate the bright and dark regions. The low-brightness areas are binarized and connected component analysis is performed to identify the connected components belonging to the canopy pore region. The area of ​​each connected component is calculated, and the areas of all canopy pore regions are summed to obtain the total area of ​​the canopy pore region. The highlighted areas are binarized and connected component analysis is performed to identify the connected components belonging to the canopy patch areas. The area and spatial location of each canopy patch area are statistically analyzed, and the area statistics and spatial concentration of the canopy patch areas are calculated respectively.

8. The method for monitoring walnut tree diseases and pests based on UAV remote sensing data analysis as described in claim 7, characterized in that, Step S3 also includes: Step S33: For each abnormal single walnut tree, at each monitoring time, the gray-level co-occurrence matrix features are calculated using the gray-level co-occurrence matrix and the local binary pattern features are calculated using the local binary pattern. The gray-level co-occurrence matrix features and the local binary pattern features are used as the crown texture index of the abnormal single walnut tree at that monitoring time. The crown texture indexes at each monitoring time are arranged in chronological order to obtain the crown texture index time series of the abnormal single walnut tree. Step S34: For each abnormal single walnut tree, the time series of the single walnut tree abnormality index, the time series of the crown structure index, and the time series of the crown texture index of the abnormal single walnut tree are jointly analyzed. Based on the time of the abnormality start time, the time series are compared in segments to obtain the changes in the single walnut tree abnormality index, the ratio of the area of ​​the crown pore region to the area of ​​the image sub-region, the spatial concentration of the crown patch region, and the crown texture index before and after the time of the abnormality start time. Step S35: Based on the preset damage mode determination logic, each abnormal single walnut tree is classified into a corresponding damage mode according to the abnormal index of a single walnut tree, the ratio of the area of ​​the canopy pore area to the area of ​​the image sub-region, the spatial concentration of the canopy patch area, and the change characteristics of the canopy texture index before and after the abnormality start time. Each abnormal single walnut tree is then assigned a damage mode identifier. The damage patterns include insect damage patterns and disease damage patterns.

9. The method for monitoring walnut tree diseases and pests based on UAV remote sensing data analysis as described in claim 8, characterized in that, Step S4 specifically includes: Step S41: Based on the spatial location information of each walnut tree in the list of individual walnut trees, for each individual walnut tree, determine C spatially adjacent individual walnut trees according to the preset spatial adjacency logic, establish the inter-tree adjacency relationship between each individual walnut tree, and form an inter-tree adjacency relationship network. Step S42: For any two single walnut trees with inter-tree adjacency in the inter-tree adjacency relationship network, calculate the corresponding inter-tree propagation weight based on the spatial location information of the two single walnut trees and in combination with the preset propagation weight calculation rules, and assign the inter-tree propagation weight to each pair of single walnut trees with inter-tree adjacency in the inter-tree adjacency relationship network. Step S43: Based on the anomalous start time and anomalous intensity index of each anomalous walnut tree in the set of anomalous individual walnut trees, under the constraint of inter-tree propagation weights, for any pair of anomalous individual walnut trees with inter-tree adjacency relationships... and abnormal single walnut trees When an abnormal single walnut tree The anomaly started later than the abnormal single walnut tree. When the abnormal start time is determined, the abnormal individual walnut tree is calculated based on the combination of the time difference of the abnormal start time and the inter-tree propagation weight. To an abnormal single walnut tree The spread score; Step S44: For each abnormal walnut tree in the set of abnormal single walnut trees, select the abnormal single walnut tree whose propagation score conforms to the preset selection logic as the candidate propagation source single walnut tree. Associate the candidate propagation source single walnut tree with the abnormal single walnut tree through directed connection. Starting from the abnormal single walnut tree with the earliest abnormal start time, connect each abnormal single walnut tree in sequence along the direction of propagation score from large to small according to the directed connection relationship to form a propagation chain composed of abnormal single walnut trees.

10. The method for monitoring walnut tree diseases and pests based on UAV remote sensing data analysis as described in claim 9, characterized in that, Step S4 also includes: Step S45: For each single walnut tree in the list of single walnut trees, map the abnormal intensity index of the single walnut tree to the initial risk value of the single walnut tree. For single walnut trees that do not belong to the set of abnormal single walnut trees, set their initial risk value of the single walnut tree to a preset non-abnormal risk value. Discrete iterative calculations are performed on the inter-tree adjacency network. In each iteration, for each individual walnut tree, the risk value of the individual walnut tree in the current iteration step is allocated to the individual walnut trees with inter-tree adjacency according to the inter-tree propagation weight. The risk value of the individual walnut tree in the previous iteration step is combined with the risk value of the individual walnut tree received from the adjacent individual walnut trees in the current iteration step according to the preset combination logic. The risk value of the individual walnut tree is updated. The above discrete iterative calculations are repeated until the preset stopping condition is met, and the final risk value of each individual walnut tree is obtained. Step S46: Associate the spatial location of each individual walnut tree with the final risk value of that individual walnut tree, and construct a risk distribution based on the spatial location information of the individual walnut trees, with the spatial location of the individual walnut trees as discrete points and the final risk value of the individual walnut trees as attribute values.

Images