Karst hillside orchard disease and insect pest spatial distribution generation method based on unmanned aerial vehicle spectral image

Abstract

The invention relates to the technical field of agricultural remote sensing plant protection, and discloses a karst hillside orchard pest and disease spatial distribution generation method based on unmanned aerial vehicle spectral images. According to the method, periodic unmanned aerial vehicle multispectral images are collected, and a standardized canopy reflectivity field is generated through illumination difference compensation and terrain shadow correction. A multi-time-sequence spectral response curve is constructed based on a reflectivity field, and a spectral response abnormal mode library is established in combination with distortion characteristics of historical disease and pest samples. And integrating the data to construct a spatial probabilistic reasoning network, carrying out pixel-by-pixel spectral state deduction on a real-time image, calculating a matching degree between each point location and an abnormal mode as a disease and pest occurrence membership degree, and finally generating a probabilistic spatial distribution thermodynamic diagram. According to the method, mountain terrain shadow interference is overcome, time sequence spectrum distortion characteristics of diseases and insect pests are utilized, and the accuracy and early warning capability of disease and insect pest recognition in a complex environment are improved.

Description

Technical Field

[0001] This invention relates to the field of agricultural remote sensing plant protection technology, specifically a method for generating spatial distribution of pests and diseases in karst mountain orchards from UAV spectral images. Background Technology

[0002] In precision agriculture, using UAV multispectral remote sensing technology to monitor orchard pests and diseases has become an important method. Existing technologies typically extract and classify spectral features from acquired canopy images to determine the areas affected by pests and diseases. However, this approach suffers from severe interference with the underlying canopy reflectance information when dealing with the unique scenario of karst mountain orchards. The non-uniform lighting and dense shadows caused by complex terrain significantly alter the spectral signals received by the sensors, resulting in vastly different reflectance values ​​for the same healthy canopy under different slope aspects or shadow conditions. This noise introduced by the terrain directly masks the true spectral characteristics of the crop itself, leading to a significant decrease in the reliability of traditional pest and disease identification methods based on reflectance thresholds or indices in karst mountain orchards.

[0003] Current remote sensing monitoring schemes for pests and diseases mostly rely on the analysis of image data from single or limited time phases. The occurrence and development of pests and diseases is a dynamic physiological process; the spectral anomalies they exhibit in single-time-phase images during their early or latent stages may be extremely weak and easily confused with spectral changes caused by other factors such as water stress and nutrient deficiency. Static single-time-phase analysis struggles to capture the unique temporal evolution patterns of pests and diseases and lacks consideration for dynamic baseline changes in the spectral spectrum during crop growth. Therefore, its ability to detect pests and diseases early is limited, and the false alarm rate is high. How to extract indicative pest and disease-specific patterns from temporal spectral changes and overcome topographic interference for accurate spatial positioning is an urgent problem to be solved. Summary of the Invention

[0004] The purpose of this invention is to provide a method for generating spatial distribution of pests and diseases in karst mountain orchards based on UAV spectral imagery, in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, this invention provides a method for generating spatial distribution data of pests and diseases in karst mountain orchards from UAV spectral imagery, the method comprising:

[0006] Collect periodic UAV multispectral images covering the entire target karst mountain orchard area;

[0007] Illumination difference compensation and terrain shadow correction are performed on periodic UAV multispectral images to generate a standardized orchard canopy reflectivity field;

[0008] Based on the standardized orchard canopy reflectance field analysis, multiple preset waveband reflectance combinations are analyzed to construct multi-time series spectral response curves;

[0009] Based on the distortion characteristics of historical pest and disease samples on multi-time-series spectral response curves, a library of abnormal spectral response patterns was identified.

[0010] A spatial probabilistic inference network is constructed by integrating standardized orchard canopy reflectivity fields, multi-time-series spectral response curves, and a library of spectral response anomalies.

[0011] The spatial probabilistic reasoning network is used to perform pixel-by-pixel spectral state inference on real-time orchard drone spectral images.

[0012] Based on the matching degree between the inference results and the spectral response anomaly pattern library, the membership degree of pest and disease occurrence at each spatial point is calculated.

[0013] Based on the spatial continuity of the membership degree of pest occurrence at various points in space, a probabilistic heat map of the spatial distribution of pests is generated.

[0014] Preferably, illumination difference compensation and terrain shading correction are performed on periodic UAV multispectral imagery to generate a standardized orchard canopy reflectance field, specifically including:

[0015] Read the raw radiance values ​​of periodic UAV multispectral images and the corresponding solar altitude angle and azimuth angle data at the time of imaging;

[0016] The topographic slope and aspect of each pixel in the karst mountain orchard were calculated using digital elevation model data.

[0017] By combining solar altitude angle, azimuth angle data, and terrain slope and aspect data, the terrain shadow coefficient is calculated and the original radiance value is compensated.

[0018] The radiance value after terrain shading compensation is converted into surface reflectance to generate a standardized orchard canopy reflectance field that eliminates the influence of terrain and light differences.

[0019] Preferably, based on the standardized orchard canopy reflectance field analysis, multiple preset waveband reflectance combinations are used to construct multi-time-series spectral response curves, specifically including:

[0020] The reflectance values ​​of the preset blue light band, green light band, red light band, red edge band and near-infrared band are extracted from the standardized orchard canopy reflectance field.

[0021] Calculate the vegetation index, which is the difference between the reflectance of the red band and the near-infrared band, and the normalized red edge index, which is the reflectance of the red edge band and the near-infrared band.

[0022] By connecting the difference vegetation index values ​​and normalized red edge index values ​​of each spatial pixel at multiple monitoring time points in chronological order, a multi-time-series spectral response curve representing the vegetation growth changes of the spatial pixel is formed.

[0023] Preferably, based on the distortion characteristics of historical pest and disease samples on multi-time-series spectral response curves, a library of abnormal spectral response patterns is identified, specifically including:

[0024] Obtain the spatiotemporal coordinates of historically verified pest and disease occurrence areas;

[0025] Locate the curve segment from the multi-time-series spectral response curves that corresponds to the spatiotemporal coordinates of historical pest and disease occurrences;

[0026] The distortion characteristics of the curve segment during the occurrence of pests and diseases, which are different from those of the healthy sample curve in terms of shape, slope and fluctuation amplitude, are analyzed, including the characteristics of earlier inflection point, attenuation of peak and rise of trough.

[0027] The statistically significant distortion features are categorized into different types, and each type is defined as a standard spectral response anomaly pattern. All standard spectral response anomaly patterns constitute a spectral response anomaly pattern library.

[0028] Preferably, a spatial probabilistic inference network is constructed by integrating standardized orchard canopy reflectance fields, multi-time-series spectral response curves, and a library of spectral response anomalies, specifically including:

[0029] The standardized orchard canopy reflectance field is used as the bottom spectral input layer of the spatial probabilistic inference network.

[0030] Multi-time-series spectral response curves are used as the temporal dynamic feature layer of the spatial probabilistic inference network;

[0031] The spectral response anomaly pattern library is used as the pattern matching rule layer of the spatial probabilistic inference network;

[0032] Within the spatial probabilistic reasoning network, a weighted connection and probability propagation path are established from the spectral input layer, through the temporal dynamic feature layer, and then to the pattern matching rule layer, enabling the network to activate corresponding abnormal patterns based on the input spectrum and dynamic features.

[0033] Preferably, a pixel-by-pixel spectral state deduction is performed on the real-time acquired orchard UAV spectral imagery using a spatial probabilistic inference network, specifically including:

[0034] Input real-time single-phase orchard UAV spectral imagery, and after preprocessing, obtain real-time standardized canopy reflectance data;

[0035] Real-time difference vegetation index and real-time normalized red edge index are calculated from real-time standardized canopy reflectance data;

[0036] The real-time difference vegetation index and the real-time normalized red edge index are connected to the end of the spatial pixel historical multi-time series spectral response curve to form an extended temporary time series curve segment.

[0037] Input this temporary time-series curve segment into the temporal dynamic feature layer of the spatial probabilistic inference network to trigger state inference along the probability propagation path within the network.

[0038] Preferably, based on the matching degree between the inference results and the spectral response anomaly pattern library, the membership degree of pest and disease occurrence at each spatial location is calculated, specifically including:

[0039] The pattern matching rule layer of the spatial probabilistic inference network measures the similarity between the inferred spectral dynamic features and each standard spectral response anomaly pattern in the spectral response anomaly pattern library.

[0040] Calculate the matching probability value between the derived spectral dynamic characteristics and each standard spectral response anomalous mode;

[0041] The matching probability values ​​corresponding to all standard spectral response anomaly modes are weighted and fused, and the fused probability value is the pest and disease occurrence membership degree of the spatial pixel point.

[0042] Preferably, based on the spatial continuity of the membership degree of pest occurrence at various points in space, a probabilistic heat map of the spatial distribution of pests is generated, specifically including:

[0043] The calculated membership degree of pest and disease occurrence at each spatial point is assigned back to the corresponding geographic spatial coordinates to form a discrete probability matrix of pest and disease occurrence.

[0044] Spatial interpolation is performed on discrete probability lattices of pests and diseases to generate a continuous probability surface covering the entire orchard area.

[0045] Based on the probability value range of the continuous probability surface, define color rendering schemes corresponding to different probability levels from low to high;

[0046] The continuous probability surface is visualized and rendered according to the color rendering scheme, and the output is a heat map of the spatial distribution of probabilistic pests and diseases.

[0047] Preferably, the curve segment corresponding to the spatiotemporal coordinates of historical pest and disease occurrences is located from the multi-time-series spectral response curves, specifically as follows:

[0048] Based on the specific dates and orchard plot locations recorded in historical records of pest and disease occurrences, determine their index numbers and corresponding spatial pixel coordinates in the time series;

[0049] Based on the time index number, a curve segment centered on the recorded date and containing a preset number of days is extracted from the multi-time-series spectral response curve corresponding to the spatial pixel coordinates. This curve segment is the target curve segment.

[0050] Ensure that the captured target curve segment completely includes the spectral changes of the pest or disease from its initial occurrence to the time of recording.

[0051] Preferably, spatial interpolation is performed on the discrete probability lattice of pest and disease occurrence to generate a continuous probability surface covering the entire orchard area, specifically:

[0052] Kriging space interpolation is used to process discrete probability lattices of pest and disease occurrence.

[0053] Based on the orchard digital elevation model, altitude and slope are used as auxiliary variables in the Kriging interpolation calculation to reflect the possible impact of topography on the spread of pests and diseases.

[0054] The predicted probability value of each unsampled location within the orchard area is calculated by Kriging space interpolation, ultimately forming a smooth and continuous surface for the probability of pest and disease occurrence.

[0055] Compared with the prior art, the beneficial effects of the present invention are:

[0056] By performing a specialized topographic shading correction step, a topographically independent standardized canopy reflectance field was generated. This eliminated the systematic distortion of spectral data caused by uneven illumination and shading effects resulting from differences in slope gradient and aspect. The corrected reflectance data more accurately reflects the biochemical and physiological properties of the canopy itself, rather than illumination artifacts caused by topographic features. This provides a reliable and consistent data foundation for all subsequent spectral analyses based on absolute or relative reflectance values, making spectral values ​​of the same type of healthy or diseased canopy collected from sunny and shady slopes, and from shaded and sunny areas comparable. This reduces misjudgments of pests and diseases caused by topographic factors in mountainous environments and improves the generalization ability and spatial distribution accuracy of monitoring models across different topographic units.

[0057] This technology constructs multi-temporal spectral response curves based on periodic images and builds a spectral response anomaly pattern library based on the distortion characteristics of historical samples. Then, it utilizes a spatial probabilistic inference network for state deduction and matching. This transforms pest and disease identification from static, single-point spectral feature matching to dynamic, temporal pathological pattern matching. It not only focuses on the spectral state at the current moment but also on the distortion trajectory of the pixel's spectrum relative to its own historical baseline or healthy reference over time. This temporal pattern analysis can keenly capture gradual spectral changes that may not be significant in a single time phase caused by the early stages of pest and disease occurrence. By calculating the matching degree with known anomaly patterns through the probabilistic inference network, it can distinguish between temporal distortions caused by pests and diseases and spectral fluctuations caused by other stress factors, thereby achieving earlier and more specific identification of pests and diseases and outputting a spatial distribution of pest and disease occurrence risk with probabilistic significance. Attached Figure Description

[0058] Figure 1 This is a schematic diagram illustrating the working principle of the method for generating spatial distribution of pests and diseases in karst mountain orchards from UAV spectral images described in this invention.

[0059] Figure 2 A flowchart for generating a standardized orchard canopy reflectivity field;

[0060] Figure 3 A flowchart for identifying a library of anomalous spectral response patterns;

[0061] Figure 4 Heat map showing the spatial distribution of the membership degree of pest and disease occurrence in karst mountain orchards;

[0062] Figure 5 A heat map showing the spatial distribution of the probability of pests and diseases occurring in orchards in karst mountainous areas. Detailed Implementation

[0063] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0064] Please see Figure 1This invention provides a method for generating spatial distribution of pests and diseases in karst mountain orchards based on UAV spectral imagery. The method includes: acquiring periodic UAV multispectral images covering the entire target karst mountain orchard; performing illumination difference compensation and terrain shadow correction on the periodic UAV multispectral images to generate a standardized orchard canopy reflectance field; analyzing the reflectance combinations of multiple preset bands based on the standardized orchard canopy reflectance field to construct multi-time-series spectral response curves; identifying a spectral response anomaly pattern library based on the distortion characteristics of historical pest and disease samples on the multi-time-series spectral response curves; integrating the standardized orchard canopy reflectance field, multi-time-series spectral response curves, and spectral response anomaly pattern library to construct a spatial probabilistic inference network; performing pixel-by-pixel spectral state inference on the real-time acquired orchard UAV spectral images through the spatial probabilistic inference network; calculating the pest and disease occurrence membership degree at each spatial location based on the matching degree of the inference results and the spectral response anomaly pattern library; and generating a probabilistic pest and disease spatial distribution heatmap based on the spatial continuity of the pest and disease occurrence membership degree at each spatial location.

[0065] In one embodiment of the present invention, see [reference] Figure 2 The system reads the original radiance values ​​and corresponding solar altitude and azimuth data from periodic UAV multispectral images. It uses digital elevation model data to calculate the terrain slope and aspect of each pixel in the karst mountain orchard. Combining the solar altitude and azimuth data with the terrain slope and aspect data, it calculates the terrain shadow coefficient and compensates for the original radiance values. The radiance values ​​after terrain shadow compensation are converted into surface reflectance to generate a standardized orchard canopy reflectance field that eliminates the influence of terrain and light differences. Based on the standardized orchard canopy reflectance field, multiple preset bands of reflectance combinations are analyzed to construct multi-time-series spectral response curves. Specifically, this process involves: extracting the reflectance values ​​of preset blue, green, red, red-edge, and near-infrared bands from the standardized orchard canopy reflectance field; calculating the difference vegetation index between the red and near-infrared bands, as well as the normalized red-edge index between the red and near-infrared bands; and connecting the difference vegetation index values ​​and normalized red-edge index values ​​of each spatial pixel at multiple monitoring time points in chronological order to form a multi-time-series spectral response curve characterizing the vegetation growth changes of the spatial pixel.

[0066] In practice, illumination difference compensation and terrain shadow correction are performed on periodic UAV multispectral images to generate a standardized orchard canopy reflectance field. Based on the standardized orchard canopy reflectance field, multiple preset band reflectance combinations are analyzed to construct multi-time-series spectral response curves. The following description uses a karst mountain orchard example scenario and data comparison. The karst mountain orchard in the example scenario is located at 30 degrees north latitude and 110 degrees east longitude, planted with citrus trees, covering an area of ​​approximately 50 hectares. The UAV multispectral image acquisition uses a UAV platform equipped with blue light band, green light band, red light band, red edge band, and near-infrared band sensors. The flight altitude is set at 100 meters, the ground resolution is 0.1 meters, and the image acquisition cycle is fixed at once a week, covering the entire growing season of the fruit trees. The acquired raw images include radiance values ​​and corresponding imaging timestamps.

[0067] In some embodiments, illumination difference compensation and terrain shadow correction are performed on periodic UAV multispectral images to generate a standardized orchard canopy reflectivity field. The original radiance value of the periodic UAV multispectral image and the corresponding solar altitude angle and azimuth angle data at the time of imaging are read. The solar altitude angle and azimuth angle data are obtained by calculating based on the solar position model using the imaging timestamp and the orchard's geographical location. The terrain slope and aspect of each pixel in the karst mountain orchard are calculated using digital elevation model data. The digital elevation model data is obtained from airborne lidar scanning, and its spatial resolution is consistent with the multispectral image. Combining the solar altitude angle, azimuth angle data, and terrain slope and aspect data, the terrain shadow coefficient is calculated and the original radiance value is compensated. The terrain shadow coefficient is calculated using an illumination geometry model, and the formula is expressed as:

[0068]

[0069] in: Indicates the terrain shading coefficient. Indicates the solar altitude angle. Indicates the slope of the terrain. Indicates the azimuth angle of the sun. The radiance value after terrain shading compensation is converted into surface reflectance to represent the terrain slope aspect. This generates a standardized orchard canopy reflectance field that eliminates the influence of terrain and light difference. The conversion from radiance value to surface reflectance adopts an atmospheric correction method based on the atmospheric radiative transfer model, and atmospheric optical thickness parameter data are introduced simultaneously.

[0070] Understandably, in the example scenario, data comparison includes the numerical difference analysis of the canopy reflectance fields before and after correction at the same pixel location. In specific implementation, based on the standardized orchard canopy reflectance field, multiple preset band reflectance combinations are analyzed to construct multi-time-series spectral response curves. Preset reflectance values ​​for blue, green, red, red-edge, and near-infrared bands are extracted from the standardized orchard canopy reflectance field. These reflectance values ​​are stored in floating-point array format. The vegetation index, calculated as the difference between the red and near-infrared reflectance, and the difference between the red-edge and near-infrared reflectance, are also calculated. The normalized red-edge index of outer band reflectance and the difference vegetation index are obtained by subtracting the near-infrared band reflectance value from the red band reflectance value. The normalized red-edge index is obtained by subtracting the near-infrared band reflectance value from the red-edge band reflectance value and then dividing by the sum of the red-edge band reflectance value and the near-infrared band reflectance value. The difference vegetation index value and the normalized red-edge index value of each spatial pixel at multiple monitoring time points are connected in chronological order to form a multi-time series spectral response curve characterizing the vegetation growth change of the spatial pixel. The time sequence is in weekly intervals, and the curve data is organized and stored in the form of a time series array.

[0071] Optionally, in data comparison, the construction of multi-time series spectral response curves involves comparing the numerical sequences of vegetation indexes for the same pixel at different time points. In some embodiments, the construction process of multi-time series spectral response curves incorporates time series smoothing to reduce random noise. It can be understood that the standardized orchard canopy reflectance field generated by the illumination difference compensation and terrain shadow correction steps provides normalized data input for the construction of multi-time series spectral response curves.

[0072] In one embodiment of the present invention, see [reference] Figure 3 The study obtains the spatiotemporal coordinates of historically verified pest and disease occurrence areas, locates the curve segments corresponding to the historical spatiotemporal coordinates of pest and disease occurrence from multi-time series spectral response curves, and analyzes the distortion characteristics of the curve segments during pest and disease occurrence that differ from healthy sample curves in terms of shape, slope, and fluctuation amplitude, including the characteristics of earlier curve inflection points, peak attenuation, and trough elevation. The statistically significant distortion characteristics are summarized into different types, and each type is defined as a standard spectral response anomaly pattern. All standard spectral response anomaly patterns constitute a spectral response anomaly pattern library. The specific process for locating the curve segment corresponding to the spatiotemporal coordinates of historical pest and disease occurrences from multi-time series spectral response curves is as follows: Based on the specific date and orchard location recorded in the historical pest and disease occurrence records, determine its index number and corresponding spatial pixel coordinates in the time series. According to the time index number, extract a curve segment centered on the recorded date and including a preset number of days before and after from the multi-time series spectral response curves corresponding to the spatial pixel coordinates. This curve segment is the target curve segment, ensuring that the extracted target curve segment completely includes the spectral change process of the pest and disease from the initial occurrence to the recorded time.

[0073] In practical implementation, the spatial distribution generation method of pests and diseases in karst mountain orchards based on UAV spectral imagery identifies a library of abnormal spectral response patterns based on the distortion characteristics of historical pest and disease samples on multi-time series spectral response curves. The following description is based on a karst mountain orchard example scenario and data comparison. The karst mountain orchard in the example scenario has undergone two years of periodic UAV multispectral monitoring, accumulating a total of 80 valid monitoring dates and establishing a historical database containing multi-time series spectral response curves. The orchard plant protection records record 15 areas of citrus Huanglongbing (HLB) occurrence that have been verified in the field, and each area records the specific date of onset and the central geographic coordinates.

[0074] In some embodiments, based on the distortion characteristics of historical pest and disease samples on multi-time-series spectral response curves, a library of abnormal spectral response patterns is identified. The spatiotemporal coordinates of historically verified pest and disease occurrence areas are obtained. These coordinates, derived from the orchard plant protection record system, include the disease name, discovery date, and the latitude and longitude coordinates of the disease patch center. The curve segment corresponding to the historical pest and disease occurrence spatiotemporal coordinates is located from the multi-time-series spectral response curves. Based on the specific date and orchard location recorded in the historical pest and disease occurrence records, its index number and corresponding time series data are determined. The spatial pixel coordinates and time index numbers are obtained by comparing the discovery date with a list of 80 monitoring dates. The spatial pixel coordinates are obtained by registering the latitude and longitude coordinates of the disease patch center with the geographic reference information of the UAV image. Based on the time index number, a curve segment centered on the recorded date and including a preset number of days is extracted from the multi-time series spectral response curves corresponding to the spatial pixel coordinates. The preset number of days is set to 35 days. This curve segment is the target curve segment, ensuring that the extracted target curve segment completely includes the spectral change process of the pest from the initial occurrence to the time of recording.

[0075] Understandably, in the example scenario, data comparison involves extracting curve segments with the same time span from the multi-time series spectral response curves of healthy years at the same pixel coordinates, and performing morphological comparison with the target curve segment of the year in which pests and diseases occur. In specific implementation, the distortion characteristics of the curve segment during the pest and disease occurrence period that differ from the healthy sample curve in terms of shape, slope, and fluctuation amplitude are analyzed, including the characteristics of earlier curve inflection points, peak attenuation, and trough rise. The analysis of the earlier curve inflection point characteristic is achieved by calculating the time offset of the occurrence of the extreme point of the first derivative of the curve. The analysis of the peak attenuation characteristic is achieved by calculating the percentage decrease of the local maximum value of the curve relative to the healthy sample curve. The analysis of the trough rise characteristic is achieved by calculating the percentage increase of the local minimum value of the curve relative to the healthy sample curve. The distortion characteristics with statistical significance are summarized into different types, and the statistical significance is determined by hypothesis testing. The significance level threshold is set at 0.05, and each type is defined as a standard spectral response anomaly pattern. All standard spectral response anomaly patterns constitute a spectral response anomaly pattern library.

[0076] Optionally, the construction of the spectral response anomaly pattern library introduces pattern similarity measurement. The process of summarizing curve segment distortion features adopts dynamic time warping algorithm to calculate the overall morphological distance between different disease and pest curve segments. Dynamic time warping algorithm is implemented by nonlinearly aligning two time series curves with potentially different lengths and calculating the minimum cumulative distance. In its implementation, the calculation process uses a multi-time-series spectral response curve of a healthy fruit tree and a multi-time-series spectral response curve of a fruit tree affected by pests and diseases as inputs. Each curve consists of a series of vegetation index values ​​arranged in chronological order. The healthy curve may contain stable periodic fluctuations, while the pest and disease curve may exhibit distortions such as peak decay or trough rise. The dynamic time warping algorithm first constructs a distance matrix, with the horizontal and vertical axes corresponding to each time point on the two curves. The value of each element in the matrix is ​​the absolute difference between the vegetation index values ​​at the corresponding time points on the two curves. This absolute difference is called the local distance. The algorithm then starts from the starting point of the distance matrix and searches for an optimal curved path that runs through the matrix to the endpoint. This path defines how to non-linearly stretch or compress and align the two curves on the time axis so that the most similar points on the two curves can match each other.

[0077] Finding the optimal curved path follows three constraints: the path must start from the top left corner of the distance matrix and end at the bottom right corner, ensuring that the beginning and end points of the sequence are considered; the path is continuous and monotonic, meaning each step on the path can only move right, down, or to the lower right, guaranteeing that the time order is not reversed; and the path typically has global constraints, with bandwidth limiting the maximum deviation of the path from the diagonal to improve computational efficiency and avoid unrealistic over-distortion. Along this optimal curved path, the local distances of all grid points traversed by the path are accumulated, and the total accumulated distance is the overall morphological distance between the two curve segments. The smaller this distance value, the more similar the overall shapes of the two curves; the larger the distance value, the greater the morphological difference.

[0078] It is understandable that the core advantage of the dynamic time warping algorithm lies in its ability to handle the scaling and phase differences of time series along the time axis. For example, pests and diseases may cause changes in spectral characteristics to occur earlier or later. By finding the optimal alignment, the algorithm can measure the essential morphological differences after eliminating the influence of simple time shifts. In some embodiments, the input curve is normalized before calculating the overall morphological distance to eliminate the influence of differences in basic reflectance between different pixels on the distance measurement. In data comparison, the difference in vegetation index numerical sequences between healthy sample curves and pest and disease sample curves within the same time window is calculated using the following formula to determine the degree of morphological difference:

[0079]

[0080] in: Indicates within the time window Average morphological differences within This indicates the health sample curve at time point. The vegetation index value, This indicates the sample curve of pests and diseases at a given time point. The vegetation index value.

[0081] In one embodiment of the present invention, a standardized orchard canopy reflectance field is used as the bottom spectral input layer of the spatial probabilistic inference network, multi-temporal spectral response curves are used as the temporal dynamic feature layer of the spatial probabilistic inference network, and a spectral response anomaly pattern library is used as the pattern matching rule layer of the spatial probabilistic inference network. Within the spatial probabilistic inference network, a weighted connection and probability propagation path are established from the spectral input layer, through the temporal dynamic feature layer, and then to the pattern matching rule layer, enabling the network to activate corresponding anomaly patterns based on the input spectrum and dynamic features. Specifically, the state of the spectral input layer nodes is assumed to be a multi-dimensional vector. The state of nodes in the temporal dynamic feature layer is a vector. In the pattern matching rule layer, the first The activation probability of each abnormal pattern discrimination node is: Its mathematical model is defined by the following formula:

[0082]

[0083] in, Indicates that at a given spectral input and temporal dynamic characteristics Under the condition of [condition], the first [condition] in the network activation pattern matching rule layer The posterior probability of a node that identifies an abnormal pattern. Indicates from the spectral input layer To the temporal dynamic feature layer The connection weight matrix is ​​used to map the original spectral data into a temporal feature representation. Indicates the dynamic feature layer of time series In the pattern matching rule layer The connection weight vector of each discriminant node encodes the strength of the association between the specific anomaly pattern and the dynamic feature. This represents a non-linear activation function, such as Sigmoid, which performs a non-linear transformation on the weighted sum, thereby introducing expressive power into the model. The formula represents the total number of anomalous modes in the standard spectral response of the pattern matching rule layer. The denominator sums the weighted outputs of all modes to normalize the probability, ensuring that the sum of the activation probabilities of all modes is 1. This mathematical model clarifies the probability propagation path in the network: First, the input spectrum... Through the weight matrix Perform a linear transformation and pass through an activation function Processing yields the state representation of the temporal dynamic feature layer. Subsequently, this state indicates Through the weight vector corresponding to each mode The dot product operation is performed to obtain the unnormalized matching degree; finally, the normalization calculation in the form of the Softmax function is used to obtain the specific activation probability of each abnormal pattern, thus completing the reasoning process from input data to probabilistic decision.

[0084] In practical implementation, the method for generating spatial distribution of pests and diseases in karst mountain orchards using UAV spectral imagery integrates standardized orchard canopy reflectance fields, multi-temporal spectral response curves, and a library of spectral response anomalies to construct a spatial probabilistic inference network. The following description uses an example scenario of a karst mountain orchard and data comparison. The karst mountain orchard in the example scenario already has processed standardized historical data sequences of orchard canopy reflectance fields, a library of multi-temporal spectral response curves built based on historical data, and a library of spectral response anomalies identified from historical pest and disease samples. The standardized orchard canopy reflectance field contains data in 5 spectral bands, the multi-temporal spectral response curves contain 80 time point sequences in two dimensions: differential vegetation index and normalized red edge index, and the spectral response anomaly library contains 5 summarized standard spectral response anomaly patterns.

[0085] In some embodiments, a spatial probabilistic inference network is constructed by integrating a standardized orchard canopy reflectance field, multi-time-series spectral response curves, and a library of spectral response anomalies. The standardized orchard canopy reflectance field serves as the bottom spectral input layer of the spatial probabilistic inference network. The spectral input layer is a multi-dimensional data node, with each dimension corresponding to a reflectance value of a preset spectral band. The multi-time-series spectral response curves serve as the temporal dynamic feature layer of the spatial probabilistic inference network. The temporal dynamic feature layer consists of multiple cascaded temporal memory units, each corresponding to a pair of difference vegetation index and normalized red edge index values ​​at a specific time point. The library of spectral response anomalies serves as the pattern matching rule layer of the spatial probabilistic inference network. The pattern matching rule layer contains a set of discriminant nodes, each corresponding to a standard spectral response anomaly pattern in the library. Within the spatial probabilistic inference network, a weighted connection and probability propagation path are established from the spectral input layer, through the temporal dynamic feature layer, and then to the pattern matching rule layer. The weighted connection is defined in the form of a connection weight matrix, and the probability propagation path is calculated forward based on conditional probability rules.

[0086] Understandably, in the example scenario, the data comparison includes the difference analysis between the structure of the three independent data sources before network construction and the integrated inference structure after network construction. In specific implementation, the process of establishing weighted connections and probability propagation paths defines state variables for each layer node in the network and assigns initial values ​​to the connection weights. The state variable of the spectral input layer is a 5-dimensional reflectance vector, the state variable of the temporal dynamic feature layer is a 160-dimensional temporal feature vector, and the state variable of the pattern matching rule layer is a 5-dimensional pattern matching probability vector. The dimension of the connection weight matrix from the spectral input layer to the temporal dynamic feature layer is 5 times 160, and the dimension of the connection weight matrix from the temporal dynamic feature layer to the pattern matching rule layer is 160 times 5. The probability propagation path defines the activation and transformation relationship of state variables from the previous layer to the next layer, and its core calculation relationship formula is:

[0087]

[0088] in: Indicates that at a given spectral input and timing state Activation under certain conditions The probability of identifying nodes based on abnormal patterns. This represents the connection weight matrix from the spectral input layer to the temporal dynamic feature layer. Represents the transition from the temporal dynamic feature layer to the 1st... Each pattern matching rule layer determines the connection weight vector of a node. This represents a non-linear activation function, and the formula realizes the propagation of probability from the lower-level input to the higher-level rule pattern.

[0089] Optionally, the parameters of the weighted connections and probability propagation paths are obtained through training on historical sample data. In some embodiments, the construction of the spatial probabilistic inference network adopts a probabilistic graphical model framework to define nodes and connections. It can be understood that the standardized orchard canopy reflectance field, as the bottom spectral input layer of the spatial probabilistic inference network, provides the network with a real-time spectral observation basis, enabling the network to activate corresponding abnormal patterns based on the input spectrum and dynamic features. This involves the calculation of the probability propagation path and the threshold judgment process of the discrimination nodes in the pattern matching rule layer.

[0090] In one embodiment of the present invention, real-time acquired single-phase orchard UAV spectral images are input, and after preprocessing, real-time standardized canopy reflectance data is obtained. Real-time difference vegetation index and real-time normalized red edge index are calculated from the real-time standardized canopy reflectance data. The real-time difference vegetation index and real-time normalized red edge index are connected to the end of the spatial pixel historical multi-temporal spectral response curve to form an extended temporary temporal curve segment. This temporary temporal curve segment is input into the temporal dynamic feature layer of the spatial probabilistic inference network to trigger state inference along the probability propagation path inside the network. Based on the matching degree between the inference results and the spectral response anomaly pattern library, the membership degree of pest and disease occurrence at each spatial point is calculated. This process is specifically implemented as follows: the pattern matching rule layer of the spatial probabilistic inference network measures the similarity between the inferred spectral dynamic features and each standard spectral response anomaly pattern in the spectral response anomaly pattern library, calculates the matching probability value between the inferred spectral dynamic features and each standard spectral response anomaly pattern, and weights and fuses the matching probability values ​​corresponding to all standard spectral response anomaly patterns. The fused probability value is the membership degree of pest and disease occurrence at the spatial pixel point.

[0091] In practical implementation, the spatial distribution generation method for pests and diseases in karst mountain orchards using UAV spectral imagery uses a spatial probabilistic inference network to perform pixel-by-pixel spectral state extrapolation on real-time UAV spectral imagery of the orchard. Based on the matching degree between the extrapolation results and the spectral response anomaly pattern library, the membership degree of pest and disease occurrence at each spatial point is calculated. The following description is based on an example scenario of a karst mountain orchard and data comparison. In the example scenario, the spatial probabilistic inference network has been constructed and loaded. The end time point of the historical multi-time series spectral response curve is May 3, 2025. The spectral response anomaly pattern library contains 5 standard spectral response anomaly patterns. The real-time single-phase UAV spectral imagery of the orchard was taken on May 10, 2025.

[0092] In some embodiments, a spatial probabilistic inference network is used to perform pixel-by-pixel spectral state extrapolation on real-time orchard UAV spectral images. Real-time single-temporal orchard UAV spectral images are input and preprocessed to obtain real-time standardized canopy reflectance data. The preprocessing process includes radiometric calibration, atmospheric correction, and terrain shading correction, consistent with historical data processing. Real-time difference vegetation index and real-time normalized red edge index are calculated from the real-time standardized canopy reflectance data, using formulas consistent with those used in historical data processing. The real-time difference vegetation index and real-time normalized red edge index are connected to the ends of the historical multi-temporal spectral response curves of spatial pixels to form an extended temporary temporal curve segment. The ends of the historical multi-temporal spectral response curves contain a sequence of 80 time points, and the temporary temporal curve segment formed after connection contains 81 time points. This temporary temporal curve segment is input into the temporal dynamic feature layer of the spatial probabilistic inference network, triggering state extrapolation along the probability propagation path within the network. The state extrapolation process involves updating the implicit state variables of the temporal dynamic feature layer based on new input data within the network.

[0093] Understandably, in the example scenario, data comparison includes the differences in spectral dynamic features obtained from the same spatial pixel using only historical curves and those obtained after incorporating real-time data. In specific implementation, based on the matching degree between the inference results and the spectral response anomaly pattern library, the membership degree of pest and disease occurrence at each spatial location is calculated. The pattern matching rule layer of the spatial probabilistic inference network performs similarity measurement between the inferred spectral dynamic features and each standard spectral response anomaly pattern in the spectral response anomaly pattern library. The inferred spectral dynamic features are the state vector representation updated by the temporal dynamic feature layer, and the standard spectral response anomaly patterns are predefined vector templates. The similarity measurement calculates the matching probability value between the inferred spectral dynamic features and each standard spectral response anomaly pattern. The similarity metric is calculated using a formula based on dynamic time warping distance, which is expressed as follows:

[0094]

[0095] in: The deduced first The spectral dynamic characteristics and spectral response anomaly patterns of the 1st pixel in the database The matching probability value of a standard spectral response anomalous mode. It is an adjustable parameter. Indicates the first Spectral dynamic feature sequence of each pixel With the A standard pattern template sequence The dynamic time-normalized distance between them is used to weight and fuse the matching probability values ​​corresponding to all standard spectral response anomaly modes. The weighting coefficients of the weighted fusion are defined by expert knowledge based on the frequency and severity of various anomaly modes in historical samples. The fused probability value is the pest and disease occurrence membership degree of the spatial pixel point.

[0096] Optionally, the weighted fusion process follows a linear weighted summation rule. In some embodiments, the calculation of the matching probability value can be processed in parallel, comparing each pixel with all patterns in the spectral response anomaly pattern library. It is understood that the introduction of the real-time difference vegetation index and the real-time normalized red edge index ensures that the temporary time-series curve segment contains the latest spectral change information, which is crucial for triggering internal network state inference. Referring to Table 1, it shows the calculated matching probability of an example pixel with each pattern in the spectral response anomaly pattern library and the final weighted fusion degree of pest and disease occurrence membership.

[0097] Table 1: Matching Probability and Weight of Example Pixels

[0098] Standard Spectral Response Anomalous Mode Number Matching probability value ( ) Preset weights ( ) Mode 1 0.15 0.10 Mode 2 0.05 0.15 Mode 3 0.60 0.50 Mode 4 0.10 0.15 Mode 5 0.10 0.10 Weighted fusion of pest and disease occurrence membership 0.3275

[0099] See Figure 4This image intuitively presents the spatial pattern of orchard pest and disease risk generated based on UAV spectral imagery and a spatial probabilistic inference network. The X and Y coordinates in the image represent the orchard's geographic location in meters, and the color gradient corresponds to the pest and disease occurrence membership degree (range -0.4 to 1.4). The transition from green to red represents a change in membership degree from low to high, and the sampling points mark the spatial locations of on-site observations. Specifically, the generation of this heatmap is based on the following process: A pixel-by-pixel temporal state deduction is performed on real-time UAV spectral imagery using a spatial probabilistic inference network. Combined with the temporary temporal curve segments updated by the real-time difference vegetation index and normalized red edge index, the matching probability of pixel spectral dynamic features and the anomaly pattern library is calculated using dynamic time-normalized distance. Weighted fusion is then used to obtain the membership degree of each point. Kriging interpolation (combined with terrain auxiliary variables) is then used to achieve spatial continuity of the discrete probability matrix, and finally, the heatmap is formed through color rendering. The red high-value areas (membership degree ≥ 1.0) in the figure correspond to clusters with a higher risk of pest and disease occurrence, while the green low-value areas are low-risk areas. Their spatial distribution characteristics can support regional decision-making for precise pest and disease control in orchards.

[0100] In one embodiment of the present invention, the calculated membership degrees of pest and disease occurrence at each spatial point are assigned back to their corresponding geographic spatial coordinates to form a discrete probability matrix of pest and disease occurrence. Spatial interpolation is performed on the discrete probability matrix of pest and disease occurrence to generate a continuous probability surface covering the entire orchard area. Based on the probability value range of the continuous probability surface, a color rendering scheme corresponding to different probability levels from low to high is defined. The continuous probability surface is visualized and rendered according to the color rendering scheme, and the output is a probabilistic heat map of the spatial distribution of pests and diseases. The specific process of performing spatial interpolation on the discrete probability matrix of pest and disease occurrence to generate a continuous probability surface covering the entire orchard area is as follows: the discrete probability matrix of pest and disease occurrence is processed using the Kriging spatial interpolation method. Based on the orchard digital elevation model, altitude and slope are used as auxiliary variables in the Kriging interpolation calculation to reflect the possible impact of terrain on the spread of pests and diseases. The predicted probability value of each unsampled location in the orchard area is obtained through Kriging spatial interpolation calculation, and finally a smooth and continuous probability surface of pest and disease occurrence is formed.

[0101] In practical implementation, the method for generating spatial distribution of pests and diseases in karst mountain orchards using UAV spectral imagery generates a probabilistic heat map of the spatial distribution of pests and diseases based on the spatial continuity of the membership degree of pest and disease occurrence at each point in space. The following description is based on an example scenario and data comparison of a karst mountain orchard. In the example scenario, the membership degree of pest and disease occurrence for all pixels in the entire orchard area has been calculated, resulting in a membership degree matrix of pest and disease occurrence containing approximately five million pixels. Each pixel is associated with its corresponding geographic spatial coordinates. The orchard digital elevation model data has the same spatial resolution and coordinate reference system.

[0102] In practical implementation, based on the spatial continuity of the membership degree of pest and disease occurrence at various spatial points, a probabilistic heat map of pest and disease spatial distribution is generated. The calculated membership degree of pest and disease occurrence at each spatial point is assigned back to the corresponding geographic spatial coordinates, forming a discrete probability matrix of pest and disease occurrence. The assignment process is completed through geocoding, converting the row and column index of each pixel into geodetic coordinates, and assigning the calculated membership degree of pest and disease occurrence as an attribute value to the coordinate point. Spatial interpolation is performed on the discrete probability matrix of pest and disease occurrence to generate a continuous probability surface covering the entire orchard area. The Kriging spatial interpolation method is used to process the discrete probability matrix of pest and disease occurrence. Kriging spatial interpolation is a spatially optimal unbiased interpolation method based on the theory of variogram. Based on the orchard digital elevation model, altitude and slope are used as auxiliary variables in the Kriging interpolation calculation to reflect the possible impact of terrain on the spread of pests and diseases. The altitude and slope data are derived from the orchard digital elevation model and used as secondary variables for co-Kriging interpolation.

[0103] In some embodiments, the spatial interpolation process sets the search neighborhood radius to control the number of sample points participating in the interpolation calculation. The level division of the color rendering scheme can adopt equal intervals or non-equal intervals based on quantiles. It can be understood that the formation of discrete probability points of pest occurrence is the basis for spatial interpolation. The continuous probability surface is visualized and rendered according to the color rendering scheme, and the output is a heat map of the spatial distribution of probabilistic pests. The visualization rendering is completed in geographic information system software or a dedicated visualization engine. The raster data of the continuous probability surface is combined with the defined color lookup table to generate a color image with geographic coordinate reference.

[0104] See Figure 5 This study presents the spatially continuous distribution characteristics of orchard pest and disease occurrence probabilities obtained based on co-kriging interpolation (including elevation / slope). Specifically, the figure uses horizontal and vertical axes (unit: meters) to represent the orchard's geographical space, and the color scale on the right corresponds to the pest and disease occurrence probability level (0 for extremely low, 1 for extremely high). The spatial heterogeneity of the probability is intuitively presented through color rendering: the central area of ​​the figure is mainly dark blue and navy blue, corresponding to the "high-extremely high" probability level, while the edge areas are mostly light blue and light green, corresponding to the "medium-low" probability level. This reflects the effect of kriging interpolation with the assistance of terrain (elevation, slope) on the spatial continuity of pest and disease probability.

[0105] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0106] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for generating spatial distribution of pests and diseases in karst mountain orchards from UAV spectral imagery, characterized in that, Includes the following steps: Collect periodic UAV multispectral images covering the entire target karst mountain orchard area; Illumination difference compensation and terrain shadow correction are performed on periodic UAV multispectral images to generate a standardized orchard canopy reflectivity field; Based on the standardized orchard canopy reflectance field analysis, multiple preset waveband reflectance combinations are analyzed to construct multi-time series spectral response curves; Based on the distortion characteristics of historical pest and disease samples on multi-time-series spectral response curves, a library of abnormal spectral response patterns was identified. A spatial probabilistic inference network is constructed by integrating standardized orchard canopy reflectivity fields, multi-time-series spectral response curves, and a library of spectral response anomalies. The spatial probabilistic reasoning network is used to perform pixel-by-pixel spectral state inference on real-time orchard drone spectral images. Based on the matching degree between the inference results and the spectral response anomaly pattern library, the membership degree of pest and disease occurrence at each spatial point is calculated. Based on the spatial continuity of the membership degree of pest occurrence at various points in space, a probabilistic heat map of the spatial distribution of pests is generated.

2. The method for generating spatial distribution of pests and diseases in karst mountain orchards from UAV spectral imagery according to claim 1, characterized in that, Illumination difference compensation and terrain shading correction are performed on periodic UAV multispectral imagery to generate a standardized orchard canopy reflectance field, specifically including: Read the raw radiance values ​​of periodic UAV multispectral images and the corresponding solar altitude angle and azimuth angle data at the time of imaging; The topographic slope and aspect of each pixel in the karst mountain orchard were calculated using digital elevation model data. By combining solar altitude angle, azimuth angle data, and terrain slope and aspect data, the terrain shadow coefficient is calculated and the original radiance value is compensated. The radiance value after terrain shading compensation is converted into surface reflectance to generate a standardized orchard canopy reflectance field that eliminates the influence of terrain and light differences.

3. The method for generating spatial distribution of pests and diseases in karst mountain orchards from UAV spectral imagery according to claim 2, characterized in that, Based on the standardized orchard canopy reflectance field analysis of multiple preset waveband reflectance combinations, multi-time-series spectral response curves are constructed, specifically including: The reflectance values ​​of the preset blue light band, green light band, red light band, red edge band and near-infrared band are extracted from the standardized orchard canopy reflectance field. Calculate the vegetation index, which is the difference between the reflectance of the red band and the near-infrared band, and the normalized red edge index, which is the reflectance of the red edge band and the near-infrared band. By connecting the difference vegetation index values ​​and normalized red edge index values ​​of each spatial pixel at multiple monitoring time points in chronological order, a multi-time-series spectral response curve representing the vegetation growth changes of the spatial pixel is formed.

4. The method for generating spatial distribution of pests and diseases in karst mountain orchards from UAV spectral imagery according to claim 3, characterized in that, Based on the distortion characteristics of historical pest and disease samples on multi-time-series spectral response curves, a library of abnormal spectral response patterns was identified, specifically including: Obtain the spatiotemporal coordinates of historically verified pest and disease occurrence areas; Locate the curve segment from the multi-time-series spectral response curves that corresponds to the spatiotemporal coordinates of historical pest and disease occurrences; The distortion characteristics of the curve segment during the occurrence of pests and diseases, which are different from those of the healthy sample curve in terms of shape, slope and fluctuation amplitude, are analyzed, including the characteristics of earlier inflection point, attenuation of peak and rise of trough. The statistically significant distortion features are categorized into different types, and each type is defined as a standard spectral response anomaly pattern. All standard spectral response anomaly patterns constitute a spectral response anomaly pattern library.

5. The method for generating spatial distribution of pests and diseases in karst mountain orchards from UAV spectral imagery according to claim 4, characterized in that, By integrating standardized orchard canopy reflectance fields, multi-time-series spectral response curves, and a library of spectral response anomaly patterns, a spatial probabilistic inference network is constructed, specifically including: The standardized orchard canopy reflectance field is used as the bottom spectral input layer of the spatial probabilistic inference network. Multi-time-series spectral response curves are used as the temporal dynamic feature layer of the spatial probabilistic inference network; The spectral response anomaly pattern library is used as the pattern matching rule layer of the spatial probabilistic inference network; Within the spatial probabilistic reasoning network, a weighted connection and probability propagation path are established from the spectral input layer, through the temporal dynamic feature layer, and then to the pattern matching rule layer, enabling the network to activate corresponding abnormal patterns based on the input spectrum and dynamic features.

6. The method for generating spatial distribution of pests and diseases in karst mountain orchards from UAV spectral imagery according to claim 5, characterized in that, A spatial probabilistic inference network is used to perform pixel-by-pixel spectral state deduction on real-time orchard UAV spectral images, specifically including: Input real-time single-phase orchard UAV spectral imagery, and after preprocessing, obtain real-time standardized canopy reflectance data; Real-time difference vegetation index and real-time normalized red edge index are calculated from real-time standardized canopy reflectance data; The real-time difference vegetation index and the real-time normalized red edge index are connected to the end of the spatial pixel historical multi-time series spectral response curve to form an extended temporary time series curve segment. Input this temporary time-series curve segment into the temporal dynamic feature layer of the spatial probabilistic inference network to trigger state inference along the probability propagation path within the network.

7. The method for generating spatial distribution of pests and diseases in karst mountain orchards from UAV spectral imagery according to claim 6, characterized in that, Based on the matching degree between the inference results and the spectral response anomaly pattern library, the membership degree of pest and disease occurrence at each spatial location is calculated, specifically including: The pattern matching rule layer of the spatial probabilistic inference network measures the similarity between the inferred spectral dynamic features and each standard spectral response anomaly pattern in the spectral response anomaly pattern library. Calculate the matching probability value between the derived spectral dynamic characteristics and each standard spectral response anomalous mode; The matching probability values ​​corresponding to all standard spectral response anomaly modes are weighted and fused, and the fused probability value is the pest and disease occurrence membership degree of the spatial pixel point.

8. The method for generating spatial distribution of pests and diseases in karst mountain orchards from UAV spectral imagery according to claim 7, characterized in that, Based on the spatial continuity of the membership degree of pest and disease occurrence at various points in space, a probabilistic heat map of the spatial distribution of pests and diseases is generated, specifically including: The calculated membership degree of pest and disease occurrence at each spatial point is assigned back to the corresponding geographic spatial coordinates to form a discrete probability matrix of pest and disease occurrence. Spatial interpolation is performed on discrete probability lattices of pests and diseases to generate a continuous probability surface covering the entire orchard area. Based on the probability value range of the continuous probability surface, define color rendering schemes corresponding to different probability levels from low to high; The continuous probability surface is visualized and rendered according to the color rendering scheme, and the output is a heat map of the spatial distribution of probabilistic pests and diseases.

9. The method for generating spatial distribution of pests and diseases in karst mountain orchards from UAV spectral imagery according to claim 4, characterized in that, Locate the curve segment from the multi-time-series spectral response curves that corresponds to the spatiotemporal coordinates of historical pest and disease occurrences, specifically as follows: Based on the specific dates and orchard plot locations recorded in historical records of pest and disease occurrences, determine their index numbers and corresponding spatial pixel coordinates in the time series; Based on the time index number, a curve segment centered on the recorded date and containing a preset number of days is extracted from the multi-time-series spectral response curve corresponding to the spatial pixel coordinates. This curve segment is the target curve segment. Ensure that the captured target curve segment completely includes the spectral changes of the pest or disease from its initial occurrence to the time of recording.

10. The method for generating spatial distribution of pests and diseases in karst mountain orchards from UAV spectral imagery according to claim 8, characterized in that, Spatial interpolation is performed on the discrete probability lattice of pests and diseases to generate a continuous probability surface covering the entire orchard area, specifically: Kriging space interpolation is used to process discrete probability lattices of pest and disease occurrence. Based on the orchard digital elevation model, altitude and slope are used as auxiliary variables in the Kriging interpolation calculation to reflect the possible impact of topography on the spread of pests and diseases. The predicted probability value of each unsampled location within the orchard area is calculated by Kriging space interpolation, ultimately forming a smooth and continuous surface for the probability of pest and disease occurrence.

Images