Wildfire smoke plume and its boundary vector automatic identification method, system, storage medium and electronic equipment

Abstract

The application discloses a kind of wild fire smoke cluster and its boundary vector automatic identification method, system, storage medium and electronic equipment, belong to satellite remote sensing and disaster monitoring technical field.Method includes: obtaining satellite active fire point, fire event and UVAI data;Fire source area is positioned based on fire event geographic location and initial plume pixel is marked;The spatially continuous smoke cluster connected domain is identified by iterative region growing algorithm;According to the specific distribution form of connected domain pixel in satellite pixel matrix, the accurate closed boundary vector is automatically generated using matching geometric algorithm;Finally, using multiple criteria, the spatial isolated independent smoke cluster is screened out.The application avoids the defects that traditional image method is easily disturbed by environment from physical principle, realizes fully automated processing without training data, and can output vector boundary that can be directly used for diffusion and emission evaluation, significantly improves the accuracy, automation degree and application value of wild fire smoke cluster monitoring.

Description

Technical Field

[0001] This invention relates to the field of satellite remote sensing and disaster monitoring technology. Specifically, it relates to a method, system, storage medium, and electronic device for automatically identifying independent smoke plumes of wildfires and extracting their precise boundary vectors by utilizing the satellite ultraviolet aerosol absorption index (UVAI) in the ultraviolet band, combined with information on active fire points and fire events. Background Technology

[0002] Wildfires, as a significant natural source of aerosols, greenhouse gases, and pollutants in the Earth's atmosphere, pose a serious threat to air quality and public health in downwind areas due to the diffusion and transport of their smoke plumes. Therefore, accurate identification, spatial morphology definition, and dynamic tracking of individual wildfire smoke plumes are crucial prerequisites for pollution early warning, exposure risk assessment, and emission quantification. However, current mainstream wildfire smoke identification technologies suffer from several key bottlenecks, hindering their application in practical operational monitoring and sophisticated scientific research.

[0003] Existing technologies largely rely on satellite or video imagery based on the visible light band and employ traditional image processing models (such as the RGB color model) or artificial intelligence methods for smoke segmentation. These methods are firstly significantly constrained by environmental conditions; their recognition performance is easily affected by light intensity, shooting angle, cloud cover, and background interference, leading to decreased reliability under complex weather conditions. More fundamentally, visible light features struggle to effectively distinguish smoke with similar spectral and textural characteristics from thin clouds, dust, or urban smog, resulting in a high false positive rate. Secondly, while methods based on deep learning and other artificial intelligence models perform well in certain scenarios, they depend on large amounts of high-quality, labeled smoke image data for training. Obtaining large-scale, diverse wildfire smoke labeling data is itself costly and extremely challenging; furthermore, model training and inference processes typically consume significant computational resources.

[0004] Finally, and most importantly, existing methods generally focus on pixel-level identification of "smoke regions," making it difficult to effectively correlate and isolate complete, independent smoke plumes originating from specific fire events in terms of time and space. They cannot automatically distinguish smoke plumes from different fire sources or smoke plumes from the same fire source at different stages of diffusion, and they lack the ability to automatically extract the precise two-dimensional spatial boundaries of smoke plumes. This makes it difficult to support quantitative analysis of the diffusion path, morphological evolution, and corresponding emissions of individual smoke plumes.

[0005] Therefore, the industry urgently needs a solution that is more reliable in principle, highly automated, and capable of outputting the precise vector spatial shape of independent smoke plumes in order to overcome the current technological limitations. Summary of the Invention

[0006] The present invention aims to solve the technical problems of existing wildfire smoke plume identification methods, such as weak anti-interference ability, reliance on training data, and inability to automatically output accurate vector boundaries of independent smoke plumes associated with specific fire events.

[0007] To address the aforementioned problems, this invention provides a method for automatic identification of wildfire smoke plumes and their boundary vectors, comprising:

[0008] S1. Acquire satellite active fire point data, fire event information, and orbit-by-orbit absorptive aerosol index (UVAI) data for the target area and target time period;

[0009] S2. For each wildfire event, calculate the center latitude and longitude of the burned area based on the fire event information, and delineate the fire source area based on the center latitude and longitude; traverse the orbital UVAI data from the start of the fire to its extinguishment, and mark the UVAI pixels in the fire source area as plume pixels on the orbit that detects active fire points within the combustion boundary of the fire event on the day the orbit passes over.

[0010] S3. For each orbit, starting with the plume pixel, search for adjacent pixels based on the row and column numbers of the pixel in the satellite orbit pixel matrix, according to the adjacency relationship of row number ±1 and column number ±1, and iteratively merge adjacent pixels that meet the threshold conditions into plume pixels according to the preset UVAI minimum threshold, until no new adjacent pixels that meet the threshold conditions are added, thus obtaining the connected domain of the wildfire event smoke plume.

[0011] S4. For each track, based on the spatial distribution of UVAI pixels in the track pixel matrix within the smoke puff connected domain, identify the boundary of the smoke puff connected domain and form the corresponding track-by-track smoke puff connected domain closed boundary vector.

[0012] S5. Based on the connected domain of each track smoke plume and its closed boundary vector, set discrimination conditions to filter spatially isolated wildfire independent smoke plumes.

[0013] As an optional implementation, in step S1, the satellite active fire point data includes at least the latitude and longitude of the active fire point and the detection time information;

[0014] The fire event information includes at least the latitude and longitude of the burn boundary of the burned area for each wildfire event and the fire start time.

[0015] The UVAI data includes at least the pixel UVAI value, the pixel center latitude and longitude, and the latitude and longitude of the four vertices of the rectangular area covered by the pixel.

[0016] As an optional implementation, in step S2, the center latitude and longitude of the fire-damaged area are obtained by averaging the latitude and longitude of all boundary points of the fire-damaged area;

[0017] The extent of the fire source area is determined by the number of UVAI pixels in the fire source area, and the number is 1 to 10.

[0018] As an optional implementation, in step S2, for the given orbital UVAI data, it is determined whether an active fire point of the satellite is detected within the combustion boundary of the fire event on the day of orbital transit;

[0019] For orbits containing active satellite fire points, UVAI pixels in the fire source region are marked as plume pixels.

[0020] As an optional implementation, the iterative merging in step S3 includes:

[0021] For the labeled plume pixels, the process of identifying its neighboring pixels and comparing them with the minimum threshold of UVAI is repeated, and neighboring pixels that meet the threshold condition are marked as plume pixels and merged, until the neighboring pixels can no longer meet the minimum threshold condition of UVAI.

[0022] As an optional implementation, in step S4, when the smoke cloud connected domain contains only one UVAI pixel, the latitude and longitude of the four vertices of the pixel are extracted and stored sequentially in the original order of the four vertices to form the boundary vector of the smoke cloud connected domain.

[0023] When the smoke plume connected domain contains no less than 2 UVAI pixels, the boundary points of the smoke plume connected domain are determined based on the latitude and longitude of the vertices of the rectangular area covered by each UVAI pixel in the smoke plume connected domain, and stored sequentially in a predetermined order to form the closed boundary vector of the smoke plume connected domain.

[0024] As an optional implementation, in step S5, the discrimination condition includes at least:

[0025] (a) Based on the maximum and minimum row and column numbers of the smoke patch connected components, determine whether the smoke patch connected components are located at the edge of the satellite UVAI orbit, and filter out smoke patches that are not at the orbit edge;

[0026] (b) The number of active fire points on satellite within a 20km radius of the smoke plume boundary during the fire, based on the maximum UVAI value of the smoke plume connected pixels, the mean UVAI value of the background pixels, and the mean UVAI value of the background pixels. bg , and the corresponding threshold THR_max smoke 、THR_mean smoke 、THR_mean bg 、THR_NOF bg Comparisons were made to filter out wildfire plumes that were spatially isolated.

[0027] (c) Set the area range with the latitude and longitude of the center of the burned area in step S2 as the center, and filter the smoke cloud connected domains smaller than the range.

[0028] In a second aspect, the present invention provides an automatic identification system for wildfire plumes and their boundary vectors for implementing the method described in the first aspect, comprising: a data acquisition unit, a center latitude and longitude calculation and fire source area delineation unit, a plume pixel marking unit, a connected component identification unit, a boundary vector generation unit, and an independent plume screening unit;

[0029] The data acquisition unit is connected to the center latitude and longitude calculation and fire source area delineation unit and the plume pixel marking unit, respectively, and is used to output fire event information, satellite active fire point data and orbit-by-orbit UVAI data.

[0030] The central latitude and longitude calculation and fire source area delineation unit is connected to the smoke plume pixel marking unit and is used to output the central latitude and longitude of the burned area and the fire source area.

[0031] The plume pixel marking unit, the connected component identification unit, and the boundary vector generation unit are connected in sequence to output plume pixels, smoke cloud connected components, and track-by-track smoke cloud connected component closed boundary vectors in sequence.

[0032] The independent smoke plume screening unit is connected to the central latitude and longitude calculation and fire source area delineation unit, the connected domain identification unit and the boundary vector generation unit. It is used to screen spatially isolated wildfire independent smoke plumes based on the threshold discrimination of orbit edge discrimination, UVAI statistics and background statistics and the number of active fire points, as well as the regional range discrimination.

[0033] Thirdly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, causes the processor to perform the method described in the first aspect.

[0034] Fourthly, the present invention provides an electronic device including a processor and a memory, wherein the memory stores a computer program, and when executed by the processor, the computer program causes the processor to perform the method described in the first aspect.

[0035] Compared to existing technologies, this invention abandons the reliance on visible light image features and instead utilizes the ultraviolet absorptivity aerosol index from satellite remote sensing. Since smoke particles specifically absorb ultraviolet light, this method significantly reduces misjudgments of thin clouds, haze, and other objects with similar shapes and colors, thus improving the anti-interference capability and accuracy of identification. Secondly, in terms of automation and practicality, the entire technical process achieves full-chain automation from multi-source data input to independent smoke plume vector boundary output, eliminating the need to rely on large amounts of manually labeled data to train complex models. This greatly reduces the dependence on computing resources and lowers the barrier to entry, making large-scale, operational smoke plume monitoring possible.

[0036] More importantly, this invention integrates UVAI data with information on active fire points and burned areas of fire events from satellites, ensuring that identified smoke plumes can be effectively associated with specific fire sources and enabling the tracking of independent smoke plumes. Its core lies in the design of a dedicated geometric boundary extraction algorithm for the spatial distribution of satellite pixel matrices, which can automatically and accurately generate closed vector polygons describing the complex shape of smoke plumes. This achievement directly solves the bottleneck of difficulty in quantifying the spatial shape of smoke plumes in existing technologies.

[0037] Ultimately, the generated precise vector boundaries can be directly input into atmospheric diffusion models to serve pollutant transport simulations; they provide key spatial morphological inputs for quantitatively assessing the emission flux of a single fire source, helping to solve the scientific challenge of uncertainty in wildfire emission inventories; at the same time, they also provide detailed spatial data support for studying the impact of environmental conditions on smoke plume evolution and the correlation of different chemical components within smoke plumes, comprehensively promoting the depth and precision of research on wildfire environmental and climate effects. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below.

[0039] Figure 1 A flowchart illustrating an automatic identification method for wildfire smoke plumes and their boundary vectors provided in an embodiment of the present invention;

[0040] Figure 2 This is a comparison diagram of the implementation effect of automatic identification of independent smoke plumes and their boundary vectors for two wildfire events in East Asia in this embodiment of the invention, wherein:

[0041] Figure 2 (a) in the figure is a schematic diagram of the distribution of UVAI of independent smoke plumes and the generated closed boundary vectors of a wildfire incident on May 30, 2018.

[0042] Figure 2 (b) in the middle is with Figure 2 (a) in the image corresponds to the visual representation of the wildfire smoke plume on a satellite RGB true-color image;

[0043] Figure 2 (c) in the figure is a schematic diagram of the connected domains and boundaries of independent smoke plumes identified in a wildfire incident on July 2, 2018.

[0044] Figure 2 (d) in the context is related to Figure 2 (c) in the image corresponds to the visual representation of the wildfire smoke plume on a satellite RGB true-color image;

[0045] Figure 3 This is a schematic diagram illustrating the spatial distribution of hundreds of independent smoke plumes and their closed boundary vectors of wildfires automatically identified in batches in East Asia in 2019, according to an embodiment of the present invention.

[0046] Figure 4 This is a structural block diagram of an automatic identification system for wildfire smoke plumes and their boundary vectors, provided in an embodiment of the present invention. Detailed Implementation

[0047] The technical solutions in the embodiments of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0048] Example 1:

[0049] This embodiment provides a method for automatically identifying wildfire smoke plumes and their boundary vectors. This method utilizes the ultraviolet aerosol absorption index (UVAI) obtained through satellite remote sensing technology, combined with active fire point data detected by satellite and specific information about fire events, to achieve automated and accurate identification of wildfire smoke plumes. Furthermore, it generates closed boundary vectors that describe the spatial morphology of independently distributed wildfire smoke plumes. This method overcomes the shortcomings of traditional visible light image-based smoke identification, which is easily affected by environmental factors such as lighting conditions, clouds, fog, and surface reflectivity. It also solves the problem of existing technologies struggling to automatically extract smoke plume vector boundaries over large areas.

[0050] like Figure 1 As shown, the method described in this embodiment mainly includes the following steps:

[0051] The first step, S1, involves collecting and extracting multi-source data for the target area and time period. This data primarily includes satellite active fire point data, fire event information, and satellite absorbent aerosol index (UVAI) data. Specifically, to accurately capture wildfire smoke plumes, data for specific time periods and geographical areas needs to be obtained from relevant open data product centers or other data sources. This embodiment uses data collected in May and July 2018, and throughout 2019, covering East Asia (80°-125°E, 10°-60°N) as an example. Satellite active fire point data can be sourced from monitoring products of geostationary satellites (such as Himawari-8 AHI) and polar-orbiting satellites (such as SNPP VIIRS); fire event information can be sourced from fire event data products published by the Global Wildfire Information System (GWIS); and the orbit-by-orbit UVAI data preferentially utilizes the UV absorbent aerosol index product provided by the polar-orbiting satellite S5P TROPOMI.

[0052] After acquiring the raw data, key information needs to be extracted and preprocessed. For satellite active fire point products, the system needs to extract the longitude and latitude coordinates of all active fire points, as well as the corresponding detection time information. This information will serve as the basis for subsequent judgment of the smoke plume origin. For fire event data products, the system needs to extract the longitude and latitude coordinates of the burning boundary of the burned area for each wildfire event, as well as the start time information of the fire event. This information defines the spatiotemporal range of the fire. For satellite UVAI products, the UVAI value of each satellite pixel, the center longitude and latitude coordinates of that pixel, and the crucial geometric information, namely the longitude and latitude coordinates of the four vertices of the rectangular area covered by each pixel, need to be extracted orbit by orbit. In addition, to facilitate subsequent matrix operations and neighborhood search, the row and column numbers of each pixel in the UVAI satellite orbit pixel matrix also need to be recorded. As a physical property-based index, UVAI can effectively distinguish between absorbing aerosols (such as black carbon) and non-absorbing aerosols (such as atmospheric sulfate), providing a solid physical basis for smoke plume identification.

[0053] After data collection and extraction are completed, step S2 is executed. For each recorded wildfire event, the central latitude and longitude of the burned area are calculated using the data products obtained in step S1. This central latitude and longitude is then used as a reference to delineate the fire source area, thereby marking the initial smoke plume pixels located within the fire source area. This step aims to establish the correlation between the smoke plume and a specific fire source, enabling smoke plume tracing. The specific operation process is as follows:

[0054] First, based on the burning boundary data of the burned area in the fire incident information, the latitude and longitude coordinates of all boundary points in the area are obtained, and the center latitude and longitude of the burned area are obtained by calculating the arithmetic mean of the latitude and longitude of these boundary points.

[0055] Subsequently, using the calculated latitude and longitude of the burned area center as the geometric center, a relatively small geographical area is defined as the fire source region based on the spatial resolution of the original UVAI satellite pixels. The size of this fire source region needs to be moderate, sufficient to cover the initial smoke plume above the fire point, but not too large to avoid introducing excessive background noise. The extent of the fire source region is determined by the number of UVAI pixels it contains, which is set between 1 and 10. Specifically, if the wildfire is small or the satellite sensor resolution is coarse, the smoke plume may be small and the ground area covered by a single satellite UVAI pixel may be large; in this case, an area containing 1 UVAI pixel can be selected as the fire source region. To increase the robustness of capturing the smoke plume above the fire source, an area containing 2 or more UVAI pixels can be selected. In cases of large fires or high satellite UVAI resolution, an area containing 10 UVAI pixels can be selected as the fire source region to ensure that the initial smoke plume covers the entire fire source region. For example, based on the spatial resolution of the TROPOMI UVAI satellite pixels (the nadir point is approximately 5.5km × 3.5km), a range of 0.15° × 0.15° latitude and longitude centered on the center can be set as the fire source area.

[0056] After delineating the fire source area, the system uses the start time of the fire event as the starting point to traverse and analyze all orbital UVAI data from the start to the end of the fire. For a given orbit of UVAI data, the system first determines the date of the orbit's transit, and then queries whether any active satellite fire points are detected within the combustion boundary of the fire event on that day. This determination step is crucial, as it ensures that the identified smoke plumes are indeed generated by the currently burning fire. If it is determined that there are indeed active satellite fire points within the combustion boundary on the day the orbit transits, the orbital data is confirmed as valid, and UVAI pixels located within the aforementioned delineated fire source area are then marked as plume pixels. These marked pixels will serve as seed points for subsequent region growing algorithms.

[0057] Next, step S3 is executed. For each wildfire event, using the plume image data of the fire source area marked in step S2 as seeds, the region where the wildfire event's smoke plume is continuously distributed in the horizontal space is identified track by track. This region is referred to as the smoke plume connected region below. The core of this step is to aggregate pixels belonging to the same smoke plume by utilizing spatial adjacency relationships and UVAI thresholds. The identification process of the smoke plume connected region for each track of UVAI data involves detailed iterative analysis.

[0058] ① Using the row and column numbers of the smoke plume pixels marked in step S2 in the satellite data matrix, identify their adjacent pixels in the four directions (top, bottom, left, and right) in the matrix by performing operations of row number ±1 and column number ±1 respectively. It should be noted that the search range does not include the pixels that have already been marked as smoke plume pixels themselves.

[0059] ② Set a preset minimum UVAI threshold, which represents the lowest signal intensity identified as smoke aerosol. Filter out pixels among the identified neighboring pixels whose UVAI values ​​are greater than or equal to this minimum threshold and mark them as new plume pixels. These newly marked pixels are then merged with the original smoke plume pixels from the fire source area to form the pixel set of the smoke plume distribution area for the current stage.

[0060] ③ For all currently labeled plume pixels (including initial and newly added ones), search again for their unlabeled neighboring UVAI pixels based on their row and column numbers, using a row number ±1 and column number ±1 method. Similarly, compare the UVAI values ​​of these neighboring pixels with the minimum UVAI threshold set in ②. If the UVAI value of a neighboring pixel is ≥ the minimum threshold, then label it as a plume pixel and merge it into the set of identified plume pixels.

[0061] ④ Iterate through sub-step ③. That is, continuously search outward from the newly discovered plume pixels as the boundary until the UVAI value no longer meets the threshold condition among all neighboring pixels at the edge of the current plume.

[0062] Thus, the algorithm identifies a complete smoke plume coverage area that is spatially connected to the fire source area, spatially coherent, and whose UVAI pixel values ​​all meet the threshold conditions, thus obtaining the smoke plume connectivity region of this wildfire event. This identification method based on connected component growth can effectively extract irregularly shaped smoke plumes from the background while ensuring the continuity of the smoke plumes in physical space.

[0063] After obtaining the smoke plume connected domain and all its contained pixels, step S4 is executed, which is the core step in generating accurate vector boundaries in this embodiment. For each wildfire event, the smoke plume connected domain generated orbit by orbit in step S3 is traversed. Based on the specific spatial distribution of UVAI pixels within the smoke plume connected domain in the satellite orbit pixel matrix, a matching geometric algorithm is used to identify the boundary of the smoke plume connected domain and form the corresponding orbit-by-orbit smoke plume connected domain closed boundary vector. Since satellite pixels are usually processed as rectangular (or approximately rectangular) shapes when projected onto the Earth's surface, and the shape of multiple pixels combined can be extremely complex, this embodiment designs four specific boundary extraction logics for different distribution scenarios:

[0064] Scenario 1: When the identified smoke plume's connected domain contains only one UVAI pixel, this situation typically corresponds to the early stages of a fire or when the smoke plume is extremely small. In this case, the wildfire smoke plume does not exceed the observation range of a single UVAI satellite pixel. In this scenario, the system directly extracts the latitude and longitude coordinates of the four vertices of this unique pixel and stores them sequentially according to the original order of the four vertices (e.g., clockwise or counterclockwise). This directly forms a closed quadrilateral, which is the boundary vector of the smoke plume's connected domain.

[0065] Scenario 2: When the smoke cloud's connected region contains ≥2 UVAI pixels and all of these pixels are located in the same row of the satellite orbit pixel matrix, the smoke cloud appears as a line segment along the orbit scan line direction. The processing method is as follows:

[0066] (i) Traverse all the cells in the connected component, determine the cell with the smallest column number and the cell with the largest column number, and mark them as the endpoint cells at both ends of the smoke cloud connected component. For ease of description, they can be denoted as cell D1 and cell D2 respectively.

[0067] (ii) Analyze pixel D1 at one end to determine if any of its four vertices overlap with the vertices of other pixels (i.e., its neighboring pixels) within the smoke cloud connected domain. Since they are in the same row and continuous, two vertices must be shared, while the other two vertices are exposed. Identify the two non-overlapping vertices as the two boundary points of the smoke cloud connected domain on this side, and mark them as boundary points p1 and p2.

[0068] (iii) Perform the same process on the other end of pixel D2, and identify two vertices among its four vertices that do not overlap with the vertices of other pixels in the smoke cloud connected domain, and use them as the two boundary points of the other end of the smoke cloud connected domain. In order to determine the connection order of these two points with p1 and p2, calculate the straight-line distance between these two new boundary points and the aforementioned boundary point p2, and mark the point that is closer to the other as the third boundary point p3 of the smoke cloud connected domain, and mark the point that is farther away as the fourth boundary point p4.

[0069] (iv) Store the latitude and longitude of these points in the order of p1, p2, p3, p4, so as to connect them into a closed rectangular or polygonal boundary, and obtain the boundary vector of the smoke cloud connected domain in this scenario.

[0070] Scenario 3: When the smoke plume's connected domain contains ≥2 UVAI pixels, all located in the same column of the satellite orbit pixel matrix, the smoke plume appears as a line segment along the orbital flight direction. The processing logic is similar to Scenario 2, but with a rotation in direction. First, identify the two pixels with the largest and smallest row numbers within the smoke plume's connected domain and mark them as the UVAI pixels at both ends of the connected domain. Second, process according to the same logic flow as steps (ii) to (iv) in Scenario 2, i.e., identify the exposed vertices of the endpoint pixels and connect them in order of distance, thereby obtaining the boundary vector of the smoke plume's connected domain in this scenario.

[0071] Scenario 4: This is the most common and complex scenario, where the number of UVAI pixels in the smoke plume's connected domain is no less than 2, and the total number of rows and columns in the smoke plume's connected domain is no less than 2. In this case, the smoke plume presents an irregular two-dimensional planar distribution. The system uses a method of row-by-row scanning and ordered concatenation of boundary points for processing. Specifically, starting from the row with the smallest row number in the smoke plume's connected domain, the row numbers are increased sequentially, and boundary extraction is performed row by row. For each row of UVAI smoke plume pixels, the goal is to mark the boundary points of the smoke plume at both ends of the row and determine the storage order of the two boundary points at each end. For ease of explanation, we will use the center point of the row as a reference, and mark the two boundary points located at the end with the smaller column number (assuming it is the left side) as the points on the boundary_1 boundary, and the two boundary points located at the end with the larger column number (assuming it is the right side) as the points on the boundary_2 boundary. We will then uniformly store the two boundary points on each of boundary_1 and boundary_2 in the direction of increasing row number.

[0072] Taking the j-th row of the smoke cloud connected domain as an example, the specific processing details are as follows:

[0073] (i) When the number of UVAI plume pixels in row j is ≥ 2, first determine the two pixels with the largest and smallest column numbers in that row, and mark them as the pixels at both ends of the plume connected domain in row j. The pixel with the smaller column number is denoted as pixel A, and the pixel with the larger column number is denoted as pixel B. For pixel A, calculate the distances of its four vertices from the geometric center of pixel B. Obviously, the two vertices with the largest distances are the outermost vertices, and these two vertices are marked as two boundary points on the boundary of plume boundary_1. To determine the storage order of these two points, construct a vector from these two boundary points and calculate the cross product between this vector and the direction vector of the increasing row number direction of the plume connected domain. The sign of the cross product indicates the relative position of these two points with respect to the direction of movement, thus determining the storage order. Store these two points in the boundary_1 list in the order of increasing row number. For pixel B, use a similar method to calculate the distances of its four vertices from the center point of pixel A, and mark the two vertices with the largest distances as two boundary points on the boundary of plume boundary_2. Similarly, the storage order of these two boundary points is determined using the vector product method, and they are stored in the boundary_2 list in chronological order along the direction of increasing row number.

[0074] (ii) When the smoke patch connected component has only one UVAI cell in row j, it is denoted as cell C. Since this is a connected component and the total number of rows is ≥2, there must be an adjacent UVAI cell in row j-1 or row j+1. The adjacent cells on both sides are processed separately.

[0075] Case 1: When pixel C has adjacent pixels only on one side (e.g., only in row j-1 or only in row j+1), this usually occurs at the protrusion or end of the smoke cloud. First, identify two vertices of pixel C that do not overlap with any other vertices in the smoke cloud's connected domain (i.e., fully exposed vertices). Form a vector from these two vertices, and calculate the direction vector of this vector relative to the column number increasing direction (denoted as...). The vector product between the two vertices is calculated. By determining the sign of the vector product, the vertex closer to the smaller column number is identified and marked as a boundary point on the boundary of boundary_1 (the smoke cloud's boundary). The other vertex is marked as a boundary point on the boundary of boundary_2. Further, two vertices of pixel C that overlap with other pixels within the smoke cloud's connected domain (i.e., vertices touching adjacent rows) are identified. The vector product formed by these two vertices is then calculated. The vector product between the two vertices is used to determine whether it is positive or negative. The vertex closer to the smaller column number is marked as a point on boundary_1, and the other vertex is marked as a point on boundary_2.

[0076] Case 2: When pixel C has adjacent UVAI pixels on both sides (i.e., in rows j-1 and j+1), this indicates that the pixel acts as a bridge connecting the upper and lower parts. In this case, calculate the distances between the four vertices of pixel C and the center of one of the adjacent pixels on one side. Construct a vector from the two closest vertices (i.e., the contact points), and calculate the distance between this vector and the center of the adjacent pixel on one side. The vector product between the points determines which points are closer to the smaller column number and are assigned to boundary_1, and which are farther away and assigned to boundary_2. The remaining two vertices of pixel C (the other contact point) are then used to construct a vector, and the same discrimination and classification process is applied.

[0077] In both case 1 and case 2, after determining the points belonging to boundary_1 and boundary_2, it is necessary to construct a vector for this pair of points on boundary_1, calculate its vector product with the row number increment direction vector, and determine the storage order; the same process is performed for this pair of points on boundary_2.

[0078] (iii) Following the above method, we can obtain two sequential points on boundary_1 and two sequential points on boundary_2 in row j+1. The data in row j+1 is appended to the data in row j to maintain the continuity of the linked list.

[0079] (iv) Iterate through the above steps row by row until all rows have been traversed, identifying the complete sequence of boundary points on the boundaries of the entire smoke cloud connected domains boundary_1 and boundary_2. Finally, to form a closed loop, the order of all boundary points in the boundary_2 list needs to be reversed (because boundary_1 is stored in ascending order of row number, while for closure, boundary_2 boundary points need to be backtracked in descending order of row number). Starting from the position after the last boundary point in the boundary_1 list, merge and store the reversed boundary points of boundary_2 in order. In this way, the point sequence extends along one side of the smoke cloud in the same direction (clockwise or counterclockwise) to the end, and then returns to the starting point along the other side in the same direction, forming a complete boundary vector of the smoke cloud connected domain.

[0080] Finally, step S5 is executed. Using the smoke plume connectivity and its closed boundary vector data identified in step S4 for each satellite orbit, strict discrimination conditions are set to filter out spatially isolated wildfire smoke plumes. This step is to eliminate smoke plumes that are background noise or, although connected, may actually be large-scale diffuse smoke rather than isolated distributions. The filtering process includes the following aspects:

[0081] Orbit edge filtering: Based on the maximum and minimum row and column numbers of the smoke plume's connected components, it is determined whether the smoke plume's connected components are located at the orbit edge of the satellite UVAI data. If the smoke plume touches the orbit edge, it means that the part of the plume beyond the edge cannot be observed. In this case, the identified smoke plume's connected components are incomplete. Therefore, these smoke plumes located at the orbit edge are filtered out, and complete smoke plumes that are not at the orbit edge are retained.

[0082] UVAI statistical characteristics and background comparison screening: Based on the physical characteristics of independent smoke plumes that have not yet spread widely or fully mixed with the background atmosphere, the UVAI value within their area is usually significantly higher than the surrounding background. Therefore, the following comprehensive criterion formula is set for screening:

[0083] ;

[0084] The meanings of each parameter are as follows:

[0085] Let $\mathbf{v}$ be the maximum value of $UVAI for all pixels within the connected domain of the smoke cloud. The maximum threshold value for UVAI is set.

[0086] The average value of UVAI for all pixels within the connected domain of the smoke cloud. The set threshold for the mean UVAI;

[0087] The UVAI background value is specifically calculated as the mean UVAI value of all pixels within the smallest bounding rectangle of the smoke plume's boundary, but not belonging to the plume's pixels. The maximum threshold for the set background mean;

[0088] This refers to the number of satellite-activated fire points within a 20-kilometer radius extending outward from the boundary of the smoke plume during the period from the fire's initiation to its eventual extinguishing. This is the minimum threshold for the number of active fire points.

[0089] Optionally, in this embodiment, It can be set to 1.5, which means that the UVAI in the core area of ​​the confirmed smoke cloud must reach a certain intensity; It can be set to 0.25 to ensure that the overall smoke concentration is higher than the background atmosphere; It can be set to 0.0, which requires that there are no absorbent aerosols similar to the smoke cloud in the surrounding background to ensure the isolation of the smoke cloud; A value of 0 can be set, meaning there are no other fire points within a 20-kilometer radius of the smoke plume, ensuring that the smoke plume is not a mixed plume formed from emissions from other nearby fires. Of course, these thresholds can also be fine-tuned based on the sensitivity of different satellite sensors and the average background characteristic concentration of local absorbent aerosols. For example... It can also be selected as 1.4, 1.5, or 1.6, etc.; It can also take values ​​between 0.2 and 0.3; It can take values ​​between -0.1 and 0.1.

[0090] Regional Scope Filtering: Using the latitude and longitude of the wildfire event's burned area calculated in step S2 as the center, a certain regional scope (e.g., a 10°×10° latitude and longitude area) is defined, and smoke plume connectivity regions located within this scope (smaller than this scope) are filtered out. This step is to exclude those that, although meeting the above conditions, are too large in scale and may belong to large-scale circulation-induced large-scale smoke bands, thereby ensuring that the identified objects are spatially relatively limited independent smoke plumes.

[0091] Through the complete processing of steps S1 to S5 described above, the method of this embodiment can automatically output all independent wildfire plumes within the target area that meet the conditions and their corresponding precise closed boundary vectors. For example... Figure 2 As shown, Figure 2 (a) shows an isolated smoke plume from a wildfire in East Asia on May 30, 2018, identified by this method. The thick black line clearly outlines the boundary of the smoke plume, and different colors within the boundary represent UVAI values ​​(warmer colors indicate higher values). (Comparison) Figure 2 In the true-color image (b), it can be seen that the identified boundary vectors highly match the actual smoke morphology. Similarly, Figure 2 (c) and Figure 2 (d) in the figure shows the accurate identification results of another case in which the method still effectively identified and automatically extracted the boundary vector of the smoke cloud despite the significant visual interference caused by the presence of clouds in the image. Figure 3 The results of batch processing in 2019 are shown. The red boxes in the figure mark hundreds of individual smoke plumes, verifying the stability and effectiveness of this method in large-scale operational monitoring.

[0092] In summary, the method in this embodiment abandons the reliance on visible light image features and utilizes the more physically defined UVAI index, combined with geometric topology algorithms, to achieve full-chain automation from multi-source data input to independent smoke plume vector boundary output. This method requires fewer computational resources, eliminates the need for training samples, and effectively tracks the spread and distribution of wildfire smoke plumes, providing crucial spatial morphological data support for early warning of wildfire emissions air pollution, reducing exposure risks, and quantitatively assessing wildfire emissions.

[0093] Example 2:

[0094] This embodiment provides an automatic identification system for wildfire smoke plumes and their boundary vectors for implementing the method described in Embodiment 1. For example... Figure 4As shown, the system includes: a data acquisition unit, a center latitude and longitude calculation and fire source area delineation unit, a plume pixel marking unit, a connected component identification unit, a boundary vector generation unit, and an independent smoke plume screening unit. These units are physically or logically interconnected and work together to complete the automated identification task.

[0095] Specifically, the data acquisition unit is connected to the center latitude and longitude calculation and fire source area delineation unit and the plume pixel marking unit. The main function of this unit is to collect and extract satellite active fire point data, fire event information, and orbit-by-orbit UVAI data from external databases or receivers. It transmits the latitude and longitude of active fire points and detection time, the combustion boundary and start time of fire events, and the numerical and geometric vertex information of UVAI pixels to subsequent units.

[0096] The center latitude and longitude calculation and fire source area delineation unit is connected to the plume pixel marking unit. This unit receives fire event information from the data acquisition unit and calculates the center latitude and longitude of the burned area by taking the arithmetic mean of the latitude and longitude of the burning boundary points. Subsequently, according to preset rules (such as including 1 to 10 pixels), this unit delineates the fire source area based on the center latitude and longitude and outputs the spatial range information of the fire source area to the plume pixel marking unit.

[0097] The plume pixel labeling unit connects the data acquisition unit, the center latitude and longitude calculation and fire source area delineation unit, and the connected component identification unit. It is responsible for traversing the UVAI data during the fire and determining whether there are active fire points within the combustion boundary on the day the orbit passes. For orbits that meet the conditions, this unit labels the UVAI pixels located within the fire source area as initial plume pixels and transmits this seed point information to the connected component identification unit.

[0098] The connected component identification unit is connected to the boundary vector generation unit. This unit executes a region growing algorithm, starting from the received plume pixels, and iteratively searches and merges neighboring pixels that meet the conditions based on the adjacency relationship of the pixel row and column numbers (row number ±1, column number ±1) and the preset minimum UVAI threshold, until it can no longer be expanded, thereby identifying the complete connected component of the wildfire event plume.

[0099] The boundary vector generation unit is one of the core computational modules of the system, connected after the connected component identification unit. For each identified smoke cloud connected component, this unit calls the corresponding geometric boundary extraction algorithm (i.e., the four scenario processing logics detailed in Example 1) based on the spatial distribution of its pixels in the matrix (such as single point, single row, single column, or irregular horizontal plane) to calculate an ordered sequence of boundary point latitude and longitude, generating a closed boundary vector for each track-by-track smoke cloud connected component.

[0100] The independent smoke plume screening unit is connected to the center latitude and longitude calculation and fire source area delineation unit, the connected component identification unit, and the boundary vector generation unit. This unit receives the generated boundary vectors and connected component statistics, and filters them based on preset discrimination conditions. These conditions include: determining whether it is located at the edge of the fire track, comparing the statistical threshold of the maximum / mean UVAI value with the background mean, and checking the number of active fire points within a 20-kilometer radius. Simultaneously, it also combines the center location of the burned area to filter smoke plumes within a specific region. Finally, this unit outputs the filtered, spatially isolated wildfire independent smoke plumes and their precise boundary vectors.

[0101] The system provided in this embodiment uses a modular design to break down the complex smoke plume identification process into independent functional units, making it easy to deploy on high-performance computing clusters or conventional servers. This system can automatically and continuously process massive amounts of satellite data, providing environmental monitoring departments with real-time spatial distribution products of wildfire smoke plumes.

[0102] Example 3:

[0103] This embodiment provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by the hardware of a computer or processor, it enables the processor to perform any step of the automatic identification method for wildfire smoke plumes and their boundary vectors as described in Embodiment 1.

[0104] The storage medium can be any medium capable of storing program code, such as read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk. The computer program contains code instructions for performing data acquisition, fire source center calculation, pixel marking, region growing iteration, complex geometric boundary extraction logic, and multi-parameter filtering algorithms. By executing this program, the computer can automatically convert raw satellite data into high-precision smoke plume vector patterns without manual intervention.

[0105] Example 4:

[0106] This embodiment provides an electronic device, which includes a processor and a memory. The memory stores a computer program, which, when executed by the processor, causes the processor to perform the automatic identification method for wildfire smoke plumes and their boundary vectors as described in Embodiment 1.

[0107] This electronic device can be a desktop computer, laptop computer, server, or cloud computing node. The processor can be a general-purpose central processing unit (CPU), application-specific integrated circuit (ASIC), digital signal processor (DSP), or other programmable logic device. The electronic device may also include a communication interface for acquiring real-time satellite active fire points and UVAI data from external data sources (such as meteorological satellite data receiving stations or internet databases). By running the aforementioned program on this electronic device, operational wildfire smoke monitoring can be achieved, providing powerful hardware computing support for atmospheric environmental research and disaster early warning.

[0108] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for automatic identification of wildfire smoke plumes and their boundary vectors, characterized in that, include: S1. Acquire satellite active fire point data, fire event information, and orbit-by-orbit absorptive aerosol index (UVAI) data for the target area and target time period; S2. For each wildfire event, calculate the center latitude and longitude of the burned area based on the fire event information, and delineate the fire source area based on the center latitude and longitude; traverse the orbital UVAI data from the start of the fire to its extinguishment, and mark the UVAI pixels in the fire source area as plume pixels on the orbit that detects active fire points within the combustion boundary of the fire event on the day the orbit passes over. S3. For each orbit, starting with the plume pixel, search for adjacent pixels based on the row and column numbers of the pixel in the satellite orbit pixel matrix, according to the adjacency relationship of row number ±1 and column number ±1, and iteratively merge adjacent pixels that meet the threshold conditions into plume pixels according to the preset UVAI minimum threshold, until no new adjacent pixels that meet the threshold conditions are added, thus obtaining the connected domain of the wildfire event smoke plume. S4. For each track, based on the spatial distribution of UVAI pixels in the track pixel matrix within the smoke puff connected domain, identify the boundary of the smoke puff connected domain and form the corresponding track-by-track smoke puff connected domain closed boundary vector. S5. Based on the connected domain of each track smoke plume and its closed boundary vector, set discrimination conditions to filter spatially isolated wildfire independent smoke plumes.

2. The method for automatic identification of wildfire smoke plumes and their boundary vectors according to claim 1, characterized in that: In step S1, the satellite active fire point data includes at least the latitude and longitude of the active fire point and the detection time information; The fire event information includes at least the latitude and longitude of the burn boundary of the burned area for each wildfire event and the fire start time. The UVAI data includes at least the pixel UVAI value, the pixel center latitude and longitude, and the latitude and longitude of the four vertices of the rectangular area covered by the pixel.

3. The method for automatic identification of wildfire smoke plumes and their boundary vectors according to claim 1, characterized in that: In step S2, the center latitude and longitude of the burned area are obtained by averaging the latitude and longitude of all boundary points of the burned area; The extent of the fire source area is determined by the number of UVAI pixels in the fire source area, and the number is 1 to 10.

4. The method for automatic identification of wildfire smoke plumes and their boundary vectors according to claim 1, characterized in that: In step S2, for the given orbital UVAI data, it is determined whether an active fire point of the satellite is detected within the burning boundary of the fire event on the day of orbital transit. For orbits containing active satellite fire points, UVAI pixels in the fire source region are marked as plume pixels.

5. The method for automatic identification of wildfire smoke plumes and their boundary vectors according to claim 1, characterized in that, The iterative merging in step S3 includes: For the labeled plume pixels, the process of identifying its neighboring pixels and comparing them with the minimum threshold of UVAI is repeated, and neighboring pixels that meet the threshold condition are marked as plume pixels and merged, until the neighboring pixels can no longer meet the minimum threshold condition of UVAI.

6. The method for automatic identification of wildfire smoke plumes and their boundary vectors according to claim 1, characterized in that: In step S4, when the smoke cloud connected domain contains only one UVAI pixel, the latitude and longitude of the four vertices of the pixel are extracted and stored sequentially according to the original order of the four vertices to form the boundary vector of the smoke cloud connected domain. When the smoke plume connected domain contains no less than 2 UVAI pixels, the boundary points of the smoke plume connected domain are determined based on the latitude and longitude of the vertices of the rectangular area covered by each UVAI pixel in the smoke plume connected domain, and stored sequentially in a predetermined order to form the closed boundary vector of the smoke plume connected domain.

7. The method for automatic identification of wildfire smoke plumes and their boundary vectors according to claim 1, characterized in that, In step S5, the discrimination conditions include at least: (a) Based on the maximum and minimum row and column numbers of the smoke patch connected components, determine whether the smoke patch connected components are located at the edge of the satellite UVAI orbit, and filter out smoke patches that are not at the orbit edge; (b) The number of active fire points on satellite within a 20km radius of the smoke plume boundary during the fire, based on the maximum UVAI value of the smoke plume connected pixels, the mean UVAI value of the background pixels, and the mean UVAI value of the background pixels. bg , and the corresponding threshold THR_max smoke 、THR_mean smoke 、THR_mean bg 、THR_NOF bg Comparisons were made to filter out wildfire plumes that were spatially isolated. (c) Set the area range with the latitude and longitude of the center of the burned area in step S2 as the center, and filter the smoke cloud connected domains smaller than the range.

8. A system for automatically identifying wildfire smoke plumes and their boundary vectors for implementing the method according to any one of claims 1 to 7, characterized in that, include: The system includes a data acquisition unit, a center latitude and longitude calculation and fire source area delineation unit, a smoke plume pixel marking unit, a connected component identification unit, a boundary vector generation unit, and an independent smoke plume screening unit. The data acquisition unit is connected to the center latitude and longitude calculation and fire source area delineation unit and the plume pixel marking unit, respectively, and is used to output fire event information, satellite active fire point data and orbit-by-orbit UVAI data. The central latitude and longitude calculation and fire source area delineation unit is connected to the smoke plume pixel marking unit and is used to output the central latitude and longitude of the burned area and the fire source area. The plume pixel marking unit, the connected component identification unit, and the boundary vector generation unit are connected in sequence to output plume pixels, smoke cloud connected components, and track-by-track smoke cloud connected component closed boundary vectors in sequence. The independent smoke plume screening unit is connected to the central latitude and longitude calculation and fire source area delineation unit, the connected domain identification unit and the boundary vector generation unit. It is used to screen spatially isolated wildfire independent smoke plumes based on the threshold discrimination of orbit edge discrimination, UVAI statistics and background statistics and the number of active fire points, as well as the regional range discrimination.

9. A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, causes the processor to perform the method according to any one of claims 1 to 7.

10. An electronic device, comprising a processor and a memory, wherein the memory stores a computer program that, when executed by the processor, causes the processor to perform the method according to any one of claims 1 to 7.

Images