A forest pest identification method based on data analysis

Abstract

This invention relates to the field of data analysis technology, and more specifically, to a method for identifying forest pests and diseases based on data analysis. The method includes: Step 1: acquiring aerial orthophotos of the target forest area and counting trapping points within a set time interval, and associating the trapping counts with the nearest-nearest-assigned sample area units; Step 2: constructing a complex simplex on a forest map grid, consisting of a patch layer, a sample area layer, and a trapping layer, and outputting suspected pest and disease triggering units after filtering out negative examples using a two-stage verifier; Step 3: establishing a pest and disease map database, describing typical patterns using disease type topological fingerprints; performing similarity assessment between the topological evidence package and the disease type topological fingerprints to generate disease type, extent, and confidence level, and labeling the data in layers on the forest map grid. This invention can capture fine-grained structural changes in pests and diseases at an early stage, and improves identification accuracy and interpretability by utilizing multi-source evidence synthesis.

Description

Technical Field

[0001] This invention belongs to the field of data analysis technology, specifically relating to a method for identifying forest pests and diseases based on data analysis. Background Technology

[0002] Monitoring and identifying forest pests and diseases has always been a crucial aspect of forestry protection and ecological management. Currently, researchers primarily rely on remote sensing imagery, ground surveys, and trapping devices to monitor the occurrence and spread of pests and diseases. For example, traditional remote sensing methods acquire infrared, visible, or multispectral images and use vegetation indices or spectral characteristics to infer the distribution of pests and diseases. These methods are efficient and have wide coverage in large-scale monitoring, but they have limitations in local identification and fine-grained lesion boundary detection. Early signs of pests and diseases often include localized discoloration of the tree canopy, leaf spots, or small-scale textural anomalies. These detailed features are easily overlooked or misjudged when image resolution is insufficient or lighting conditions are complex.

[0003] On the other hand, ground surveys can directly observe the type and severity of pests and diseases with high accuracy. However, due to the vast area and complex terrain of forest regions, manual surveys are time-consuming and labor-intensive, making it impossible to achieve high-frequency, full-coverage dynamic monitoring. Furthermore, manual survey results are easily influenced by the experience level of the surveyors, resulting in insufficient data standardization and difficulty in forming quantifiable and reproducible identification standards. The application of trapping devices provides direct information on the dynamic changes in pest populations. By counting the number of pests in the traps, the intensity of pest activity in the area can be determined. However, trapping data has spatial limitations, only representing the situation within a certain radius around the device. If the trapping points in the forest area are sparsely distributed or unevenly distributed, it often leads to insufficient data support, making it difficult to effectively integrate with remote sensing or ground imagery. In addition, trapping counts are greatly affected by weather, season, and environmental factors, exhibiting a certain degree of fluctuation. Relying solely on a single counting curve can easily lead to false alarms or missed alarms. Summary of the Invention

[0004] The main objective of this invention is to provide a data analysis-based method for identifying forest pests and diseases. This method constructs a forest atlas grid by acquiring aerial orthophotos and trapping counts. Based on this, a complex simplex layer is formed, comprising patch layers, sample area layers, and trapping layers. Topological features such as connectivity weakening, hole expansion, and expansion persistence are extracted using scale sequences and guard windows. Backfilled bridging edges are established when key features co-occur within the same time window. Furthermore, the spread process of pests and diseases is revealed through anomalous vortex kernels and trajectory chains. Finally, the topological evidence package is matched with the topological fingerprint of the disease type to output the disease type, extent, and confidence level. The advantages of this method are that it can capture fine-grained structural changes of pests and diseases at an early stage, improves identification accuracy and interpretability through multi-source evidence synthesis, ensures the stability of results through neighborhood consistency and cross-time window verification, and clearly distinguishes between different disease types, thus providing a reliable basis for the monitoring, early warning, and precise control of forest pests and diseases.

[0005] To solve the above problems, the technical solution of the present invention is implemented as follows:

[0006] A data analysis-based method for identifying forest pests and diseases, comprising:

[0007] Step 1: Obtain aerial orthophotos of the target forest area and the trapping point count within a set time interval. Generate a forest atlas grid according to a fixed grid, divide the sample area into units, generate patch fragments within the sample area units, and associate the trapping counts with the nearest-last-place sample area units.

[0008] Step 2: Construct a complex simplex on the forest map grid, consisting of a patch layer, a sample area layer, and a trapping layer. Scale sequences are formed based on color, texture, and count, and guard windows are set. Persistent events of connectivity, porosity, and expansion are recorded along the scale to obtain persistent pairs. When persistent patch porosity, trapping upscaling, and sample area connectivity weakening co-occur within the same guard window, a backfilled bridging edge is established in the sample area layer. Based on this, anomalous vortex kernels are extracted spatiotemporally, and anomalous vortex trajectory chains are grown. Topological evidence packages are synthesized according to the asymmetric weights of connectivity weakening, porosity expansion, and trapping upscaling. After two-stage verification to filter out counterexamples, suspected pest and disease triggering units are output.

[0009] Step 3: Establish a pest and disease map library, describe typical patterns with disease type topological fingerprints; determine the similarity between the topological evidence package and the disease type topological fingerprints, generate disease type, range and confidence level, and label them in layers on the forest map grid.

[0010] Further, in step 1, an aerial orthophoto of the target forest area and a set of daily counts of trapping points for the past 14 days are obtained. The aerial orthophoto includes three channels: red, green, and blue, with a spatial resolution of no less than 20 centimeters. The target forest area is divided into several sample area units according to a fixed grid with a grid side length of 10 meters. Within each sample area unit, patch fragments are generated using a superpixel segmentation method, limiting the number of patch fragments in each sample area unit to between 20 and 60. The trapping count is associated with the sample area unit based on the nearest trapping point within a distance of no more than 80 meters. If no trapping point falls within this range, it is marked as a trap-free sample area unit. After geometric registration and color white balance, a forest atlas grid is formed, which includes sample area units, patch fragments, and trapping point affiliations.

[0011] Furthermore, the complex simplex complex in step 2 consists of three layers: a patch layer, a sample area layer, and a trapping layer. The patch layer uses patch segments as nodes, establishing adjacency edges between contacting patch segments and creating triangular units in the shared boundary region of three patch segments. The sample area layer uses sample area units as nodes, establishing adjacency edges between four adjacent sample area units and creating quadrilateral units for any four square regions forming a grid. The trapping layer uses trapping points as nodes, establishing cross-layer association edges with the sample area units to which they belong, and creating cluster units for sets of sample area units belonging to the same trapping point. A cross-layer mapping table is established, recording one-to-one or one-to-many correspondences between the patch layer and the sample area layer, and between the sample area layer and the trapping layer.

[0012] Furthermore, the process of forming scale sequences and setting guard windows in step 2 includes: constructing scale sequences in the patch layer, sample area layer, and trap layer respectively, and managing and advancing them with a unified guard window; the scale sequence of the patch layer is discretized into 256 levels from low to high according to the green channel mean of the patch fragments; the scale sequence of the sample area layer is discretized into 100 levels from low to high according to the texture roughness level of the sample area unit, and the texture roughness is obtained by statistical analysis through the gray-level co-occurrence matrix; the scale sequence of the trap layer is discretized into 20 levels from low to high according to the quantile level of the cumulative count over the past 14 days; the patch layer, sample area layer, and trap layer all use guard windows with a length of 5 levels, and merging, rollback, and bridging operations are only allowed within the guard window during scale advancement.

[0013] Furthermore, in step 2, the sequence is synchronously advanced along the scales of the patch layer, sample area layer, and trapping layer to record the generation and merging events of connected components, thus obtaining persistent connectivity pairs; the occurrence and disappearance events of holes are simultaneously recorded in the patch layer and sample area layer to obtain persistent hole pairs; the continuous expansion events of clustered units are recorded in the trapping layer to obtain persistent expansion pairs; and a persistent event list is established for each sample area unit.

[0014] Furthermore, in step 2, when the same sample unit meets the following three conditions, a backfill bridging edge is introduced in the sample layer to stabilize the topological evidence chain: the first condition is that there is a persistent pair of holes in the patch layer and the persistence level is not less than 30; the second condition is that the trapping layer shows a monotonically increasing trend on two consecutive days and the quantile level increases by not less than 2 levels; the third condition is that a connectivity weakening event occurs in the sample layer within the current guard window; if all three conditions are met, a backfill bridging edge is established between the node where connectivity weakening occurs and the nearest node with stable connectivity within the same guard window in the sample layer, and the backfill bridging edge is retained until the next guard window.

[0015] Furthermore, in step 2, at the end of each guard window, anomaly vortex kernel determination is performed on all sample units. A sample unit is marked as an anomalous vortex kernel if it meets the following three anomaly criteria: The first criterion is that the number of pores in the patch layer is dominant within the neighborhood of the sample unit; the second criterion is that the connectivity of the sample layer decreases by at least one level compared to the previous guard window; the third criterion is that the quantile level of the trapping layer is located in the upper half of the highest 20th percentile in the region. The anomalous vortex kernel record includes the core... Sample area unit identifier, corresponding hole index and trapping quantile level; trajectory growth is performed in both space and scale starting from the anomalous vortex core. The trajectory growth strategy is: firstly, expand along the 8-neighborhood in the direction of decreasing connectivity, and secondly, expand along the scale sequence to higher levels; the step size of a single expansion is one neighborhood unit or one scale level; when two trajectories meet within 3 steps, they are merged into one trajectory chain. If the difference in the total persistence of the two trajectories exceeds 10 levels, the one with the larger total persistence is retained and the smaller one is marked as the subordinate trajectory.

[0016] Furthermore, in step 2, evidence for each abnormal vortex trajectory chain is summarized at three layers and synthesized with asymmetric weights: connectivity weakening evidence has a weight of 50, hole expansion evidence has a weight of 30, and trapping ascent evidence has a weight of 20. A topological evidence package is generated when the synthesized score is not less than 70 and contains at least two types of evidence. The topological evidence package includes a list of evidence types, support layer distribution, spatial coverage, guard window span, synthesized score, and the number of times backfill bridging edges are used. A two-level checker is executed on the topological evidence package of each sample area unit. The first-level robust consistency checker requires that connectivity weakening and hole expansion in the same direction occur simultaneously in two consecutive guard windows. The second-level neighborhood counterexample exclusion checker requires that the number of dominant features opposite to the topological evidence package in 8 neighborhoods does not exceed 1. Sample areas that pass the two-level checker are output as suspected pest and disease triggering units and are output together with their corresponding topological evidence packages.

[0017] Furthermore, in step 3, a pest and disease map library is established. The pest and disease map library describes typical patterns using disease type topological fingerprints, including at least three types of disease type topological fingerprints: leaf spot type, branch borer type, and canopy wilting type. The disease type topological fingerprint consists of essential items and optional items. The essential items must include at least one of connectivity weakening dominance or hole expansion dominance, and the optional items must include at least one of trapping rising support or backfilling bridging edge multiple triggers. The topological evidence package output in step 2 is compared with the pest and disease map library for similarity determination. If all essential items and at least one optional item of the corresponding disease type topological fingerprint are satisfied, it is determined to be a matching disease type. The matching disease types are divided into three confidence levels: high, medium, and low, according to the composite score. Pest and disease identification results are generated and displayed in layers on the forest map grid. When adjacent guard windows give different disease types for the same area unit, a safety backoff rule is executed, and the result with the higher confidence level is used, while the other result is recorded as pending review.

[0018] This invention provides a data analysis-based method for identifying forest pests and diseases, offering the following advantages: The invention simultaneously integrates aerial orthophotos and trapping count information within a forest atlas grid, constructing a complex simplex consisting of a patch layer, a sample area layer, and a trapping layer. It then extracts persistent features such as connectivity weakening, hole expansion, and anomalous vortices through topological data analysis. Compared to existing methods relying on single spectral indicators or empirical thresholds, this invention significantly improves the accuracy and stability of pest and disease identification. First, by limiting the scale progression range through a guard window mechanism, random fluctuations are confined to local areas, making the determination of persistent events more robust and effectively avoiding misjudgments caused by short-term illumination or count anomalies. Second, by employing a backfill bridging strategy, adjacent units are bridged when structural damage and trapping increases coexist, resolving the problem of false segmentation caused by shadows, roads, or local occlusion, and ensuring the integrity of the spatial continuity of lesions. Third, through bidirectional spatial and scale tracking of anomalous vortex kernels and trajectory chains, the dynamic process of pest and disease spread is presented in an interpretable trajectory manner, providing a basis for revealing the evolutionary path of lesions. Furthermore, the asymmetric weighting of evidence synthesis reinforces the dominance of structural evidence while also taking into account the biological support of trap data, ensuring that the judgment results are both physically verifiable and ecologically sound. Finally, through matching disease type topological fingerprints, clear distinctions can be made between leaf spot, branch borer, and canopy wilting types, enabling precise identification of different pest and disease types. The method's output not only includes confidence level stratification and spatial range but also performs safe backoff in conflict situations, ensuring consistency and traceability of results across guard windows, thus providing reliable technical support for early warning and precise control of forest pests and diseases. Attached Figure Description

[0019] Figure 1A schematic diagram of the method flow for the forest pest and disease identification method based on data analysis provided in an embodiment of the present invention;

[0020] Figure 2 This is a schematic diagram illustrating the statistical analysis of the occurrence and disappearance events of hole persistence provided in an embodiment of the present invention;

[0021] Figure 3 This is a schematic diagram showing the relationship between the number of times the backfill bridging edge is used and the span of the guard window, provided in an embodiment of the present invention. Detailed Implementation

[0022] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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 should fall within the scope of protection of the present invention.

[0023] refer to Figure 1 A data analysis-based method for identifying forest pests and diseases, comprising:

[0024] A data analysis-based method for identifying forest pests and diseases, comprising:

[0025] Step 1: Obtain aerial orthophotos of the target forest area and the trapping point count within a set time interval. Generate a forest atlas grid according to a fixed grid, divide the sample area into units, generate patch fragments within the sample area units, and associate the trapping counts with the nearest-last-place sample area units.

[0026] In the specific implementation process, firstly, aerial orthophotos of the target forest area are acquired, including red, green, and blue channels, with a ground resolution of no less than 20 cm. The georeferenced information inherent in the image is read, and the coordinates are checked for consistency with the projection to avoid deviations in subsequent grid positioning. The requirement of no less than 20 cm is because canopy boundaries, leaf textures, and fine linear structures of forest paths are difficult to distinguish at coarser resolutions, leading to over-smoothing of patch boundaries. Daily counts and planar coordinates of trapping points within a set time interval are obtained. The coordinates are checked for consistency with the image coordinate system; if inconsistencies exist, only coordinate system transformation is performed, without altering the original count sequence. The timestamps of the daily counts are retained, as subsequent step 2 will derive quantile levels in chronological order; therefore, they are not merged into a single cumulative value here. Count sequences with severe row gaps (e.g., more than 20% of the total days in the time interval) are removed to avoid misleading the most recently assigned sample unit. For the image, cloud and shadow coverage are checked; if cloud and shadow coverage exceeds 20% of the area, it is recommended to replace it with an image of better time and phase to prevent patch fragments from being mis-segmented under large areas of shadow.

[0027] Without introducing external data, the first step is to standardize the seams between image frames to eliminate displacement and abrupt brightness changes at overlapping seams. High-contrast details within overlapping areas are prioritized for alignment during stitching, as the repetitive features of detailed areas are more stable, reducing misalignment caused by local deformation. The image is then cropped to a buffer zone 20 meters beyond the target forest area boundary. This buffer prevents gaps in the boundary grid caused by overly tight cropping when dividing the image into 10-meter fixed grids, thus maintaining the continuity and traceability of sample unit numbering. If slight differences in ground resolution exist between different areas of the image, the image is resampled to achieve a uniform resolution, preventing sample units of the same size from containing different numbers of pixels at different locations, which could affect the stable control of patch fragment numbers.

[0028] A color white balance based on overall statistics is adopted. Specifically, the average brightness of the red, green, and blue channels across the entire image is statistically analyzed, and the average brightness of the three channels is brought to the same level. This reduces the overall color cast caused by differences in the color temperature of the light source during shooting, making the relative brightness relationship between green and earth tones in different areas more comparable. This is particularly important for patch generation because superpixel segmentation relies on color and local contrast; color cast can cause similar tree canopies to appear with different color clusters, thus resulting in fragmentation. Then, shadow and highlight masks are generated, specifically: shadow areas are usually significantly less bright and have reduced color saturation, while highlight areas (such as gravel roads or bare soil) are abnormally bright. These shadow and highlight areas are marked only for subsequent patch boundary constraints and quality marking, without modifying pixel values ​​at this stage. The reason is that forcibly increasing shadow brightness would introduce noise textures, which would interfere with superpixel aggregation; masking avoids over-segmentation in these areas.

[0029] Within the image space, a fixed square grid with 10-meter sides is established, and sample area units are generated row by row and column by column, each assigned a unique number. The advantage of the square grid is that it naturally aligns with the image pixel grid, reducing coordinate conversion errors, and facilitating the eight-neighborhood consistency check and cross-unit trajectory growth in subsequent step 2. For edge sample area units that cross forest boundaries, if the effective forest pixel ratio is less than 50%, they are marked as low-effective sample area units but retain their numbers. This strategy is better than discarding them directly because retaining the numbers ensures the integrity of adjacency relationships in subsequent spatial calculations, while the low-effectiveness marking reminds users to carefully interpret the segmentation results of this unit. Pixel counts, shadow mask coverage, and highlight mask coverage are recorded for each sample area unit as a basis for subsequent quality control. When the shadow or highlight coverage is higher than 50%, only the target number of patch fragments is reduced, without directly skipping the unit, thus avoiding the formation of spatial holes.

[0030] Superpixel segmentation is performed within each sample unit, with the number of target patch fragments controlled between 20 and 60. The reason for setting these upper and lower limits is that below 20 fragments will cause different canopies to be merged, resulting in the loss of canopy boundary information; above 60 fragments will generate a large number of fragments in texture-rich areas, increasing the noise density of subsequent topological events. The specific number of targets is adaptively determined based on the texture complexity within the unit. Texture complexity can be represented by contrast or energy indices obtained from common co-occurrence matrix statistics. When texture complexity is low (e.g., pure canopy without road intersections), a value close to 20 is used; when texture complexity is high (e.g., forest land interspersed with forest roads and open spaces), a value close to 60 is used. This adaptive approach can match the average size of patch fragments with scene complexity without changing the interval constraints, reducing over-segmentation or under-segmentation. Boundary merging and small patch merging strategies are enabled during segmentation. Boundary merging uses color gradients to bring the edges of patch fragments closer to the real canopy or road edges; for fragments that are too small and elongated in shape, if their color is similar to that of adjacent large patches, they are merged into adjacent patches. The benefit of this approach is reducing unstable small fragments, which can easily introduce false connectivity or false holes in subsequent connectivity event statistics. If the shadow mask coverage is high, the number of targets in patch fragments is reduced in the shadow area, prioritizing shape continuity. This is because color similarity deteriorates in shadow areas, and forcibly maintaining a high number of targets would unnecessarily segment the same canopy, compromising the reliability of subsequent connectivity measurements. After patch fragments are generated, the number of fragments within each sample unit, the number of pixels in each patch fragment, the average color, and the outer boundary are recorded. For units with fewer than 20 fragments, re-segmentation is triggered to adequately represent details; for units with more than 60 fragments, adjacent fragments are merged until they return to the interval. This backtracking and merging only occurs within the current sample unit, ensuring independence and repeatability between different units.

[0031] Using the center of the sample area unit as a representative location, the planar distance to all trapping points is calculated, and the nearest trapping point within 80 meters is selected to establish the attribution relationship. When no trapping point meets the criteria, the sample area unit is marked as a trap-free sample area unit. The reason for using 80 meters is that the representativeness of the activity of nearby insects under the small-scale diffusion conditions of common trapping devices in the forest is mainly concentrated within 100 meters. Using 80 meters can balance representativeness and spatial resolution. If the radius is too large, sample areas crossing forest roads or abrupt changes in forest type will be incorrectly assigned to the same trapping point, while if the radius is too small, a large number of units will lack data support. The nearest attribution is accelerated by spatial indexing, so that the whole-domain calculation can be completed within an acceptable time. For overlapping or dense trapping points, the nearest attribution naturally forms their own influence domain, which can avoid assigning a sample area unit to multiple counting sources at the same time, thereby maintaining the one-to-one correspondence between subsequent quantile levels and sample area units. For trapping point counting sequences with consecutive days of missing data, the original missing markers are retained without filling in the values. This approach allows for an explicit distinction between "true low activity" and "unobserved" scenarios during the scaling process in step 2, reducing the risk of misclassifying missing data as low activity. Time-series reference pointers are added to the assigned sample units, pointing to the daily counts within the defined time interval. Averaging or smoothing is not performed here to avoid making any transformations in step 1 that might mask short-term upward trends, providing complete input for step 2 to calculate the cumulative total and quantile ranking over the past 14 days.

[0032] A quality label is written for each sample unit, including shadow mask coverage, highlight mask coverage, whether the number of patch fragments is between 20 and 60, and whether the most recent affiliation of the trap point has been successfully established. The quality label is used for subsequent interpretation and verification of the identification results. A mapping relationship is established between the unique sample unit number and its geographical location, the list of patch fragments, and the trap point identifier. This mapping relationship ensures that the same input data can reproduce the same forest atlas grid results on different devices and implementations, which is a prerequisite for subsequent time-series comparison of topological data analysis.

[0033] After completing the above process, a forest atlas grid is formed, containing three basic elements and their quality labels: sample area units, patch fragments, and trap point affixes. This data structure is directly used as input for step 2 to construct the complex simplex, set the scale sequence and guard windows, and statistically analyze connectivity, voids, and expansion events during scale progression. Since step 1 does not perform any smoothing or filling on the count sequence, it can preserve the details of short-term rises, providing a realistic basis for triggering backfill bridging edges within the same guard window.

[0034] Optionally, the fixed grid can be divided using equal-area hexagonal honeycomb, with the side length corresponding to a 10-meter square according to the equal-area principle. Hexagons have a more isotropic adjacency structure, which can reduce directional bias in areas with significant topographic relief. When using hexagons, the subsequent eight-neighborhood check is replaced with a six-neighborhood check that matches the hexagon, while the rest of the process remains unchanged. Patch fragment generation can be achieved using region-growing-based segmentation or watershed-based segmentation. Region growing is suitable for coniferous forests with relatively uniform texture, which can reduce the jaggedness of the boundaries; watersheds are suitable for areas where roads and woodlands intersect, which can better separate along strong gradient lines. Regardless of which method is used, a target quantity range of 20 to 60 and a small patch merging strategy should be maintained to ensure consistency with subsequent topological event statistics.

[0035] Optionally, in forested areas with significant shadows, a shadow-referenced white balance can be used. This involves prioritizing the average color value of open or bare soil areas within the same time phase as a reference before adjusting the entire image. This approach more accurately restores color relationships under varying illumination and reduces segmentation drift caused by excessive darkening in shadowed areas. If a standard color chart is available on-site, it can be used as a reference. Using 80 meters as the default value, values ​​between 60 and 80 can be used in dense coniferous forests to reduce cross-forest mismatches, while values ​​between 80 and 100 can be used in sparse broadleaf forests to reduce the proportion of untrapped sample units. Adaptation can be triggered by estimating the average canopy diameter of sample units or by classifying road density indices. When using adaptive methods, the specific radius value needs to be recorded in the output to ensure the interpretability of subsequent results. For obvious water bodies, bare land, and large roads, a forest-free mask can be generated on the image. No patch fragments are generated within the mask; only the sample unit number is retained and marked as a non-forest unit. This avoids generating a large number of meaningless fragments in non-target areas, reducing noise in subsequent topological event statistics. If multiple usable images exist within a set time interval, the one with lower cloud cover and more uniform illumination can be selected first. When a single image is insufficient to cover the entire area, multiple images from the same day or adjacent days can be stitched together, still following the stitching and white balance process described above to ensure color and geometric consistency.

[0036] Step 2: Construct a complex simplex on the forest map grid, consisting of a patch layer, a sample area layer, and a trapping layer. Scale sequences are formed based on color, texture, and count, and guard windows are set. Persistent events of connectivity, porosity, and expansion are recorded along the scale to obtain persistent pairs. When persistent patch porosity, trapping upscaling, and sample area connectivity weakening co-occur within the same guard window, a backfilled bridging edge is established in the sample area layer. Based on this, anomalous vortex kernels are extracted spatiotemporally, and anomalous vortex trajectory chains are grown. Topological evidence packages are synthesized according to the asymmetric weights of connectivity weakening, porosity expansion, and trapping upscaling. After two-stage verification to filter out counterexamples, suspected pest and disease triggering units are output.

[0037] In the specific implementation process, the input is the forest atlas grid formed in step 1, which includes sample area units, patch fragments, and the attribution and quality labels of trap points. Step 2 uses this input as a basis to sequentially complete the construction of the complex simplex, the setting of scale sequences and guard windows, persistent event recording, the establishment of backfilled bridging edges, the extraction of anomalous vortex kernels, the growth of anomalous vortex trajectory chains, the synthesis of asymmetric weighted evidence, and two-level verification, outputting suspected pest and disease triggering units and topological evidence packages.

[0038] Step 2 begins by establishing adjacency edges for patch fragments that meet at their boundaries on the image. The smallest closed region enclosed by three fragments sharing a common boundary is denoted as a triangular unit. Triangular units are used because three fragments meeting easily form the smallest loop of canopy damage or holes; recording this structure allows for sensitive detection of the appearance and disappearance of holes during subsequent scale progression. Next, adjacency edges are established between four adjacent sample area units, using sample area units as nodes. Any four square regions enclosing a grid are denoted as quadrilateral units. Square units conform to the natural partitioning of a fixed grid, stably characterizing regional scale changes in canopy continuity and facilitating eight-neighborhood statistics. Finally, cross-layer association edges are established between the trap point and the sample area unit to which it belongs. A set of sample area units belonging to the same trap point is recorded as a cluster unit. The cluster unit characterizes the common influence domain of a trap point on several surrounding sample area units; this set will exhibit continuous expansion events as the count increases. A mapping table is established from the patch layer to the sample area layer, and from the sample area layer to the trap layer, allowing both one-to-one and one-to-many relationships. The mapping table is used to point events within a layer back to common sample area units, so as to determine whether the three types of evidence co-occur within the same guard window.

[0039] Mapping tables are established from patch layers to sample area layers and from sample area layers to trap layers, allowing both one-to-one and one-to-many relationships. These mapping tables are used to refer events within a layer back to common sample area units, enabling the determination of co-occurrence of the three types of evidence within the same guard window. A uniform guard window of 5 levels is used across all three layers, allowing merging, rollback, and bridging operations only within the same guard window. The guard window limits the "step size" of scale advancement, confining transient random fluctuations to local levels and preventing them from affecting consistent judgments across windows, thereby improving the robustness of persistent events.

[0040] Simultaneously advancing along the respective scale sequences of the three layers, the birth and merging of connected components are recorded through suprathreshold connectivity analysis, forming persistent connectivity pairs. Recording birth and merging measures the "length" of the maintained connectivity structure; a longer length indicates that the structure is less dependent on incidental chromatic aberration or noise. In the patch and sample area layers, the appearance and disappearance of pores are monitored based on loop units, forming persistent pore pairs. Recording pores first and then disappearance accurately describes the duration of the common lesion morphology of "internal cavities caused by boundary damage." In the trapping layer, the spatial coverage of swarming units is statistically analyzed level by level. Any coverage that continuously expands at adjacent levels is recorded as continuous expansion, forming persistent expansion pairs. The continuity of expansion reflects the spillover trend of insect activity from point to area, and is better able to exclude incidental factors than single-day peaks.

[0041] Within the same monitoring window, a backfill bridging edge is established if the same sample unit meets three conditions: a persistent pair of pores exists in the patch layer with a persistence level of not less than 30; the trapping layer shows a monotonically increasing quantile level over two consecutive days with an increase of not less than 2 levels; and a connectivity weakening event (decreased connectivity) occurs in the sample area layer. The purpose of setting these three conditions to occur simultaneously is to ensure that structural bridging only occurs when "structural damage, increased activity, and deteriorated regional connectivity" are met simultaneously, avoiding mistaking noise from a single source for lesion expansion. In the sample area layer, a backfill bridging edge is established between the node experiencing connectivity weakening and the nearest node with stable connectivity within the same monitoring window. Prioritizing nearest neighbors is to reconnect the same canopy layer that may be temporarily severed by shadows or small roads, preventing the amplification of connectivity and pores in subsequent statistics due to occasional shading. This backfill bridging edge is retained until the next monitoring window to observe whether bridging is still needed; if the triggering conditions are no longer met in the next monitoring window, it is automatically cleared to avoid long-term distortion of the true topology.

[0042] At the end of each guard window, a judgment is performed on all sample units to reduce false triggers caused by repeated fluctuations within the guard window. An anomalous vortex core is marked if the following three conditions are met: the number of pores in the patch layer is dominant within the eight-neighborhood of the sample unit; the connectivity of the sample layer decreases by at least one level compared to the previous guard window; and the trapping layer quantile is located in the upper half of the highest 20th percentile of the region. The design intent of combining these three conditions is to simultaneously incorporate "internal cavity expansion," "regional connectivity degradation," and "upward shift of biological pressure." The combined occurrence of these three factors is more consistent with the typical pathways of lesion formation, significantly reducing false positives caused by random textures and occasional count spikes. The core sample unit identifier, corresponding pore index, and trapping quantile are recorded as the starting point and constraint for subsequent trajectory growth.

[0043] Starting from the anomalous vortex core, the trajectory expands both spatially and scale-wise. It prioritizes expansion along the eight-neighborhood towards decreasing connectivity, followed by expansion along the corresponding scale sequence to higher levels. The spatial priority ensures the trajectory adheres to the actual damage propagation path, rather than being deviated from by single-point extrema; the scale-increasing suboptimal selection helps capture the "shallow-to-deep" intensification process. Each expansion step is limited to one neighborhood unit or one scale level. Limiting the step size prevents skipping multiple potential inflection points when crossing complex boundaries, thus preserving trajectory interpretability. Expansion stops upon encountering a boundary or when two consecutive expansions fail to meet the priority criteria. When two trajectories meet within three steps, they are merged into one trajectory chain; if the difference in the total persistence of two trajectories exceeds 10 levels, the one with the larger persistence is retained, and the smaller one is marked as the subordinate trajectory. Close-range merging reduces the risk of repeatedly labeling the same lesion boundary, and prioritizing the retention of total persistence makes the more stable and wider-covering trajectory the primary description of the region.

[0044] Along each anomalous vortex trajectory chain, the coverage of connectivity weakening evidence, aperformation evidence, and entrapment ascent evidence across the three layers was statistically analyzed. Asymmetric weighting was employed: connectivity weakening evidence received a weight of 50%, aperformation evidence received a weight of 30%, and entrapment ascent evidence received a weight of 20%. This weighting design reflects the fact that structural signals are less susceptible to short-term weather and sampling randomness compared to count signals, thus giving them a higher weight in synthesis. A topological evidence package was generated when the synthesized score was not less than 70 and contained at least two types of evidence. This dual condition ensured both strength and multi-source consistency in the results. The evidence type list, support layer distribution, spatial coverage, guard window span, synthesized score, and the number of times backfill bridging edges were used were recorded. The number of times backfill bridging edges were used was recorded separately because frequent bridging indicates the presence of persistent occlusion or fragmented cutting structures in the area; this information is instructive for subsequent review and maintenance.

[0045] The requirement is that connectivity weakening and hole expansion occur simultaneously in the same direction within two consecutive guard windows. The design of two consecutive windows is to eliminate isolated, one-off events, such as brief traces left by mechanical operations. It is also required that the number of dominant features opposite to the topological evidence package within eight neighborhoods does not exceed one. The inclusion of a counterexample count limit is to create a spatially consistent judgment environment, preventing occasional anomalies at the unit level from being mistakenly amplified across the entire area. Sample units that pass the two-level verification are output as suspected pest and disease triggering units and are output together with the corresponding topological evidence package, providing structured input for pest and disease map library matching in step 3.

[0046] When a sample unit has no trapping affixation, backfill bridging edges are not triggered, but scores can still be synthesized based on two types of evidence: connectivity weakening and void expansion. If such a unit passes two levels of verification, it is also output as a suspected pest-triggered unit, marked "no count support" for reference during subsequent review. When the quality label shows a shadow or highlight coverage rate higher than 50%, evidence packages are allowed to be generated, but a "quality concern" label is added to the output; simultaneously, the maximum number of steps for trajectory growth within that unit is limited to 1 to prevent large areas of low quality from causing trajectory deviation. If the three triggering conditions are no longer simultaneously met in subsequent guard windows, the backfill bridging edges established in the previous guard window are automatically deleted. The rollback mechanism ensures that structural bridging only exists when necessary and does not permanently alter the true topology.

[0047] For each sample unit, an event log is maintained, recording three types of persistent pairs, bridging, kernel, and trajectory operations in the order of the guard window, ensuring that any result can be traced back to a specific scale location and time sequence. The execution order is fixed as follows: construct the complex, set the scale and guard window, record the three types of persistent pairs, establish backfill bridging edges, extract abnormal vortex kernels, trajectory growth and merging, evidence synthesis, two-level verification, and output. The fixed order avoids discrepancies caused by inconsistencies in the order of operations between different implementations. The rank, percentile, and window length appearing in step 2 are all fixed values ​​to facilitate comparison across batches of data; if an optional implementation method is used, the specific values ​​must be specified in the output to ensure interpretability.

[0048] Optionally, the color sequence of the patch layer can be replaced with a single-channel sequence based on green saliency. This sequence improves the separability of leaf proportion by enhancing green and suppressing red-blue differences, making it suitable for evergreen forests with indistinct seasonal color changes. The advantage of this sequence is that leaf regions maintain a relatively stable order even with significant changes in light conditions, thus improving the stability of persistent pairs. The remaining procedures and guard window length remain unchanged. Hole detection can employ a "fill-difference" strategy: a global fill is performed on the closed regions of the patch layer and the sample area layer, and then the difference is calculated with the original region; the difference region is the candidate for holes. This strategy is more robust when there is significant boundary noise because it directly measures "internal gaps" and does not rely on the continuity of fragmented edges. Persistent pairs of holes are still recorded based on their appearance and disappearance.

[0049] Optionally, in forest transition zones, the trapping quantile level can use the sliding upper percentile of the historical distribution within the region as a threshold. For example, the boundary of the "highest 20th percentile" can be determined by the distribution dynamics of the most recent guard windows. This approach maintains stable sensitivity as the overall activity level shifts up or down, avoiding over- or under-detection caused by a fixed threshold. When multiple parallel trajectories are close together, a one-time match can be established between the ends of the trajectories, replacing the "meeting within 3 steps" condition with the merging condition of "end distance and direction consistency being satisfied." This approach reduces false merging in areas with many parallel structures such as roads or forest edges, preserving multiple true advance fronts. If a hexagonal grid is used in step 1, the eight-neighborhood in this step is replaced with a six-neighborhood; correspondingly, the neighborhood statistics in the anomaly vortex kernel and the secondary neighborhood counterexample exclusion checker are simultaneously adjusted to a six-neighborhood count. The remaining thresholds and procedures do not need to be changed. When traps are positioned high within multiple consecutive guard windows but there are no significant changes in the holes or connections, the retention period for the backfilled bridging edge can be extended from one guard window to two guard windows. This preserves the potential connection in situations where biological pressure precedes structural damage, preventing premature removal that could break the subsequent chain of evidence.

[0050] Step 3: Establish a pest and disease map library, describe typical patterns with disease type topological fingerprints; determine the similarity between the topological evidence package and the disease type topological fingerprints, generate disease type, range and confidence level, and label them in layers on the forest map grid.

[0051] The input to step 3 is the suspected pest and disease triggering unit output from step 2, along with its corresponding topological evidence package, anomalous vortex trajectory chain, and quality label. The objective of step 3 is to: based on the topological fingerprint of the pest and disease map library, perform similarity assessment on each suspected pest and disease triggering unit, generate the disease type, range, and confidence level, and display it in layers on the forest map grid; and review and save the discrepancies generated by adjacent guard windows according to the safety backoff rules.

[0052] The pest and disease atlas database stores "disease type topological fingerprints" by disease type. Each fingerprint contains required and optional items and records three types of information: evidence combination, temporal relationship, and cross-layer coverage.

[0053] Leaf spot type: A dominant characteristic is pore expansion; commonly, it presents as a sequence where multi-grained small pores appear first in the patch layer and later in the sample area layer; optional features include trapping upward supports or backfilling bridging edges for minor triggering. This design is based on the fact that leaf spots are mostly formed by the aggregation of small areas of necrosis on the leaf surface, first forming sparse cavities at the fine-grained level, and then appearing at the regional level. Branch borer type: A dominant characteristic is weakened connectivity; commonly, it presents as linear degradation along the sample area layer first, with pores in the patch layer being optional; optional features include repeated triggering of backfilling bridging edges or trapping upward supports. Degradation along the linear path corresponds to the spatial morphology of damaged branches, and repeated bridging suggests minor fragmentation. Canopy wilting type: A dominant characteristic is weakened connectivity, with pore expansion as a secondary feature; trapping upwards may be insignificant; commonly, it presents as a large-scale, continuous decrease in connectivity in the sample area layer and slow outward expansion. This type of fingerprint emphasizes overall degradation at the regional scale, weakening the dependence on local small pores.

[0054] Each fingerprint entry is derived from historical confirmed samples and expert review. Once an entry is approved, its required and optional items are fixed and do not change spontaneously in step 3. This ensures reproducibility across batches and regions, preventing changes in judgment criteria due to the randomness of individual cases in the field. For first-time applications in new regions, the example description and reference layer can be replaced without altering the entry structure, enhancing local adaptability.

[0055] For each suspected pest / disease triggering unit, the disease type that fully satisfies all the essential criteria is first screened from the pest / disease atlas database to form a candidate disease type set. The advantage of screening for essential criteria first is that it quickly eliminates fingerprints with inconsistent directions. For example, units with strong hole expansion but almost no connectivity degradation should not be initially classified as branch-boring types. The candidate disease type set is usually small in size, which facilitates subsequent detailed similarity determination.

[0056] Compare the fingerprints of candidate lesions, and verify item by item whether the evidence type, guard window span, and support layer distribution of the topological evidence package are consistent with the fingerprint description. The order of verification is: first check whether the essential items are fully covered, then check whether at least one of the optional items is hit. The order of essential items first and then optional items can grasp the lower limit of the reliability of identification and avoid being misled by accidental matching of individual optional items. Check whether the order of appearance and expansion direction of the abnormal vortex trajectory chain in the patch layer and sample area layer is consistent with the temporal relationship in the fingerprint. The reason for emphasizing temporal sequence is that lesions often follow a fixed evolutionary path of "local first, then regional" or "skeleton first, then pore". Static evidence alone can easily lead to certain construction cuts or short-term shadows being mistaken for lesions. Statistical evidence of simultaneous occurrence in the patch layer, sample area layer, and trap layer. The higher the proportion of simultaneous occurrence across layers, the more likely it is that the observed phenomenon is the projection of the same bio-structural event at different scales, rather than random phenomena from different sources. Consistency scores are obtained by weighting and synthesizing three aspects: evidence consistency, temporal relationship, and cross-layer coverage, and then normalized to a range of 0 to 100. The influence of structure-related items is appropriately increased during weighting because structure items are more stable than count items and are less sensitive to weather and sampling intervals. The scores are used for ranking rather than individual judgment; the final result still adheres to the principle of "all essential items are satisfied and at least one optional item is matched." The highest-scoring disease type in the candidate disease type set is selected as the matching disease type for the suspected pest or disease triggering unit. When the difference between the highest and second-highest scores is very small and they match different optional items, they are marked as "parallel candidates," and both are retained for further differentiation through spatial context during subsequent range aggregation and neighborhood consistency assessment.

[0057] At the unit level, regional growth is performed on the forest atlas grid based on matching disease types. The initial set consists of all suspected disease / pest triggering units with the same matching disease type. Expansion extends outwards along eight neighboring regions, with the condition that adjacent units have the same disease type and a consistency score not lower than a predetermined proportion of the initial unit. Using relative conditions instead of absolute thresholds accommodates differences in overall contrast across different forest stands, avoiding misclassification of large areas as low-confidence scatter points when overall contrast is weak. Expansion stops when adjacent units have different disease types or the consistency score significantly decreases, resulting in a connected region set. Holes within the set are then filled in a single operation, provided the hole area is much smaller than the surrounding area and the units within the hole do not show contrary evidence. The filling eliminates occasional gaps caused by small patches of shadow or road seams, making lesion boundaries more natural. The count of opposite disease types within the edge zone is calculated for each connected region. When the proportion of opposite counts within the edge zone is too high, the confidence level of that region is reduced, or the region is split into multiple smaller sub-regions. This measure can inhibit the excessive aggregation of "multi-disease cross-linking zones" and avoid treating the contact surface of two different diseases as a single lesion.

[0058] The matching results are categorized into three levels: high, medium, and low. High: All essential criteria are met, at least two optional criteria are met, or one optional criteria is met with a completely consistent temporal relationship, and neighborhood consistency is good. Medium: All essential criteria are met, at least one optional criteria is met, and there is a slight deviation in temporal relationship or a moderate proportion of opposite counts at the edges. Low: Only essential criteria are met, but optional criteria are not matched or neighborhood consistency is poor. This categorization method incorporates "evidence structure," "evolutionary rationality," and "spatial stability" simultaneously, avoiding relying solely on strong single-point evidence to elevate the overall judgment. A result record is generated for each connected region, including the disease type name, region number, outer boundary, area, centroid coordinates, guard window span, list of main supporting evidence, and credibility level. A one-to-one link is established between the result record and the event log from step 2, enabling full traceability from the final result back to the specific guard window and specific evidence.

[0059] Using sample area units as the basic drawing unit, colors are applied according to disease type, and transparency is adjusted according to confidence level. Higher confidence levels are less transparent, while lower confidence levels are more transparent, facilitating prioritization of high-confidence areas during large-scale browsing. Within suspected pest and disease triggering units, representative patch fragments in the patch layer are marked with markers indicating the type of evidence they originate from; these markers do not cover all fragments, only selecting representative locations to reduce visual clutter. Boundary lines are drawn for each connected region, with line types differentiated based on disease type. Boundary layers are used for overlay inspection with features such as roads and waterways to aid in understanding spatial relationships. A structured list is output simultaneously, containing core fields and quality notes for each region (e.g., "no count support," "high shadow coverage") for archiving and batch statistical analysis.

[0060] When adjacent guard windows provide different disease types for the same area cell, a conflict is identified. The result with the higher confidence level is saved first, and the other result is marked as pending review. If both results have the same confidence level, the disease type consistent with the previously confirmed result is selected to maintain temporal continuity; simultaneously, the backfill bridging edge near the cell is marked as "review focus," prompting step 2 to focus on checking the bridging rationality at that location in the next round of scale advancement. The backed-out result is recorded completely but not included in range aggregation. If two consecutive guard windows support the backed-out disease type, the backing-out is automatically lifted and the range is updated.

[0061] When a unit is marked as a "non-trapping sample area unit," the "trapping up support" option is automatically set to undetermined, ensuring the morphology is not downgraded due to the lack of this option; however, under the same conditions, fingerprints that do not depend on the trapping option are preferred. When the shadow or highlight coverage exceeds 50%, the confidence level is increased to medium at most, and no high is given. This limitation reminds users to be cautious with visual evidence while retaining structured output for subsequent on-site review. For edge sample areas with an effective forest pixel ratio of less than 50%, an "edge unit" note is added to the results list after successful matching. When aggregating ranges, at least one adjacent non-edge unit is required to provide common support before the region is included.

[0062] Optionally, while maintaining the essential and optional framework, the pest and disease atlas database can be expanded to include a "canopy striping pattern." This pattern is commonly seen in repetitive linear damage along roads or forest edges, and its fingerprint emphasizes the co-occurrence of directional connectivity degradation in the sample layer and a small number of elongated holes in the patch layer. Adding this pattern helps to separate non-planar lesions from "linear perturbations" and reduces confusion with branch-boring patterns. A seasonal adaptation table is attached to each pattern fingerprint, recording the weighting tendencies of trap support and hole salience in different seasons. For example, in early spring, trap rise often precedes structural damage, allowing for a moderately reliable match based on connectivity weakening and trap rise even in the absence of short-term hole expansion; while in late summer, more pronounced hole expansion is required. Seasonal adaptation reduces the bias of the same fingerprint throughout the year.

[0063] Optionally, when multiple discrete clusters of suspected pests and diseases of the same type are presented, aggregation and confidence classification can be completed within each cluster first, and then at a higher level, clusters can be checked to see if they are connected by narrow corridors. If the corridor is only supported by low-confidence units, it is retained as multiple independent regions; this approach avoids the "weak corridors" from incorrectly piecing together independent lesions. For units determined to be "parallel candidates," the decision can be delayed until the range aggregation is completed: if the vast majority of the eight neighboring regions of the unit match a certain type of disease and the confidence level is not lower than medium, then the unit is classified into that type of disease; otherwise, it remains in a parallel state and is marked as pending review. Delayed decision-making utilizes spatial context to enhance robustness. When displaying in layers, the boundary lines of high-confidence regions can be thickened and the span of the guard window can be marked at the center of gravity; medium-confidence regions can be represented by dashed boundaries; low-confidence regions can only be presented in a semi-transparent color in the raster display layer. Through differentiated display, users can focus their attention on more critical areas in a single browsing session.

[0064] A rectangular area with dimensions of 40 meters on each side within forest area A was selected as the example area. The ground resolution of the aerial orthophoto was set to 0.2 meters per pixel. A forest atlas grid was established using fixed squares with a side length of 10 meters, dividing the area into 16 sample units of 4x4. Two trapping points were set, located at planar coordinates (12,18) and (37,5) respectively. The time interval was set to daily counts over a continuous 14 days. Geometric homogenization and color white balance were performed on all pixels before calculation.

[0065] Let the row and column indices of the sample area cells be... and ,in and Define the grid side length as... Values Sample area unit The center coordinates are ,in These are the two-dimensional coordinates of the center of the sample area unit. For the target unit, we obtain Define the distance function as follows: ,in The coordinates of the center of the sample area unit are: These are the coordinates of the trapping point. The distance to the two trapping points is calculated as follows: and Recently assigned to ,and Satisfies the nearest home radius constraint.

[0066] make The location of the trapping point is at the Daily count, value sequence Define the cumulative total over the past 14 days as... ,get .make The empirical quantile is defined as the 14-day cumulative set of all sample units on the same day. ,in Let be the cardinality of the set. Define the 20-level quantile as . ,in To round up. In this embodiment, the global set is set on the 13th day. Make this unit ,get Set the global collection on the 14th day. Make this unit ,get Therefore, the monotonically increasing trend over two consecutive days with a cumulative increase in the percentile level is... .make For the number of target unit patch fragments, this embodiment uses... get The recorded shadow coverage was 0.18 and the highlight coverage was 0.06, both below the quality attention threshold of 0.5.

[0067] A complex simplex is constructed, consisting of a patch layer, a sample area layer, and a trapping layer. In the patch layer, triangular units are constructed with patch segments as nodes, adjacent segments as edges, and three segments sharing a common boundary. In the sample area layer, quadrilateral units are constructed with sample area units as nodes, four adjacent units as edges, and four units forming a quadrilateral unit. In the trapping layer, cross-layer association edges are constructed with trapping points as nodes and their corresponding units, and clustered units are recorded for units belonging to the same trapping point. Mapping tables are established from the patch layer to the sample area layer and from the sample area layer to the trapping layer. Let... The green channel level for patchy layers, the range to ;make The roughness level of the sample area layer texture, range to ;make The range is based on the 20 percentile levels accumulated over the past 14 days in the trap layer. to Define the length of the guard window as... In this example, three consecutive guard windows were selected in the patch layer. .

[0068] make For the sample area layer in the guard window The in-window average of the number of connected components within the window, let For the plaque layer in the guard window The average area of ​​the total area of ​​the holes inside the window (in square meters) is given by the following formula: For the trapping layer The 20th percentile level of the day. Obtained through suprathreshold connectivity and loop detection. ; By definition, a persistent connected pair and a persistent hole pair have appeared, and the hole continues to expand between two adjacent guard windows.

[0069] make A backfill bridging edge is established when all three of the following conditions are met: the patch layer has persistent pores with a grade of not less than 30; and Established; a connectivity weakening event occurred in the current guard window of the sample area layer. Based on the numerical values, the hole is in... to Durability rating over 30 Furthermore, the count showed a monotonic increase from the 13th to the 14th. This indicates that connectivity weakening holds, therefore... In the sample area layer, a backfill bridge edge is established between the node experiencing weakened connectivity and the nearest node with stable connectivity within the same window, and this edge is retained until... .make Let be the number of holes. This represents the connectivity level of the sample area. Statistical analysis shows that... and its eight neighboring domains, Dominant within the eight neighboring domains Compared Descend by at least one level, and Located in the upper half of the region's highest 20th percentile. Satisfies all three criteria, and is marked as such. This is the anomalous vortex core. The spatial step size is defined as one eight-neighborhood unit, and the scale step size is one level. The starting point is... Extend along the eight-neighbor region in a direction where connectivity continues to decrease. and One step each; then proceed along the green levels from Advance to Each step involves one step, resulting in two branches of length 2. The third step expands spatially to... It then meets the branch from the other side within 3 steps, merging into a single anomalous vortex trajectory chain. Let... The total persistence of the trajectory chain is measured as the sum of the persistence levels of each segment; in this example, we obtain... .

[0070] Define connectivity weakening metric, hole expansion metric, and trapping rise metric as follows: , , Let the composite score be... for ,in Let represent the normalized strengths of connectivity weakening, hole expansion, and trapping ascent, respectively. Substituting the values ​​yields... Considering the anomalous vortex trajectory chain in Keep it up and Increase cross-window stability gain Include, define The cross-window synthesis score is .because If at least two of the three types of evidence are non-zero, a topological evidence package is generated. Record the list of evidence types, the three-layer distribution, and the coverage area. The span of the guard window is to The backfill bridging edge can be used once.

[0071] make To guard the window The simultaneous occurrence of weakened connectivity and expanded voids indicates that... Count the negative examples for the eight neighboring domains. Upon inspection, the first-level verification was found to be in... and All meet Level 2 verification obtained Based on this, output Its extended units are suspected pest and disease triggering units.

[0072] The pest and disease atlas database contains topological fingerprints for three types of diseases: leaf spot, branch borer, and canopy wilt. Based on essential criteria, leaf spot is characterized by predominantly enlarged pores; branch borer is characterized by predominantly weakened connectivity along a linear path; and canopy wilt is characterized by decreased connectivity at the regional scale. Since this case exhibits strong pore enlargement, which is prominent initially in the patch layer, leaf spot and canopy wilt are included in the candidate disease set. A consistency score for evidence is defined. Time series consistency score Cross-layer coverage score The value ranges from 0 to 1. Let the overall score be... For the leaf spot type, the consistency of evidence was obtained by checking the essential and optional items. (Potential expansion dominates, with the option of hitting the upward support level), time series consistency. (The order of patch layer followed by sample area layer is basically consistent), cross-layer cover (The patch layer and the sample area layer are distinct, and the trapping layer is supported by a high quantile.) Based on this, we obtain... For the canopy wilting type, the main characteristic is a continuous decline in regional-scale connectivity, but the weight of void expansion is relatively low. ,get Therefore, the leaf spot pattern was chosen as the matching disease type. All suspected pest / disease triggering units matching the leaf spot pattern were used as the starting set. Let... The relative expansion threshold is defined as follows: ,in This is the highest overall score in the initial set. (Example) ,get The overall score in the eight neighboring domains is not lower than Furthermore, the units with the same disease type underwent region growth, ultimately resulting in a connected region composed of 5 sample units, with an area of ​​[area missing]. .

[0073] make The proportion of opposite disease types within the regional periphery was calculated. According to the trust level classification rules: essential requirements are met, at least one optional item is met, and the timing is consistent. The confidence level of this area is determined to be high. Checking the conflict records of adjacent guard windows, no instances were found where the same sample area unit was judged as different disease types in adjacent guard windows, so no backoff trigger is needed. If there are parallel alternatives, the disease type with the higher confidence level or consistent with the previous confirmation result will be used for determination. The output disease type name is "Leaf Spot Type," the suspected area covers 5 sample area units, and the unit with the highest overall score is... The span of the guard window is to The main supporting evidence includes the coexistence of hole expansion and connectivity weakening, the high position of the trapping quantile and its rise over two adjacent days, and one instance of backfill bridging. Three types of layers are generated using the forest atlas grid as the base map: the raster display layer is colored according to the disease type and the transparency is adjusted according to the confidence level; the tree node annotation layer places point markers at the locations of representative patch fragments; and the lesion boundary layer draws the circumscribed polygons of the connected regions.

[0074] Figure 2 This study illustrates the dynamic changes in porosity persistence in both patch and sample layer layers during forest pest and disease identification, following a sequence of guard windows. The horizontal axis represents the guard window number, and the vertical axis represents the porosity level, with the threshold line marked at persistence level 30. The experimental data shows a significant upward trend in porosity occurrence events (solid line bar chart) in the patch layer from the 1st to the 4th guard window, peaking at 85 porosity events in the 4th window. This reflects a rapid change in the topological structure of the pest-affected area during this period, with numerous cavities appearing within the patch fragments. Subsequently, the number of porosity occurrence events gradually decreased from the 5th to the 8th guard windows, to 45, 35, 35, and 15 respectively, indicating that pest and disease spread entered a relatively stable phase. The trend of porosity disappearance events in the sample layer (dashed line bar chart) is basically consistent with that in the patch layer, but with slight numerical differences. The 4th guard window also reaches its highest value, subsequently showing a downward trend. It is noteworthy that the rate of pore disappearance events in the sample area layer slows down after the fifth guard window, indicating that topological repair at the sample area scale requires a longer time period. When the pore persistence level in the patch layer exceeds the threshold of 30, the system triggers a backfill bridging mechanism to stabilize the topological evidence chain. As shown in the figure, the values ​​of the third and fourth guard windows significantly exceed the threshold line, meeting the triggering condition of "the presence of persistent pore pairs in the patch layer with a persistence level of not less than 30". This spatiotemporal distribution characteristic provides an important basis for the accurate identification of anomalous vortex nuclei and verifies the effectiveness of the complex simplex in multi-scale pest and disease detection.

[0075] Figure 3A dual Y-axis design was adopted to comprehensively demonstrate the intrinsic relationship between the number of times backfill bridging edges were used (scatter plot, left Y-axis) and the span of the guard window (step line, right Y-axis). The horizontal axis represents the guard window number, reflecting the evolution process over time. The number of times backfill bridging edges were used showed a clear unimodal distribution, reaching a peak of 27 times in the 5th guard window. This peak occurred in accordance with the triggering mechanism: when three conditions are met simultaneously—a porosity level of ≥30 in the patch layer, a quantile level increase of ≥2 in the trapping layer, and a connectivity weakening event in the sample area layer—the system will establish a backfill bridging edge in the sample area layer. From the 1st to the 5th guard window, the number of uses rapidly increased from 2 times to 27 times, reflecting the intensification of topological instability during the spread of pests and diseases. After the 5th guard window, the number of times backfill bridging edges were used showed a downward trend, dropping to 5 times by the 9th guard window. This indicates that as the anomalous vortex trajectory chain fully developed, the topological structure gradually stabilized, reducing the need for human intervention. The guard window span increases progressively within the first four windows, from level 1 to level 4, and then remains at level 5. This design ensures sufficient analysis windows during the rapid spread of pests and diseases, while maintaining a fixed span during the stable period to preserve detection accuracy. A trend line clearly reveals the negative correlation between the two: when the guard window span reaches its maximum value, the demand for backfill bridging edges begins to decrease. This phenomenon validates the design concept of retaining backfill bridging edges to the next guard window in this invention. By dynamically adjusting the lifespan of the bridging edges, the optimal stabilization effect of the topological evidence chain is achieved, laying a solid foundation for the accurate output of suspected pest and disease triggering units.

[0076] The above-described 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 the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying forest pests and diseases based on data analysis, characterized in that, The method includes: Step 1: Obtain aerial orthophotos of the target forest area and the trapping point count within a set time interval. Generate a forest atlas grid according to a fixed grid, divide the sample area into units, generate patch fragments within the sample area units, and associate the trapping counts with the nearest-last-place sample area units. Step 2: Construct a complex simplex on the forest map grid, consisting of a patch layer, a sample area layer, and a trapping layer. Scale sequences are formed based on color, texture, and count, and guard windows are set. Persistent events of connectivity, porosity, and expansion are recorded along the scale to obtain persistent pairs. When persistent patch porosity, trapping upscaling, and sample area connectivity weakening co-occur within the same guard window, a backfilled bridging edge is established in the sample area layer. Based on this, anomalous vortex kernels are extracted spatiotemporally, and anomalous vortex trajectory chains are grown. Topological evidence packages are synthesized according to the asymmetric weights of connectivity weakening, porosity expansion, and trapping upscaling. After two-stage verification to filter out counterexamples, suspected pest and disease triggering units are output. Step 3: Establish a pest and disease map library, describe typical patterns using disease type topological fingerprints; determine the similarity between the topological evidence package and the disease type topological fingerprints, generate disease type, range and confidence level, and add layered annotations on the forest map grid; Step 2, which involves forming scale sequences and setting guard windows, includes: constructing scale sequences for the patch layer, sample area layer, and trap layer respectively, and managing and advancing them using a unified guard window; the scale sequence for the patch layer is discretized into 256 levels based on the green channel mean of the patch fragments from low to high; the scale sequence for the sample area layer is discretized into 100 levels based on the texture roughness level of the sample area unit from low to high, with texture roughness obtained statistically through the gray-level co-occurrence matrix; the scale sequence for the trap layer is discretized into 20 levels based on the quantile level of the cumulative count over the past 14 days from low to high; the patch layer, sample area layer, and trap layer all use guard windows of 5 levels in length, and during scale advancement, merging, reversing, and bridging operations are only allowed within the guard window; In step 2, at the end of each guard window, all sample area units are judged for abnormal vortex kernels. A sample area unit is marked as an abnormal vortex kernel if it meets the following three anomaly criteria: The first criterion is that the number of pores corresponding to the persistence of the patch layer is dominant in the neighborhood of the sample area unit; the second criterion is that the connectivity of the sample area layer decreases by at least one level compared to the previous guard window; the third criterion is that the trapping layer quantile level is located in the upper half of the highest 20th percentile of the region. The abnormal vortex kernel record includes the core sample area unit identifier, the corresponding pore index, and the trapping quantile level. Trajectory growth is performed in both spatial and scale directions starting from the abnormal vortex kernel. The trajectory growth strategy is: firstly, expand along the 8-neighborhood towards the direction where connectivity continues to decrease, and secondly, expand along the scale sequence to higher levels; the step size for each expansion is one neighborhood unit or one scale level; when two trajectories meet within 3 steps, they are merged into one trajectory chain. If the difference in the total persistence of the two trajectories exceeds 10 levels, the one with the larger total persistence is retained, and the smaller one is marked as the subordinate trajectory. In step 2, evidence for each anomalous vortex trajectory chain is summarized at three layers and synthesized with asymmetric weights: connectivity weakening evidence has a weight of 50, hole expansion evidence has a weight of 30, and trapping ascent evidence has a weight of 20. A topological evidence package is generated when the synthesized score is not less than 70 and contains at least two types of evidence. The topological evidence package includes a list of evidence types, support layer distribution, spatial coverage, guard window span, synthesized score, and the number of times backfill bridging edges are used. A two-level checker is executed on the topological evidence package of each sample area unit. The first-level robust consistency checker requires that connectivity weakening and hole expansion in the same direction occur simultaneously in two consecutive guard windows. The second-level neighborhood counterexample exclusion checker requires that the number of dominant features opposite to the topological evidence package in 8 neighborhoods does not exceed 1. Sample areas that pass the two-level checker are output as suspected pest and disease triggering units and are output together with their corresponding topological evidence packages.

2. The forest pest and disease identification method based on data analysis as described in claim 1, characterized in that, In step 1, an aerial orthophoto of the target forest area and a set of daily counts of trapping points for the past 14 days are acquired. The aerial orthophoto includes three channels: red, green, and blue, with a spatial resolution of no less than 20 cm. The target forest area is divided into several sample area units according to a fixed grid with a side length of 10 meters. Within each sample area unit, patch fragments are generated using a superpixel segmentation method, limiting the number of patch fragments in each sample area unit to between 20 and 60. The trapping count is associated with the sample area unit based on the nearest trapping point within a distance of no more than 80 meters. If no trapping point falls within this range, it is marked as a trap-free sample area unit. After geometric registration and color white balance, a forest atlas grid is formed, which includes sample area units, patch fragments, and trapping point affiliations.

3. The forest pest and disease identification method based on data analysis as described in claim 2, characterized in that, The complex simplex in step 2 consists of three layers: a patch layer, a sample area layer, and a trapping layer. The patch layer uses patch segments as nodes, establishing adjacency edges between contacting patch segments and creating triangular units in the shared boundary region of three patch segments. The sample area layer uses sample area units as nodes, establishing adjacency edges between four adjacent sample area units and creating quadrilateral units for any four square regions forming a grid. The trapping layer uses trapping points as nodes, establishing cross-layer association edges with the sample area units to which they belong, and creating cluster units for sets of sample area units belonging to the same trapping point. A cross-layer mapping table is established, recording one-to-one or one-to-many correspondences from the patch layer to the sample area layer and from the sample area layer to the trapping layer.

4. The forest pest and disease identification method based on data analysis as described in claim 1, characterized in that, In step 2, the sequence is synchronously advanced along the scales of the patch layer, sample area layer, and trapping layer to record the generation and merging events of connected components, thus obtaining persistent connectivity pairs; the occurrence and disappearance events of holes are simultaneously recorded in the patch layer and sample area layer to obtain persistent hole pairs; the continuous expansion events of clustered units are recorded in the trapping layer to obtain persistent expansion pairs; and a persistent event list is established for each sample area unit.

5. The forest pest and disease identification method based on data analysis as described in claim 4, characterized in that, In step 2, when the same sample unit meets the following three conditions, a backfill bridging edge is introduced in the sample layer to stabilize the topological evidence chain: the first condition is that there is a persistent pair of holes in the patch layer and the persistence level is not less than 30; the second condition is that the trapping layer shows a monotonically increasing trend on two consecutive days and the quantile level increases by not less than 2 levels; the third condition is that a connectivity weakening event occurs in the sample layer within the current guard window; if all three conditions are met, a backfill bridging edge is established between the node where connectivity weakening occurs and the nearest node with stable connectivity within the same guard window in the sample layer, and the backfill bridging edge is retained until the next guard window.

6. The forest pest and disease identification method based on data analysis as described in claim 1, characterized in that, In step 3, a pest and disease map database is established. The pest and disease map database describes typical patterns using disease type topological fingerprints, and includes at least three types of disease type topological fingerprints: leaf spot type, branch borer type, and canopy wilting type. The disease type topological fingerprint consists of essential items and optional items. The essential items must include at least one of connectivity weakening dominance or hole expansion dominance, and the optional items must include at least one of trapping upward support or backfilling bridging edge multiple triggers. The topological evidence package output in step 2 is compared with the pest and disease map database for similarity determination. If all essential items and at least one optional item of the corresponding disease type topological fingerprint are satisfied, it is determined to be a matching disease type. The matched disease types are classified into three levels of credibility: high, medium, and low based on the composite score. The disease and pest identification results are generated and displayed in layers on the forest map grid. When adjacent guard windows give different disease types for the same area unit, the safety retreat rule is executed, and the result with the higher credibility level is used, while the other result is recorded as pending review.

Images