A forestry disaster three-dimensional monitoring and early warning method, system, terminal and medium

Abstract

This invention discloses a method, system, terminal, and medium for three-dimensional monitoring and early warning of forestry disasters, specifically relating to the field of forestry disaster monitoring and early warning. It includes collecting multi-height, multi-view forest area images of the same monitoring region within the same monitoring period, and writing a region code, acquisition time, height marker, viewpoint marker, and source marker into each forest area image, outputting an observation record sequence. This invention establishes a mechanism for identifying and supplementing observations related to the continuous development process of the same disaster by performing continuous correlation, missing level localization, supplementary observation generation, and insertion verification on asynchronously generated abnormal image results at different heights and viewpoints. This solves the problem that asynchronous abnormal image results are difficult to use for determining the continuous development process of the same disaster.

Description

Technical Field

[0001] This invention relates to the field of forestry disaster monitoring and early warning technology, and more specifically, to a three-dimensional monitoring and early warning method, system, terminal and medium for forestry disasters. Background Technology

[0002] Existing forestry disaster monitoring and early warning work usually revolves around whether disaster signs can be detected as early as possible and form usable early warning results. Common processing methods include comprehensively acquiring forest area image information through satellite remote sensing, drone patrols, tower-mounted equipment and ground terminals, then using machine learning to complete the identification of abnormal areas, the judgment of disaster categories and the marking of suspected locations, and organizing supplementary shooting, patrols or on-site verification based on the identification results. Taking continuous monitoring of understory smoldering and early-stage pests in mountainous forests as an example, satellites mainly acquire information from the top of the canopy, drones cover more of the canopy surface and forest gaps, and tower-mounted or ground-based equipment is more likely to capture changes in lateral branches, understory smoke, and lower leaves. However, they are also limited by conditions such as canopy obstruction, slope shading, cloud and fog disturbance, limited flight time, and insufficient verification capabilities. Under these circumstances, the abnormal signs of the same disaster formed at different heights and from different perspectives often have obvious differences in sequence. The current processing method of supplementing observations based on existing identification results is prone to the following: the first round of identification can only obtain fragmented local anomalies, and subsequent supplementary observations cannot be carried out in a targeted manner around the current ambiguity. Ultimately, in terms of operations, anomalies have appeared, but it is difficult to determine whether they belong to the continuous development process of the same disaster. Even after multiple supplementary observations, it is still difficult to eliminate misjudgments in time, and disasters that are truly in the early stage miss the opportunity for early warning in the process of repeated verification. Therefore, in the context of three-dimensional monitoring of forestry disasters, how to establish a mechanism for identifying and supplementing observations of abnormal image results that appear asynchronously at different heights and from different perspectives, in order to reduce the problems of delayed early disaster confirmation and excessively long misjudgment resolution cycles, has become an urgent technical problem to be solved. Summary of the Invention

[0003] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a method, system, terminal, and medium for three-dimensional monitoring and early warning of forestry disasters. By performing continuous correlation, missing level localization, supplementary observation generation, and insertion verification on abnormal image results asynchronously formed at different heights and perspectives, a mechanism for identifying and supplementing observations in connection with the continuous development process of the same disaster is established to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a three-dimensional monitoring and early warning method for forestry disasters, comprising: S1. Collect multi-height, multi-view forest area images of the same monitoring area within the same monitoring period, and write the area code, acquisition time, height mark, view mark and source mark for each forest area image, and output the observation record sequence; S2. For each observation record in the observation record sequence, divide the forest area image into image segments with corresponding region codes, input the image segments into the image recognition network to extract anomaly blocks, write the contour, center position, anomaly category, exposure level and observation record mark for each anomaly block, and output the anomaly block record set. S3. Based on the center position, outline, exposure level and acquisition time of the abnormal block record set, connect the preceding and following abnormal blocks in the same area code and between adjacent area codes in chronological order, and output the candidate development chain set. S4. For each candidate development chain in the candidate development chain set, locate the missing hierarchical segment between the previous anomaly block and the subsequent anomaly block, and generate supplementary observation instructions based on the center position of the subsequent anomaly block, the missing observation height of the previous anomaly block, the missing observation angle of the previous anomaly block, and the most recent executable time period after the subsequent anomaly block. S5. Obtain supplementary images and extract supplementary anomaly blocks according to the supplementary observation instructions. Insert the supplementary anomaly blocks into the missing level segment of the corresponding candidate development chain for re-verification. When a continuous hierarchical advancement relationship is formed, output the disaster confirmation result and early warning result. When a continuous hierarchical advancement relationship is not formed, output the misjudgment resolution result.

[0005] In a preferred embodiment, S1 includes: S1-1. Using the coordinates of the upper left corner of the monitoring area boundary as the starting coordinates, divide the area codes in an ascending row number and column number manner, and establish a mapping between each area code and its corresponding spatial location to output the set of area codes. S1-2. Extract the coverage area of ​​each forest area image within the same monitoring period, compare the coverage area with the set of area codes one by one, take the area code contained in the covered area and write it into the forest area image, and write the acquisition time, height mark, view mark and source mark at the same time, and output the image record set. S1-3. Based on the region code and acquisition time in the image record set, arrange the image records with the same region code in ascending order of acquisition time, and output the observation record sequence according to the arrangement result.

[0006] In a preferred embodiment, S2 includes: S2-1. For each observation record in the observation record sequence, extract the corresponding image fragment from the forest area image according to the area code, and output the image fragment set; S2-2. Input each image segment in the image segment set into the image recognition network, extract abnormal pixel regions, and merge the abnormal pixel regions with the boundary into abnormal blocks. Use the outer edge pixels of each abnormal block to form a contour, and use the geometric center of the region surrounded by the contour to form the center position. At the same time, read the corresponding abnormal category and exposure level, and output the abnormal block information set. S2-3. Write the information of each abnormal block in the abnormal block information set and the corresponding observation record's area code, acquisition time, height marker, viewpoint marker, and source marker in a fixed order to generate observation record markers. Then write the observation record markers into the corresponding abnormal blocks and output the abnormal block record set.

[0007] In a preferred embodiment, S3 includes: S3-1. Arrange the abnormal block record set in ascending order according to the acquisition time, and extract the previous and subsequent abnormal blocks within the same region code and between adjacent region codes between two adjacent acquisition times, and output the set of previous and subsequent abnormal blocks. S3-2. For each pair of anomaly blocks in the set of anomaly blocks, when the anomaly block and the anomaly block belong to the same region code, count the number of overlapping pixels in the outline-enclosed area of ​​the anomaly block and the outline-enclosed area of ​​the anomaly block. When the anomaly block and the anomaly block belong to adjacent region codes, count the number of overlapping pixels in the corresponding projections of the anomaly block outline and the anomaly block outline on the shared boundary of the region code. Calculate the direction of the line connecting the center position of the anomaly block and the center position of the anomaly block, as well as the direction of change from the exposure level of the anomaly block to the exposure level of the anomaly block. When the number of overlapping pixels is greater than zero, the number of overlapping pixels in the corresponding projections is greater than zero, and the direction of the line and the direction of change are consistent, determine that the anomaly block and the anomaly block satisfy the outline continuity relationship, the center continuity relationship, and the hierarchical advancement relationship, and output the set of anomaly blocks. S3-3. Connect the preceding and following abnormal blocks in the continuation abnormal block pair set end to end according to the acquisition time sequence, and write the sequence of abnormal blocks formed by continuous connection as a candidate development chain, and output the candidate development chain set.

[0008] In a preferred embodiment, S4 includes: S4-1. For each candidate development chain in the candidate development chain set, extract the previous and subsequent anomalous blocks in the order of collection time, calculate the difference between the exposure level of the subsequent anomalous block and the exposure level of the previous anomalous block, write complete tags to the previous and subsequent anomalous blocks when the difference is equal to one, and determine the missing level segment between the previous and subsequent anomalous blocks when the difference is greater than one. S4-2. For each missing level segment, read all height markers in the observation record sequence and arrange them in ascending order of value to form a height sequence. Read all viewpoint markers in the observation record sequence and arrange them in numerical order to form a viewpoint sequence. Then, according to the positions of the height markers of the preceding and following anomalies in the height sequence, extract the middle height marker to form a missing height sequence. According to the positions of the viewpoint markers of the preceding and following anomalies in the viewpoint sequence, extract the middle viewpoint marker to form a missing viewpoint sequence. Finally, determine the center position of the following anomaly as the target position of the missing level segment.

[0009] In a preferred embodiment, S4 further includes: S4-3. For each missing level segment, first determine the height candidate based on the missing height sequence. If the missing height sequence is not empty, write each height marker in the missing height sequence into the height candidate in sequence. If the missing height sequence is empty, write the height marker of the previous abnormal block into the height candidate. Then determine the view candidate based on the missing view sequence. If the missing view sequence is not empty, write each view marker in the missing view sequence into the view candidate in sequence. If the missing view sequence is empty, write the view marker of the previous abnormal block into the view candidate. S4-4. Combine the height candidates and viewpoint candidates one by one to form an observation combination, and perform region determination on each observation combination. When the region code of the previous anomaly block is the same as the region code of the subsequent anomaly block, keep the target position unchanged. When the region code of the previous anomaly block is adjacent to the region code of the subsequent anomaly block, write the intersection of the line connecting the center position of the previous anomaly block and the center position of the subsequent anomaly block with the boundary shared by the region codes into the target position. When the region code of the previous anomaly block is neither the same nor adjacent to the region code of the subsequent anomaly block, write a backtracking mark into the missing level segment. S4-5. For each observation combination formed by each missing level segment, if no backoff flag is written in the missing level segment, read the executable time period after the next anomaly block and take the first executable time period. Combine the target position, the height flag and the view flag in the observation combination to generate supplementary observation instructions and write them into the corresponding candidate development chain. If a backoff flag is written in the missing level segment, replace the previous anomaly block corresponding to the missing level segment with the anomaly block in the candidate development chain that is located before the previous anomaly block. Based on the replaced previous anomaly block and the next anomaly block, redetermine the missing height sequence, the missing view sequence, the target position and the supplementary observation instructions.

[0010] In a preferred embodiment, S5 includes: S5-1. Obtain supplementary images according to supplementary observation instructions, and input the supplementary images into the image recognition network to extract supplementary anomaly blocks. Write the outline, center position, anomaly category, exposure level and observation record mark for the supplementary anomaly blocks. S5-2. Insert the supplementary abnormal block into the missing level segment of the corresponding candidate development chain. Arrange the previous abnormal block, the supplementary abnormal block, and the subsequent abnormal block in the order of acquisition time. Calculate the difference in the exposure level between the previous abnormal block and the supplementary abnormal block, and the difference in the exposure level between the supplementary abnormal block and the subsequent abnormal block. Count the number of overlapping pixels between the outlines of the previous abnormal block and the supplementary abnormal block, and the number of overlapping pixels between the outlines of the supplementary abnormal block and the subsequent abnormal block. Calculate the direction of the line connecting the center position of the previous abnormal block and the center position of the supplementary abnormal block, and the direction of the line connecting the center position of the supplementary abnormal block and the center position of the subsequent abnormal block. S5-3. When the difference in the exposure level between the preceding anomaly block and the supplementary anomaly block is equal to 1, the difference in the exposure level between the supplementary anomaly block and the following anomaly block is equal to 1, the number of overlapping pixels between the outlines of the preceding anomaly block and the supplementary anomaly block is greater than zero, the number of overlapping pixels between the outlines of the supplementary anomaly block and the following anomaly block is greater than zero, and the direction of the line connecting the center positions of the preceding anomaly block and the supplementary anomaly block is consistent with the direction of the line connecting the center positions of the supplementary anomaly block and the following anomaly block, output the disaster confirmation result and the early warning result; if any one of these conditions is not met, output the misjudgment resolution result.

[0011] In a preferred embodiment, a three-dimensional monitoring and early warning system for forestry disasters includes: The observation and processing module is used to collect multi-height and multi-view forest area images of the same monitoring area within the same monitoring cycle, and write the area code, acquisition time, height mark, view mark and source mark for each forest area image, and output the observation record sequence. The anomaly extraction module divides the forest area image into image segments with corresponding region codes for each observation record in the observation record sequence, inputs the image segments into the image recognition network to extract anomaly blocks, writes the contour, center position, anomaly category, exposure level and observation record mark for each anomaly block, and outputs an anomaly block record set. The link construction module connects consecutive anomaly blocks within the same region code and between adjacent region codes in chronological order according to the center position, outline, exposure level and collection time of the anomaly block record set, and outputs a candidate development chain set. The observation generation module locates the missing hierarchical segment between the previous and subsequent anomalous blocks for each candidate development chain in the candidate development chain set, and generates supplementary observation instructions based on the center position of the subsequent anomalous block, the missing observation height of the previous anomalous block, the missing observation angle of the previous anomalous block, and the most recent executable time period after the subsequent anomalous block. The result verification module acquires supplementary images and extracts supplementary anomaly blocks according to the supplementary observation instructions. It then inserts the supplementary anomaly blocks into the missing level segment of the corresponding candidate development chain for re-verification. When a continuous hierarchical progression relationship is formed, it outputs disaster confirmation results and early warning results. When a continuous hierarchical progression relationship is not formed, it outputs misjudgment resolution results.

[0012] In a preferred embodiment, a three-dimensional monitoring and early warning terminal for forestry disasters includes: A processor, and a memory connected to the processor; The memory is used to store computer programs; The processor is used to call and execute the computer program in the memory to perform the described three-dimensional monitoring and early warning method for forestry disasters.

[0013] In a preferred embodiment, a three-dimensional monitoring and early warning medium for forestry disasters includes: The monitoring and early warning medium stores a computer program, which, when executed by a processor, implements the various steps of a three-dimensional monitoring and early warning method for forestry disasters.

[0014] The technical effects and advantages of this invention are as follows: 1. By uniformly writing forest area images from different heights, perspectives, and sources into regional codes and organizing them into observation record sequences, and then combining candidate development chains and missing hierarchical segments to generate supplementary observation instructions, it is possible to organize identification and supplementary evidence around the continuous development process of the same disaster, which can relatively alleviate the problems of delayed early disaster confirmation and excessively long misjudgment resolution cycle. 2. By extracting image segments according to region codes and merging abnormal pixel regions in the image segments, anomaly block records with contours, center positions, anomaly categories, exposure levels, and observation record markers are generated. This ensures that the anomaly blocks maintain a stable correspondence with their respective observation records, thereby providing a consistent data foundation for subsequent cross-time-sequence determination. 3. By calculating the contour continuity, center continuity, and hierarchical progression relationship of the anomaly blocks within the same area code and between adjacent area codes between adjacent acquisition times, and constructing candidate development chains accordingly, scattered local anomalies can be organized into a continuous anomaly sequence, thereby improving the ability to identify the continuous development process of the same disaster. 4. By calculating the difference in exposure levels between the preceding and following anomalous blocks in the candidate development chain, and further determining the missing height sequence, missing view sequence, and target location to generate supplementary observation instructions, the supplementary observations can directly correspond to the missing level segments, thereby relatively reducing the situation where the supplementary observations are disconnected from the current identification ambiguity. 5. By inserting the supplementary abnormal blocks extracted from the supplementary images into the missing layer segments, and re-verifying the difference in the exposure layer, the number of overlapping pixels in the contour, and the consistency of the connection direction between the previous abnormal block, the supplementary abnormal block, and the subsequent abnormal block, the candidate development chain can be supplemented or eliminated, thereby improving the pertinence and reliability of the early warning results to a certain extent. 6. By outputting misjudgment resolution results when supplementary observations cannot form a continuous hierarchical progression relationship, and outputting disaster confirmation and early warning results when supplementary observations form a continuous hierarchical progression relationship, the supplementary observation results can be directly converted into confirmation or exclusion conclusions, reducing the resource consumption of handling caused by repeated verification. Attached Figure Description

[0015] Figure 1 This is a flowchart of the method steps of the present invention.

[0016] Figure 2 This is a schematic diagram of the system modules of the present invention. Detailed Implementation

[0017] 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.

[0018] Refer to the instruction manual appendix Figure 1-2 The present invention provides a three-dimensional monitoring and early warning method for forestry disasters, comprising: S1. Collect multi-height, multi-view forest area images of the same monitoring area within the same monitoring period, and write the area code, acquisition time, height mark, view mark and source mark for each forest area image, and output the observation record sequence; In this implementation process, a unified spatial numbering system is first established for the monitoring area, then forest area images entering the system within the same monitoring period are assigned to the corresponding spatial locations, and finally, they are organized into an observation record sequence according to the area code and the time of acquisition. The purpose of this process is to unify forest area images formed by different observation heights, different observation angles and different sources to the same spatial location and the same time sequence, so that the image records corresponding to the same area code have a clear spatial ownership and sequential relationship. The implementation process includes the following steps: When using the top-left corner coordinates of the monitoring area boundary as the starting coordinates, first read the minimum x-coordinate, maximum x-coordinate, minimum y-coordinate, and maximum y-coordinate of the monitoring area boundary, and determine the minimum x-coordinate and maximum y-coordinate as the top-left corner coordinates; then, starting from these top-left corner coordinates, increase the column number horizontally by a fixed width and the row number vertically by a fixed height to divide the monitoring area into multiple rectangular areas, and generate area codes in the order of row number first, column number last; for the edge parts located within the monitoring area boundary but not enough to form a complete rectangular area, merge them into the rectangular area with the largest overlapping area, and use the area code of that rectangular area; after the area codes are generated, write the top-left corner coordinates, bottom-right corner coordinates, row number, and column number of the rectangular area corresponding to each area code into the area mapping table to obtain the area code set; the fixed width and fixed height can use the same values ​​for the same monitoring task, for example, if the monitoring area is divided into rectangular areas with a side length of 20 meters, then each area code corresponds to a ground area with a side length of 20 meters; When extracting the coverage area for each forest area image within the same monitoring period, the location data of that forest area image is first read. If the location data contains coverage area coordinates, these coordinates are directly taken as the coverage area. If the location data does not contain coverage area coordinates but contains the center coordinates of the shooting location, the shooting height, and the field of view, the top-left and bottom-right coordinates of the coverage area are calculated based on the ground projection width and height corresponding to the shooting height, using the center coordinates of the shooting location as the center. This gives the coverage area of ​​the forest area image. Then, the coverage area is compared one by one with the set of area codes, and the area codes that fall completely within the coverage area are written into the forest area image. For each forest area image, the region code that falls only partially within the coverage area and whose center coordinates fall within the coverage area is written into the image. Region codes that do not meet either of these conditions are not written into the image. After writing the region code, the acquisition time, height marker, viewpoint marker, and source marker are written into the image. The acquisition time is the time when the acquisition of the forest area image is completed, the height marker is the observation height number of the acquisition, the viewpoint marker is the observation viewpoint number of the acquisition, and the source marker is the acquisition device or platform number. When the same forest area image corresponds to multiple region codes, image records are generated one by one according to the region codes to obtain an image record set. When outputting the observation record sequence based on the region code and acquisition time in the image record set, all image records in the image record set are first grouped according to the region code, so that image records with the same region code are placed in the same record group. Then, the image records within each record group are arranged in ascending order of acquisition time, with the image records acquired earlier placed first and the image records acquired later placed last. When there are multiple image records with the same acquisition time under the same region code, they are further arranged in ascending order of height mark. If the height marks are the same, they are arranged in ascending order of viewing angle mark. If the viewing angle marks are the same, they are arranged in ascending order of source mark. After the arrangement of each record group is completed, the image records in each record group are output sequentially in ascending order of region code to obtain the observation record sequence. After the above processing, each forest area image corresponds to a specific region code, acquisition time, height marker, viewpoint marker, and source marker. Image records under the same region code have a specific chronological order. When extracting image fragments under the corresponding region code, the corresponding image records can be directly called according to the observation record sequence without having to perform spatial attribution and temporal sorting again. In practical applications: When a mountainous forest area is used as the monitoring area, the boundary coordinates of the forest area are first read and the area codes are divided into 20-meter by 20-meter sections to form an area mapping table. Within the same monitoring period, the system receives satellite images, UAV images, and tower-mounted images. The coverage area of ​​each image is extracted and written into the corresponding area code, along with the acquisition time, altitude marker, viewing angle marker, and source marker. For example, if a UAV image covers area codes A0105, A0106, and A0205, three image records are generated respectively. Then, the system sorts the data under each area code according to the acquisition time, altitude marker, viewing angle marker, and source marker, thus obtaining an observation record sequence that can be directly used for subsequent processing.

[0019] S2. For each observation record in the observation record sequence, divide the forest area image into image segments with corresponding region codes, input the image segments into the image recognition network to extract anomaly blocks, write the contour, center position, anomaly category, exposure level and observation record mark for each anomaly block, and output the anomaly block record set. In this implementation, image segments corresponding to each observation record are first obtained from the observation record sequence according to the region code. Then, anomaly blocks are extracted from the image segments, and the time, space, and observation attributes of the corresponding observation records are written into the same record structure to form the anomaly block record set required for connecting the previous and subsequent anomaly blocks. In specific processing, the entire forest area image is not directly used as the anomaly identification object. Instead, spatial truncation is first completed according to the region code, so that each image segment corresponds to only one region code. Then, the abnormal pixel regions are extracted within the image segment and merged into anomaly blocks according to a unified pixel connection rule. Then, the contour, center position, anomaly category, and exposure level are generated for each anomaly block. Finally, the anomaly block information is written into the region code, acquisition time, height marker, viewpoint marker, and source marker of the observation record to which the anomaly block belongs in a fixed order to form an anomaly block record set that can be directly used in subsequent calculations. The implementation process includes the following steps: For each observation record in the observation record sequence, first read the corresponding forest area image and region code, then read the upper left and lower right corner coordinates of the rectangular region corresponding to the region code from the region mapping table, and use this rectangular region as the cropping range; when a one-to-one correspondence between image coordinates and spatial coordinates has been established in the forest area image, directly convert the upper left and lower right corner coordinates into the cropping boundary in the image coordinates, and crop image segments from the forest area image according to the cropping boundary; when the forest area image only has coverage area coordinates, first match the upper left to lower right corner coordinates of the coverage area with the image coordinates. The width and height correspondence is used to establish horizontal and vertical proportions. Then, the spatial boundary of the rectangular area corresponding to the region code is converted into the pixel boundary in the image, and image segments are cropped according to the converted pixel boundary. When the rectangular area corresponding to the region code is located at the edge of the forest area image and partially exceeds the image range, only the part that falls within the image range is cropped and the region code is retained unchanged. The above cropping process is performed on each observation record to obtain a set of image segments that correspond one-to-one with each observation record. After this processing, each image segment has a unique region code source, and the spatial attribution is no longer repeatedly judged when generating subsequent anomaly blocks. When image segments from a set of image segments are input into an image recognition network, the network first outputs abnormal pixel regions belonging to anomalous targets within each segment. The network then scans each abnormal pixel region pixel-by-pixel, merging adjacent abnormal pixel regions into a single anomalous block. The boundary alignment follows an eight-neighbor rule, meaning two anomalous pixels are considered adjacent horizontally, vertically, or diagonally. After merging, the outermost abnormal pixels of each anomalous block are extracted to form a contour. The average horizontal and vertical coordinates of all abnormal pixels within the contour are used as the geometric center of the block, and this geometric center is recorded as the center position. Subsequently, the anomalous category and exposure level corresponding to the block are read. The anomalous category is the category result output by the image recognition network for that block, and the exposure level is the position of the anomalous block within the image recognition network. In the hierarchical results of the network output, if the image recognition network output provides multiple exposure levels for the same abnormal block, the exposure level with the most occurrence pixels is taken as the exposure level of the abnormal block. The above processing is performed on each image segment to obtain the abnormal block information set. Here, the outline, center position, abnormal category, and exposure level of the abnormal block are all generated from the same abnormal block to avoid the problem of inconsistent field sources when performing subsequent calculations between previous and subsequent abnormal blocks. For example, in the forest smoldering monitoring scenario, an abnormal pixel area composed of smoke and high temperature edges may appear in the image segment under the same area code. After merging eight neighborhoods, an abnormal block is formed. The outline is then formed by the outer edge pixels of the abnormal block, and the center position is obtained from the average coordinate of the abnormal pixels inside. The smoldering category and the corresponding exposure level are read simultaneously. When writing the information of each anomaly block in the anomaly block information set along with the corresponding observation record's region code, acquisition time, height marker, viewpoint marker, and source marker in a fixed order, the observation record corresponding to the image segment that generated the anomaly block is read first. Then, the region code, acquisition time, height marker, viewpoint marker, and source marker in the observation record are concatenated in the following order: region code first, acquisition time second, height marker second, viewpoint marker third, and source marker last, forming an observation record marker. This observation record marker is then written into the corresponding anomaly block and saved together with the anomaly block's outline, center position, anomaly type, and exposure level as an anomaly. Block recording; when multiple anomalous blocks are extracted from the same image segment, the same observation record tag is written to each anomalous block; when anomalous blocks are extracted from the same region code at different acquisition times, each anomalous block is written with the observation record tag of its respective observation record; after all are written, an anomalous block record set is obtained; through this processing, each anomalous block record retains both the spatial morphology and category hierarchy information of the anomalous block itself, as well as the time, space and observation attributes of its respective observation record. When extracting the previous and subsequent anomalous blocks in S3, the required fields can be directly read from the anomalous block record without having to backtrack to the original image or the original observation record; After the above processing, each observation record in the observation record sequence is converted into an anomaly block record with anomaly block information and observation record tags. The outline, center position, anomaly category and exposure level of the anomaly block have a clear source. The composition order of the observation record tags is fixed. The correspondence between the anomaly block and the observation record is consistent. When connecting the previous anomaly block and the next anomaly block in the same area code and between adjacent area codes, the anomaly block record set can be directly used as input. In practical applications: When an observation record corresponding to a certain region code comes from a UAV forest area image, the system first extracts an image segment from the forest area image based on the boundary coordinates of the region code in the region mapping table; then, the image segment is input into the image recognition network to obtain several abnormal pixel regions, which are then merged into two abnormal blocks according to the eight-neighbor adjacency rule. One abnormal block is identified as a pest category, and the other abnormal block is identified as a smoldering category. The system extracts the outline and center position of the two abnormal blocks respectively and determines their respective exposure levels; then, the region code, acquisition time, altitude marker, viewpoint marker, and source marker in the observation record are read, and observation record markers are generated in a fixed order and written into the two abnormal blocks respectively. Finally, two abnormal block records are formed and written into the abnormal block record set for use in subsequent steps.

[0020] S3. Based on the center position, outline, exposure level and acquisition time of the abnormal block record set, connect the preceding and following abnormal blocks in the same area code and between adjacent area codes in chronological order, and output the candidate development chain set. In this implementation process, the anomaly block record set is first organized into an anomaly block sequence with a clear time order. Then, between two adjacent acquisition times, preceding and following anomaly blocks with a basis for continuity are extracted. Subsequently, based on regional relationships, contour continuity relationships, center continuity relationships, and hierarchical progression relationships are calculated respectively. Finally, preceding and following anomaly blocks that meet the continuity conditions are connected end to end in chronological order to form a candidate development chain. The focus of this process is not on judging whether an anomaly block exists at a single moment, but on judging whether the anomaly blocks at two adjacent acquisition times belong to the continuous development process of the same disaster in terms of spatial location and exposure level. Therefore, it is necessary to first fix the extraction criteria for preceding and following anomaly blocks, and then perform continuity calculations separately for the two cases of the same regional code and adjacent regional codes to avoid mixing calculation fields under different regional relationships. The implementation process includes the following steps: When sorting the abnormal block record set by acquisition time in ascending order, first read the acquisition time, region code, center position, outline, and exposure level of each abnormal block record, and sort them from earliest to latest acquisition time. When two abnormal block records have the same acquisition time, they are sorted in ascending order by region code; if the region codes are the same, they are then sorted in ascending order by the horizontal coordinate of the center position; if the horizontal coordinates are the same, they are sorted in ascending order by the vertical coordinate, thus forming a unique sorting result for the abnormal block record set. After sorting, two adjacent acquisition times are taken as a time pair, and the abnormal block under the previous acquisition time is taken as the previous abnormal block, and the subsequent abnormal block is taken as the next abnormal block. An abnormal block at a given acquisition time is designated as the subsequent abnormal block. For each time pair, the preceding and subsequent abnormal blocks are first extracted within the same region code to form a pair of abnormal blocks in the same region. Then, according to the adjacency relationship of the region codes recorded in the region mapping table, the abnormal block pairs adjacent to the region codes of the preceding and subsequent abnormal blocks are extracted to form a pair of abnormal blocks in adjacent regions. The pairs of abnormal blocks in the same region and the pairs of abnormal blocks in adjacent regions are merged to obtain a set of preceding and following abnormal block pairs. After this processing, both the preceding and subsequent abnormal blocks are limited to the time between two adjacent acquisition times, and each pair of preceding and following abnormal blocks has a clear regional relationship. For each pair of preceding and following anomalous blocks in the set of preceding and following anomalous blocks, first determine the region code relationship between the preceding and following anomalous blocks; when they belong to the same region code, first map the contour bounding areas of the preceding and following anomalous blocks to the same image coordinate system, then count the number of pixels that simultaneously fall within the contour bounding areas of both the preceding and following anomalous blocks, and determine this number as the number of overlapping pixels; when the number of overlapping pixels is greater than zero, it is determined that the preceding and following anomalous blocks satisfy the contour continuity relationship; when the number of overlapping pixels is equal to zero, it is determined that the preceding anomalous block... The preceding and following anomaly blocks do not satisfy the contour continuity relationship; subsequently, the direction of the line connecting the center positions of the preceding and following anomaly blocks is calculated. Specifically, the horizontal difference is obtained by subtracting the horizontal coordinate of the preceding anomaly block's center position from the horizontal coordinate of the following anomaly block's center position, and the vertical difference is obtained by subtracting the vertical coordinate of the preceding anomaly block's center position from the vertical coordinate of the following anomaly block's center position. When the horizontal difference is zero and the vertical difference is greater than zero, the line direction is recorded as downward; when the horizontal difference is zero and the vertical difference is less than zero, the line direction is recorded as upward; when the vertical difference is zero and the horizontal difference is greater than zero... When the vertical difference is zero and the horizontal difference is less than zero, the direction of the connection is recorded as to the right; when the vertical difference is zero and the horizontal difference is less than zero, the direction of the connection is recorded as to the left; when both the horizontal and vertical differences are not zero, they are recorded as upper left, upper right, lower left, and lower right respectively, according to the positive and negative combinations of the horizontal and vertical differences; then, the change direction is generated based on the change in the exposure level of the previous anomaly block to the exposure level of the subsequent anomaly block. When the exposure level of the subsequent anomaly block is higher than that of the previous anomaly block, the change direction is recorded as upward; when the exposure level of the subsequent anomaly block is lower than that of the previous anomaly block, the change direction is recorded as downward; when both are exposed... When the levels are the same, the direction of change is recorded as translation. For the same area code, when the outline continuity relationship is established, the center position has a connecting direction, and the direction of change corresponds to the connecting direction, it is written into the same continuity judgment result. This determines that the previous abnormal block and the subsequent abnormal block satisfy the outline continuity relationship, the center continuity relationship, and the hierarchical advancement relationship. Among them, the correspondence between the direction of change and the connecting direction is established in advance according to the exposure level table. For example, when the exposure level increases from the understory to the canopy, pushing up corresponds to upward, upper right, and upper left; moving down corresponds to downward, lower right, and lower left; and translating corresponds to left and right. When the current anomalous block and the subsequent anomalous block belong to adjacent region codes, first read the shared boundary of the two region codes in the region mapping table. If the two region codes are vertically adjacent, the shared boundary is the boundary line formed by the coincidence of the lower boundary of the upper region code and the upper boundary of the lower region code. If the two region codes are horizontally adjacent, the shared boundary is the boundary line formed by the coincidence of the right boundary of the left region code and the left boundary of the right region code. After determining the shared boundary, project the outlines of the previous and subsequent anomalous blocks onto the shared boundary respectively, and count the number of pixels that overlap on the shared boundary. This number of pixels is determined as the corresponding number of overlapping pixels. When the corresponding number of overlapping pixels is greater than zero, it is determined that the previous and subsequent anomalous blocks satisfy the outline continuity relationship. When the corresponding number of overlapping pixels is equal to zero, it is determined that the previous and subsequent anomalous blocks do not satisfy the outline continuity relationship. Then, calculate the center position of the previous anomalous block and the subsequent anomalous block in the same way as in the case of the same region code. The direction of the line connecting the center of the constant block is determined, and the change direction is generated according to the change in the exposure level from the previous abnormal block to the subsequent abnormal block. When the number of overlapping pixels in the corresponding projection is greater than zero, there is a line connecting the center and the change direction corresponds to the line connecting direction, it is determined that the previous abnormal block and the subsequent abnormal block satisfy the contour continuity relationship, the center continuity relationship, and the hierarchical advancement relationship. When any of the above conditions are not met, it is determined that the previous abnormal block and the subsequent abnormal block do not satisfy the continuity condition. The above same region code branch and adjacent region code branch calculations are performed on all abnormal block pairs in the set of previous and subsequent abnormal block pairs respectively, and the abnormal block pairs that satisfy the continuity condition are written into the continuous abnormal block pair set. The same region code and adjacent region code are processed separately here because their contour continuity calculation basis is different: the former is based on the number of overlapping pixels in the contour enclosed area, and the latter is based on the number of overlapping pixels in the shared boundary projection. The two are not taken at the same time, so as to avoid the mixing of calculation conditions under different regional relationships. When connecting the preceding and following abnormal blocks in the continuation abnormal block pair set according to the acquisition time sequence, first establish an initial connection segment with the preceding abnormal block as the chain head and the following abnormal block as the chain tail; then read the subsequent abnormal block pairs in the continuation abnormal block pair set, and determine whether its preceding abnormal block is consistent with the chain tail of the existing connection segment. If they are consistent, then the following abnormal block is continuated to the end of the connection segment; if they are inconsistent, then a new connection segment is established with the preceding abnormal block. When a following abnormal block corresponds to multiple preceding abnormal blocks, the preceding abnormal block with the acquisition time closest to the following abnormal block is retained first. When the acquisition times are the same, the preceding abnormal block with the larger number of pixels corresponding to the contour continuation relationship is retained. The preceding anomalous blocks are selected; when a preceding anomalous block corresponds to multiple subsequent anomalous blocks, the subsequent anomalous blocks with more corresponding pixels in the contour continuity relationship are retained first. When the corresponding pixel counts are the same, the subsequent anomalous blocks with shorter center distances from the center of the preceding anomalous block are retained. After the above uniqueness processing is completed, the sequence of anomalous blocks formed by continuous connection is written as a candidate development chain, and all candidate development chains are written into the candidate development chain set. Through this processing, the anomalous blocks in each candidate development chain are arranged continuously according to the acquisition time, and the continuity relationship between adjacent anomalous blocks has been verified. They can then be directly used to locate missing level segments and generate supplementary observation instructions. After the above processing, the abnormal block record set is transformed into a candidate development chain set with clear time order, regional relationship and succession relationship. Abnormal continuation within the same regional code and abnormal expansion between adjacent regional codes are both included in the same processing chain, and corresponding contour succession calculation methods are used respectively to avoid the mixing of judgment conditions caused by different regional relationships. In addition, the extraction caliber of the previous abnormal block and the subsequent abnormal block, the generation method of the connection direction, the generation method of the change direction, and the uniqueness rules of the candidate development chain are all determined in this step. In practical applications: When a pest anomalous block exists in a region code at the first acquisition time, and a pest anomalous block with an expanded range exists in the same region code at the second acquisition time, the system first counts the number of overlapping pixels in the outlines of the two anomalous blocks. If the number of overlapping pixels is greater than zero, the system calculates the direction of the line connecting the two center positions and, combined with the change in the exposure level, determines that they belong to the same continuous development process, and then writes them into the same candidate development chain. When the anomalous block at the first acquisition time is located in region code A0105, and the anomalous block at the second acquisition time is located in region code A0106, which is adjacent to the right side of A0105, the system reads the shared boundary between A0105 and A0106 and counts the number of overlapping pixels projected onto the shared boundary of the two anomalous block outlines. If the corresponding number of overlapping pixels is greater than zero, the system, combined with the direction of the line connecting the center positions and the direction of the change in the exposure level, determines that they meet the continuation conditions, and then continues to connect them into the same candidate development chain. If a later anomalous block meets the continuation conditions with two earlier anomalous blocks, one set of connection relationships is retained according to the order of acquisition time and corresponding pixel count, ensuring that the candidate development chain is unique.

[0021] S4. For each candidate development chain in the candidate development chain set, locate the missing hierarchical segment between the previous anomaly block and the subsequent anomaly block, and generate supplementary observation instructions based on the center position of the subsequent anomaly block, the missing observation height of the previous anomaly block, the missing observation angle of the previous anomaly block, and the most recent executable time period after the subsequent anomaly block. In this process, the exposed hierarchical positions between the preceding and following anomalies that have not been filled in by existing observations are first identified within the candidate development chain. Then, based on the existing height and viewpoint markers in the observation record sequence, observation conditions for filling in the missing hierarchical segments are generated in reverse. Subsequently, the target position is corrected by combining the regional relationship between the preceding and following anomalies, and finally, supplementary observation instructions are generated. The purpose of this process is to avoid directly adding supplementary images to the entire candidate development chain. Instead, the missing height, missing viewpoint, and target position corresponding to the missing hierarchical segments are determined item by item, and these fields are organized into executable supplementary observation instructions, so that the supplementary observations always revolve around the hierarchical segments in the candidate development chain that have not yet been closed. The implementation process includes the following steps: For each candidate development chain in the candidate development chain set, first, all abnormal blocks in the candidate development chain are read in the order of acquisition time. Then, two adjacent abnormal blocks are grouped together, and the abnormal block with the earlier acquisition time is taken as the previous abnormal block, and the abnormal block with the later acquisition time is taken as the next abnormal block. Then, the exposure level of the next two abnormal blocks is subtracted from the exposure level of the previous abnormal block to obtain the level difference between the previous and next abnormal blocks in this group. When the level difference is equal to one, it means that there are no incomplete exposure levels between the previous and next abnormal blocks in this group, so a complete marker is written to the previous and next abnormal blocks. When the level difference is equal to one, it means that there are no incomplete exposure levels between the previous and next abnormal blocks in this group, so a complete marker is written to the previous and next abnormal blocks. When the value is greater than one, it indicates that there are still unseen levels between the preceding and following anomalous blocks in the group. Therefore, a missing level segment is determined between the preceding and following anomalous blocks. When there are multiple preceding and following anomalous blocks with a level difference greater than one in a candidate development chain, multiple missing level segments are determined respectively, and each missing level segment corresponds to its respective preceding and following anomalous blocks. Here, the complete marker is only used to indicate that the preceding and following anomalous blocks in the group will no longer enter the supplementary observation generation process, while the missing level segment is used to indicate that the preceding and following anomalous blocks in the group still need to continue to supplement the observations. For each missing hierarchical segment, first read all height markers in the observation record sequence and arrange them in ascending order of value to form a height sequence; then read all viewpoint markers in the observation record sequence and arrange them in ascending order of number to form a viewpoint sequence; subsequently, find the height marker positions of the preceding and following anomalies in the height sequence; when the two positions are not adjacent, extract all height markers between the two positions to form the missing height sequence for that missing hierarchical segment; when the two positions are adjacent, the missing height sequence for that missing hierarchical segment is empty; then, in the same way, find the viewpoint marker positions of the preceding and following anomalies in the viewpoint sequence; when the two positions are not adjacent, extract all viewpoint markers between the two positions to form the missing viewpoint sequence for that missing hierarchical segment; when the two positions are adjacent, the missing viewpoint sequence for that missing hierarchical segment is empty; if the height marker position of the preceding anomaly is greater than that of the following anomaly... For the height marker position of the block, the smaller of the two positions is taken as the starting position and the larger position as the ending position, and then the middle height marker is extracted. The viewpoint marker is processed in the same way to ensure that the order of the preceding and following anomaly blocks in the height and viewpoint sequences does not affect the generation of the missing sequence. After the above processing is completed, the center position of the following anomaly block is determined as the initial target position of the missing level segment. After this processing, each missing level segment corresponds to a clear missing height sequence, missing viewpoint sequence and initial target position, providing direct input for the generation of subsequent observation combinations. For example, in multi-height monitoring in mountainous forest areas, the height markers in the observation record sequence are one, two, three and four in sequence. If the height marker of the preceding anomaly block is one and the height marker of the following anomaly block is four, then the missing height sequence is two and three; if the height marker of the preceding anomaly block is two and the height marker of the following anomaly block is three, then the missing height sequence is empty. For each missing level segment, height candidates are first determined based on the missing height sequence. When the missing height sequence is not empty, height markers are written into the height candidates in the order of the missing height sequence. When the missing height sequence is empty, the height markers of the previous anomaly block are written into the height candidates. Then, viewpoint candidates are determined based on the missing view sequence. When the missing view sequence is not empty, view markers are written into the viewpoint candidates in the order of the missing view sequence. When the missing view sequence is empty, the viewpoint markers of the previous anomaly block are written into the viewpoint candidates. The reason for this setting is that when a missing level segment does have an intermediate gap in the height or view direction, supplementary observations should be organized around the intermediate gap first. When there is no intermediate gap in a certain direction, the observation conditions already corresponding to the previous anomaly block are used to continue to fill the gap in the other direction. After writing, the missing level segment corresponds to one height candidate and one viewpoint candidate, or multiple height candidates and multiple viewpoint candidates. Subsequent item combinations can obtain all the observation combinations for the missing level segment. When combining height and viewpoint candidates item by item, first fix one height candidate, then match all viewpoint candidates sequentially to generate all observation combinations under that height candidate; then read the next height candidate and repeat the above process until all height and viewpoint candidates are combined; each observation combination includes at least the target position, one height marker, and one viewpoint marker; after the observation combination is formed, perform region determination on each observation combination; if the region code of the preceding anomaly block is the same as the region code of the following anomaly block, it means that the missing level segment is located within the same region code, the target position of the missing level segment remains unchanged, and the center position of the following anomaly block is retained as the target position ... is retained within the same region code, the target position of the following anomaly block is retained within the same region code, the target position of the following anomaly block is retained within the same region code, the target position of the following anomaly block is retained within the same region code, the target position of the following anomaly block is retained within the same region code, the target position of the following anomaly block is If the codes are adjacent, the shared boundary between the two is first read from the region mapping table, then the line connecting the center position of the previous anomalous block and the center position of the subsequent anomalous block is calculated, and the intersection point of the line and the shared boundary is obtained and written to the target position. If the region code of the previous anomalous block and the region code of the subsequent anomalous block are neither the same nor adjacent, it means that there is a lack of necessary regional continuity between the previous anomalous block and the subsequent anomalous block corresponding to the current missing level segment. Therefore, a backtracking mark is written to the missing level segment. After this processing, the supplementary observations within the same region code are still carried out around the center position of the subsequent anomalous block, while the supplementary observations across adjacent region codes are changed to be carried out around the regional transition position, so that the position of the supplementary observations is consistent with the spatial continuity of the candidate development chain. For each observation combination formed by a missing hierarchical segment, first determine whether the missing hierarchical segment has been marked with a rollback flag. If the missing hierarchical segment has not been marked with a rollback flag, then read the executable time period after the subsequent anomaly block. The executable time period is taken from the time period record in the observation equipment task table, and from all time period records whose start time is later than the acquisition time of the subsequent anomaly block, sort them by start time from earliest to latest and take the first time period as the first executable time period. Then, write the first executable time period together with the target position, height marker, and viewpoint marker in the observation combination into a supplementary observation instruction, and write the supplementary observation instruction into the corresponding candidate development chain. If the missing hierarchical segment has been marked with a rollback flag, then read the anomaly block in the candidate development chain where the missing hierarchical segment is located, which is located before the preceding anomaly block, and search in reverse order from nearest to farthest acquisition time. After finding the first anomalous block with a difference in exposure level greater than one between it and the subsequent anomalous block, it is replaced with the preceding anomalous block corresponding to the missing level segment. Then, based on the replaced preceding anomalous block and the original subsequent anomalous block, the process of determining the missing height sequence, missing view sequence, and target position is re-executed, and supplementary observation instructions are generated accordingly. If no anomalous block that meets the above conditions is found, the generation of supplementary observations for the missing level segment is stopped, and the candidate development chain in which the missing level segment is located is written into the misjudgment candidate state. Here, the role of the backtracking process is that when there is no effective regional connection between the current preceding anomalous block and the subsequent anomalous block, the candidate development chain is traced back, and an earlier preceding anomalous block is selected to form a new completion object with the current subsequent anomalous block, thereby avoiding the continued generation of invalid supplementary observation instructions when the regional relationship has been broken. After the above processing, each missing level segment can obtain its corresponding missing height sequence, missing view sequence, target position and supplementary observation instructions; the spatial relationship within the same region code and between adjacent region codes is determined in this step, and the writing conditions of the backtracking marker, the replacement order of the previous anomaly block and the generation method of the supplementary observation instructions are also completed in this step. In practical applications: If a candidate development chain contains a first anomalous block with a disclosure level of one and a second anomalous block with a disclosure level of four, and the first anomalous block has a height marker of one and a view marker of one, while the second anomalous block has a height marker of four and a view marker of three, the system first determines that there is a missing level segment between the two. Then, it extracts height markers two and three from the height sequence to form a missing height sequence, and extracts view marker two from the view sequence to form a missing view sequence. The center position of the second anomalous block is used as the initial target position. If the first and second anomalous blocks are located within the same region code, the target position remains unchanged, and the height markers are adjusted accordingly. The system combines the second and third observation markers with the second observation marker to form multiple observation combinations. Then, it selects the first executable time period from the observation equipment task table after the acquisition time of the later anomaly block and generates corresponding supplementary observation instructions. If the previous and later anomaly blocks are located within adjacent region codes, the system uses the intersection of the calculation center line and the shared boundary as the target location and generates supplementary observation instructions. If the previous and later anomaly blocks are neither located in the same region code nor in adjacent region codes, the system writes a backtracking marker to the missing level segment and searches forward along the candidate development chain for an earlier anomaly block to replace the previous anomaly block, and then regenerates supplementary observation instructions.

[0022] S5. Obtain supplementary images and extract supplementary anomaly blocks according to the supplementary observation instructions. Insert the supplementary anomaly blocks into the missing level segment of the corresponding candidate development chain for re-verification. When a continuous hierarchical advancement relationship is formed, output the disaster confirmation result and early warning result. When a continuous hierarchical advancement relationship is not formed, output the misjudgment resolution result. In this process, supplementary images are first obtained according to supplementary observation instructions, and supplementary anomaly blocks are extracted from the supplementary images. Then, the supplementary anomaly blocks are inserted into the missing hierarchical segments of the corresponding candidate development chain. The hierarchical connection, contour connection, and positional connection between the preceding anomaly block, the supplementary anomaly block, and the following anomaly block are checked in sequence. Finally, the disaster confirmation result and early warning result, or the misjudgment resolution result, are output based on the verification results. The purpose of this process is not only to determine whether the supplementary observation has obtained new anomaly blocks, but also to determine whether the supplementary anomaly block can form a continuous progression relationship with the preceding and following anomaly blocks at both ends of the missing hierarchical segment, thereby determining whether the anomaly changes corresponding to the candidate development chain belong to the continuous development process of the same disaster. The implementation process includes the following steps: When acquiring supplementary images according to supplementary observation instructions, the target location, height marker, viewpoint marker, and executable time period in the supplementary observation instructions are read first. Then, within the executable time period, supplementary images corresponding to the target location and satisfying the height marker and viewpoint marker are retrieved. When multiple supplementary images satisfying the conditions exist within the same executable time period, the supplementary image with the earliest acquisition time is taken as the current supplementary image. When the same supplementary image contains multiple region codes, only the image range corresponding to the region code in which the target location falls is retained as the current processing range. Subsequently, the current supplementary image is input into the image recognition network to extract abnormal pixel regions, and abnormal pixel regions with adjacent boundaries are merged into supplementary abnormal blocks, where the boundary adjacency uses the same eight-neighbor phase as S2. Following the rules, for each supplementary anomaly block, extract its outer edge pixels to form a contour. Use the average horizontal and vertical coordinates of all anomaly pixels within the contour-enclosed area to form the center position, and read the anomaly category and exposure level corresponding to the supplementary anomaly block. Then, generate observation record markers in a fixed order of region code, acquisition time, height marker, viewpoint marker, and source marker, and write the observation record markers into the corresponding supplementary anomaly block. To ensure that the supplementary anomaly blocks correspond one-to-one with the corresponding missing level segment, when multiple supplementary anomaly blocks are extracted in the same supplementary image, take the supplementary anomaly block with the shortest distance between the center position and the target position as the target supplementary anomaly block. The remaining supplementary anomaly blocks are retained in the recognition results of the supplementary image but are not inserted into the current missing level segment. When inserting the supplementary anomaly block into the missing level segment of the corresponding candidate development chain, first read the preceding and following anomaly blocks corresponding to the missing level segment, and then arrange the preceding, supplementary, and following anomaly blocks into three sequences according to the acquisition time order; then calculate the difference in the exposure level between the preceding and supplementary anomaly blocks, and the difference in the exposure level between the supplementary and following anomaly blocks, where the former is the exposure level of the supplementary anomaly block minus the exposure level of the preceding anomaly block, and the latter is the exposure level of the following anomaly block minus the exposure level of the supplementary anomaly block; then map the outline bounding areas of the preceding and following anomaly blocks to the same coordinate system, and count the number of pixels that fall within the outline bounding areas of both blocks pixel by pixel to obtain the number of overlapping pixels between the outlines of the preceding and following anomaly blocks; count the number of overlapping pixels between the outlines of the supplementary and following anomaly blocks in the same way. Subsequently, the first connecting direction is generated by the center position of the previous anomaly block and the center position of the supplementary anomaly block, and the second connecting direction is generated by the center position of the supplementary anomaly block and the center position of the subsequent anomaly block. The method of generating the connecting direction is consistent with S3, that is, the direction of upward, downward, left, right, upper left, upper right, lower left, or lower right is determined according to the difference in the horizontal coordinate and the difference in the vertical coordinate, respectively. If the same missing layer segment corresponds to multiple supplementary observation instructions, the supplementary anomaly blocks are inserted sequentially from the earliest to the latest time of the supplementary image acquisition. After each supplementary anomaly block is inserted, the exposure layer difference between the current previous anomaly block and the current subsequent anomaly block is recalculated. When the remaining exposure layer difference is still greater than one, the next supplementary observation instruction is executed until the current missing layer segment is completely divided into multiple adjacent connecting segments with an exposure layer difference of one, or all supplementary observation instructions are executed. When outputting results based on the calculation results between the preceding anomaly block, the supplementary anomaly block, and the following anomaly block, the process first checks whether the difference in the exposure level between the preceding and supplementary anomaly blocks is equal to one, and then checks whether the difference in the exposure level between the supplementary and following anomaly blocks is equal to one. If both exposure level differences are equal to one, the process then checks whether the number of overlapping pixels between the outlines of the preceding and supplementary anomaly blocks is greater than zero, and whether the number of overlapping pixels between the outlines of the supplementary and following anomaly blocks is greater than zero. If all of the above conditions are met, the process then checks whether the direction of the first connection line is consistent with the direction of the second connection line. When all of the above conditions are met, it is determined that the supplementary anomaly block forms a continuous hierarchical progression relationship with the preceding and following anomaly blocks, and the candidate development chain corresponding to the missing hierarchical segment is written into the disaster confirmation result. Simultaneously, based on this candidate development... The anomaly category, center location, and latest acquisition time in the chain generate early warning results. If any of the above conditions are not met, it is determined that the supplementary anomaly block has not formed a continuous hierarchical progression relationship, and the candidate development chain corresponding to the missing hierarchical segment is written into the misjudgment resolution result. If multiple supplementary anomaly blocks have been inserted into the same missing hierarchical segment, the disaster confirmation result and early warning result are output only when all adjacent connecting segments meet the above conditions, and the misjudgment resolution result is output when any connecting segment does not meet the above conditions. Here, the disaster confirmation result includes at least the candidate development chain identifier, anomaly category, target location, and confirmation time; the early warning result includes at least the monitoring area, anomaly category, target location, confirmation time, and early warning status; and the misjudgment resolution result includes at least the candidate development chain identifier, target location, and resolution time. After the above processing, the supplementary images obtained from the supplementary observation instructions are no longer used only for re-identifying anomalies, but are used to verify whether a complete and continuous progressive relationship is formed between the preceding and following anomaly blocks at both ends of the missing hierarchical segment; the selection rules, insertion order, recalculation method, and output conditions of disaster confirmation results and misjudgment resolution results of the supplementary anomaly blocks are all determined in this step, so that there is a clear basis for judging whether the candidate development chain is valid; In practical applications: If the exposure level of the preceding anomaly block corresponding to a certain missing layer segment is one and the exposure level of the following anomaly block is three, the system obtains a supplementary image with a height marker of two and a viewing angle marker of two according to the supplementary observation command, and extracts multiple supplementary anomaly blocks from the supplementary image. The supplementary anomaly block with the shortest distance between its center position and the target position is selected as the current supplementary anomaly block. Then, this supplementary anomaly block is inserted between the preceding and following anomaly blocks. The exposure level difference between the preceding and following anomaly blocks is calculated to be one, and the exposure level difference between the following and following anomaly blocks is also one. Furthermore, the number of overlapping pixels between the outlines of the preceding and following anomaly blocks is greater than zero, the number of overlapping pixels between the outlines of the following and following anomaly blocks is greater than zero, and the two connecting lines are in the same direction. In this case, the system outputs the disaster confirmation result and the warning result. If the exposure level difference between the following and following anomaly blocks is not equal to one, or the number of overlapping pixels in any set of outlines is equal to zero, or the two connecting lines are in the same direction, the system writes the candidate development chain into the misjudgment resolution result.

[0023] Furthermore, it also includes a three-dimensional monitoring and early warning system for forestry disasters, the system comprising: The observation and processing module is used to collect multi-height and multi-view forest area images of the same monitoring area within the same monitoring cycle, and write the area code, acquisition time, height mark, view mark and source mark for each forest area image, and output the observation record sequence. The anomaly extraction module divides the forest area image into image segments with corresponding region codes for each observation record in the observation record sequence, inputs the image segments into the image recognition network to extract anomaly blocks, writes the contour, center position, anomaly category, exposure level and observation record mark for each anomaly block, and outputs an anomaly block record set. The link construction module connects consecutive anomaly blocks within the same region code and between adjacent region codes in chronological order according to the center position, outline, exposure level and collection time of the anomaly block record set, and outputs a candidate development chain set. The observation generation module locates the missing hierarchical segment between the previous and subsequent anomalous blocks for each candidate development chain in the candidate development chain set, and generates supplementary observation instructions based on the center position of the subsequent anomalous block, the missing observation height of the previous anomalous block, the missing observation angle of the previous anomalous block, and the most recent executable time period after the subsequent anomalous block. The result verification module acquires supplementary images and extracts supplementary anomaly blocks according to supplementary observation instructions. These anomaly blocks are then inserted into the missing level segments of the corresponding candidate development chain for re-verification. If a continuous hierarchical progression relationship is formed, the module outputs disaster confirmation and early warning results; otherwise, it outputs misjudgment resolution results. Furthermore, it also includes a three-dimensional monitoring and early warning terminal for forestry disasters, the terminal comprising: The processor, and the memory connected to the processor; Memory is used to store computer programs; The processor is used to call and execute computer programs in memory to implement a three-dimensional monitoring and early warning method for forestry disasters.

[0024] Furthermore, it also includes a three-dimensional monitoring and early warning medium for forestry disasters, the medium comprising: The monitoring and early warning medium stores a computer program. When the computer program is executed by a processor, it implements the various steps of a three-dimensional monitoring and early warning method for forestry disasters.

[0025] Working Principle: This scheme first uniformly writes forest area images from different observation heights, viewing angles, and sources within the same monitoring area into a region code, acquisition time, height marker, viewing angle marker, and source marker, forming an observation record sequence arranged in chronological order. Then, image segments are extracted from the forest area images according to the region code, and machine learning is used to extract anomaly blocks, resulting in an anomaly block record set with contours, center locations, anomaly categories, and exposure levels. Subsequently, preceding and following anomaly blocks that satisfy contour continuity, center continuity, and hierarchical progression relationships are connected between adjacent acquisition times to form a candidate development chain. If the preceding and following anomaly blocks in the candidate development chain... If missing hierarchical segments still exist, supplementary observation instructions are generated based on the missing height sequence, missing viewpoint sequence, and target location. Supplementary images are acquired and supplementary anomaly blocks are extracted. These supplementary anomaly blocks are then inserted into the missing hierarchical segments for re-verification. When the supplemented anomaly blocks can form a continuous hierarchical progression relationship, disaster confirmation and early warning results are output. If a continuous hierarchical progression relationship cannot be formed, misjudgment resolution results are output. Overall, this scheme does not only consider whether anomalies occur at a certain moment, but makes judgments along the continuous exposure process of the same disaster at different times, heights, and viewpoints, thereby organizing scattered anomaly image results into a verifiable continuous development chain. For example, in mountainous forest areas, satellite imagery first detects abnormal changes in the upper canopy at a certain region code. Subsequently, drone imagery captures abnormalities at a lower altitude near the same region in adjacent time periods. Ground equipment then captures abnormal images closer to the forest floor in later time periods. The system first assigns these images to the corresponding region code, then uses machine learning to extract anomalous blocks and determines whether these blocks belong to the same continuous development process in terms of outline location, center location, and exposure level. If the system finds that two anomalous blocks are missing an intermediate level, it will not draw a conclusion directly but will automatically generate a supplementary layer. The system can issue supplementary observation commands, such as requiring images to be taken at a specific target location, a missing height, and a missing viewpoint. If the newly extracted anomaly blocks can connect the two anomaly blocks after the supplementary images are taken, it indicates that the anomaly in the area is indeed developing continuously, and the system outputs disaster confirmation and warning results accordingly. If the images still cannot be connected after the supplementary images are taken, the candidate development chain is eliminated to avoid mistaking unrelated local anomalies for the same disaster. In this way, in actual forestry disaster monitoring scenarios, the system can not only use existing images to discover anomalies, but also continue to supplement evidence around the missing locations, thus making the warning results more coherent.

[0026] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A three-dimensional monitoring and early warning method for forestry disasters, characterized in that, include: S1. Collect multi-height, multi-view forest area images of the same monitoring area within the same monitoring period, and write the area code, acquisition time, height mark, view mark and source mark for each forest area image, and output the observation record sequence; S2. For each observation record in the observation record sequence, divide the forest area image into image segments with corresponding region codes, input the image segments into the image recognition network to extract anomaly blocks, write the contour, center position, anomaly category, exposure level and observation record mark for each anomaly block, and output the anomaly block record set. S3. Based on the center position, outline, exposure level and acquisition time of the abnormal block record set, connect the preceding and following abnormal blocks in the same area code and between adjacent area codes in chronological order, and output the candidate development chain set. S4. For each candidate development chain in the candidate development chain set, locate the missing hierarchical segment between the previous anomaly block and the subsequent anomaly block, and generate supplementary observation instructions based on the center position of the subsequent anomaly block, the missing observation height of the previous anomaly block, the missing observation angle of the previous anomaly block, and the most recent executable time period after the subsequent anomaly block. S5. Obtain supplementary images and extract supplementary anomaly blocks according to the supplementary observation instructions. Insert the supplementary anomaly blocks into the missing level segment of the corresponding candidate development chain for re-verification. When a continuous hierarchical advancement relationship is formed, output the disaster confirmation result and early warning result. When a continuous hierarchical advancement relationship is not formed, output the misjudgment resolution result.

2. The method for three-dimensional monitoring and early warning of forestry disasters according to claim 1, characterized in that: S1 includes: S1-1. Using the coordinates of the upper left corner of the monitoring area boundary as the starting coordinates, divide the area codes in an ascending row number and column number manner, and establish a mapping between each area code and its corresponding spatial location to output the set of area codes. S1-2. Extract the coverage area of ​​each forest area image within the same monitoring period, compare the coverage area with the set of area codes one by one, take the area code contained in the covered area and write it into the forest area image, and write the acquisition time, height mark, view mark and source mark at the same time, and output the image record set. S1-3. Based on the region code and acquisition time in the image record set, arrange the image records with the same region code in ascending order of acquisition time, and output the observation record sequence according to the arrangement result.

3. The method for three-dimensional monitoring and early warning of forestry disasters according to claim 2, characterized in that: S2 includes: S2-1. For each observation record in the observation record sequence, extract the corresponding image fragment from the forest area image according to the area code, and output the image fragment set; S2-2. Input each image segment in the image segment set into the image recognition network, extract abnormal pixel regions, and merge the abnormal pixel regions with the boundary into abnormal blocks. Use the outer edge pixels of each abnormal block to form a contour, and use the geometric center of the region surrounded by the contour to form the center position. At the same time, read the corresponding abnormal category and exposure level, and output the abnormal block information set. S2-3. Write the information of each abnormal block in the abnormal block information set and the corresponding observation record's area code, acquisition time, height marker, viewpoint marker, and source marker in a fixed order to generate observation record markers. Then write the observation record markers into the corresponding abnormal blocks and output the abnormal block record set.

4. The method for three-dimensional monitoring and early warning of forestry disasters according to claim 3, characterized in that: S3 includes: S3-1. Arrange the abnormal block record set in ascending order according to the acquisition time, and extract the previous and subsequent abnormal blocks within the same region code and between adjacent region codes between two adjacent acquisition times, and output the set of previous and subsequent abnormal blocks. S3-2. For each pair of anomaly blocks in the set of anomaly blocks, when the anomaly block and the anomaly block belong to the same region code, count the number of overlapping pixels in the outline-enclosed area of ​​the anomaly block and the outline-enclosed area of ​​the anomaly block. When the anomaly block and the anomaly block belong to adjacent region codes, count the number of overlapping pixels in the corresponding projections of the anomaly block outline and the anomaly block outline on the shared boundary of the region code. Calculate the direction of the line connecting the center position of the anomaly block and the center position of the anomaly block, as well as the direction of change from the exposure level of the anomaly block to the exposure level of the anomaly block. When the number of overlapping pixels is greater than zero, the number of overlapping pixels in the corresponding projections is greater than zero, and the direction of the line and the direction of change are consistent, determine that the anomaly block and the anomaly block satisfy the outline continuity relationship, the center continuity relationship, and the hierarchical advancement relationship, and output the set of anomaly blocks. S3-3. Connect the preceding and following abnormal blocks in the continuation abnormal block pair set end to end according to the acquisition time sequence, and write the sequence of abnormal blocks formed by continuous connection as a candidate development chain, and output the candidate development chain set.

5. The method for three-dimensional monitoring and early warning of forestry disasters according to claim 4, characterized in that: S4 includes: S4-1. For each candidate development chain in the candidate development chain set, extract the previous and subsequent anomalous blocks in the order of collection time, calculate the difference between the exposure level of the subsequent anomalous block and the exposure level of the previous anomalous block, write complete tags to the previous and subsequent anomalous blocks when the difference is equal to one, and determine the missing level segment between the previous and subsequent anomalous blocks when the difference is greater than one. S4-2. For each missing level segment, read all height markers in the observation record sequence and arrange them in ascending order of value to form a height sequence. Read all viewpoint markers in the observation record sequence and arrange them in numerical order to form a viewpoint sequence. Then, according to the positions of the height markers of the preceding and following anomalies in the height sequence, extract the middle height marker to form a missing height sequence. According to the positions of the viewpoint markers of the preceding and following anomalies in the viewpoint sequence, extract the middle viewpoint marker to form a missing viewpoint sequence. Finally, determine the center position of the following anomaly as the target position of the missing level segment.

6. The method for three-dimensional monitoring and early warning of forestry disasters according to claim 5, characterized in that: S4 further includes: S4-3. For each missing level segment, first determine the height candidate based on the missing height sequence. If the missing height sequence is not empty, write each height marker in the missing height sequence into the height candidate in sequence. If the missing height sequence is empty, write the height marker of the previous abnormal block into the height candidate. Then determine the view candidate based on the missing view sequence. If the missing view sequence is not empty, write each view marker in the missing view sequence into the view candidate in sequence. If the missing view sequence is empty, write the view marker of the previous abnormal block into the view candidate. S4-4. Combine the height candidates and viewpoint candidates one by one to form an observation combination, and perform region determination on each observation combination. When the region code of the previous anomaly block is the same as the region code of the subsequent anomaly block, keep the target position unchanged. When the region code of the previous anomaly block is adjacent to the region code of the subsequent anomaly block, write the intersection of the line connecting the center position of the previous anomaly block and the center position of the subsequent anomaly block with the boundary shared by the region codes into the target position. When the region code of the previous anomaly block is neither the same nor adjacent to the region code of the subsequent anomaly block, write a backtracking mark into the missing level segment. S4-5. For each observation combination formed by each missing level segment, if no backoff flag is written in the missing level segment, read the executable time period after the next anomaly block and take the first executable time period. Combine the target position, the height flag and the view flag in the observation combination to generate supplementary observation instructions and write them into the corresponding candidate development chain. If a backoff flag is written in the missing level segment, replace the previous anomaly block corresponding to the missing level segment with the anomaly block in the candidate development chain that is located before the previous anomaly block. Based on the replaced previous anomaly block and the next anomaly block, redetermine the missing height sequence, the missing view sequence, the target position and the supplementary observation instructions.

7. The method for three-dimensional monitoring and early warning of forestry disasters according to claim 6, characterized in that: S5 includes: S5-1. Obtain supplementary images according to supplementary observation instructions, and input the supplementary images into the image recognition network to extract supplementary anomaly blocks. Write the outline, center position, anomaly category, exposure level and observation record mark for the supplementary anomaly blocks. S5-2. Insert the supplementary abnormal block into the missing level segment of the corresponding candidate development chain. Arrange the previous abnormal block, the supplementary abnormal block, and the subsequent abnormal block in the order of acquisition time. Calculate the difference in the exposure level between the previous abnormal block and the supplementary abnormal block, and the difference in the exposure level between the supplementary abnormal block and the subsequent abnormal block. Count the number of overlapping pixels between the outlines of the previous abnormal block and the supplementary abnormal block, and the number of overlapping pixels between the outlines of the supplementary abnormal block and the subsequent abnormal block. Calculate the direction of the line connecting the center position of the previous abnormal block and the center position of the supplementary abnormal block, and the direction of the line connecting the center position of the supplementary abnormal block and the center position of the subsequent abnormal block. S5-3. When the difference in the exposure level between the preceding anomaly block and the supplementary anomaly block is equal to 1, the difference in the exposure level between the supplementary anomaly block and the following anomaly block is equal to 1, the number of overlapping pixels between the outlines of the preceding anomaly block and the supplementary anomaly block is greater than zero, the number of overlapping pixels between the outlines of the supplementary anomaly block and the following anomaly block is greater than zero, and the direction of the line connecting the center positions of the preceding anomaly block and the supplementary anomaly block is consistent with the direction of the line connecting the center positions of the supplementary anomaly block and the following anomaly block, output the disaster confirmation result and the early warning result; if any one of these conditions is not met, output the misjudgment resolution result.

8. A three-dimensional monitoring and early warning system for forestry disasters, characterized in that, include: The observation and processing module is used to collect multi-height and multi-view forest area images of the same monitoring area within the same monitoring cycle, and write the area code, acquisition time, height mark, view mark and source mark for each forest area image, and output the observation record sequence. The anomaly extraction module divides the forest area image into image segments with corresponding region codes for each observation record in the observation record sequence, inputs the image segments into the image recognition network to extract anomaly blocks, writes the contour, center position, anomaly category, exposure level and observation record mark for each anomaly block, and outputs an anomaly block record set. The link construction module connects consecutive anomaly blocks within the same region code and between adjacent region codes in chronological order according to the center position, outline, exposure level and collection time of the anomaly block record set, and outputs a candidate development chain set. The observation generation module locates the missing hierarchical segment between the previous and subsequent anomalous blocks for each candidate development chain in the candidate development chain set, and generates supplementary observation instructions based on the center position of the subsequent anomalous block, the missing observation height of the previous anomalous block, the missing observation angle of the previous anomalous block, and the most recent executable time period after the subsequent anomalous block. The result verification module acquires supplementary images and extracts supplementary anomaly blocks according to the supplementary observation instructions. It then inserts the supplementary anomaly blocks into the missing level segment of the corresponding candidate development chain for re-verification. When a continuous hierarchical progression relationship is formed, it outputs disaster confirmation results and early warning results. When a continuous hierarchical progression relationship is not formed, it outputs misjudgment resolution results.

9. A three-dimensional monitoring and early warning terminal for forestry disasters, characterized in that, include: A processor, and a memory connected to the processor; The memory is used to store computer programs; The processor is used to invoke and execute the computer program in the memory to perform the method as described in any one of claims 1-7.

10. A three-dimensional monitoring and early warning medium for forestry disasters, characterized in that, include: The monitoring and early warning medium stores a computer program, which, when executed by a processor, implements the various steps of the monitoring and early warning method as described in any one of claims 1-7.

Images