Post-processing method based on remote sensing image fire point recognition result

Abstract

The invention discloses a post-processing method based on a remote sensing image fire point identification result, and relates to the technical field of fire point identification, a sliding time window method is used to acquire a fire point set when a T + 0 period task starts, and fire points comprise coordinates, generation time TG and affiliated attributes of a solar flare angle SG. According to the post-processing method based on the fire point recognition result of the remote sensing image, the correlation of fire points in the space-time dimension is fully utilized, the fire points with adjacent geographic positions in the same time phase are aggregated into a'fire scene ', then the fire scene is filtered by utilizing statistical information of a similar fire point set, 'non-fire scenes' caused by most solar flare angles and the like are eliminated, and the fire point recognition result of the remote sensing image is obtained. The false alarm rate of fire point identification is greatly reduced; and overlapping fire scenes in different time phases are aggregated into a fire event by using a space-time aggregation algorithm, and meanwhile, most of accidental non-fire scenes are further eliminated, so that the false alarm rate of fire point identification is further reduced.

Description

Technical Field

[0001] This invention relates to the field of fire point recognition technology, specifically a post-processing method based on fire point recognition results from remote sensing images. Background Technology

[0002] Wildfires, as a sudden disaster, pose a huge threat to people's lives and property as well as the natural environment. They not only cause property damage but also cause serious environmental pollution. In order to effectively monitor and warn of wildfires, infrared satellite remote sensing technology is widely used in fire point monitoring. By capturing information on surface thermal anomalies, infrared satellite remote sensing technology can quickly identify fire points. However, although existing fire point identification methods, such as traditional methods and deep learning methods, have achieved certain results in fire point identification, there are still some problems that need to be solved.

[0003] Currently, fire point identification methods based on infrared satellite remote sensing images generally use fire points as the basic unit, ignoring the spatiotemporal correlation between fire points. They fail to scientifically and intuitively organize and aggregate fire point data, resulting in a large number of false alarms in the identification results. Non-fire points such as clouds, water bodies, and snow reflecting sunlight are mistakenly identified as fire points. Furthermore, in visualization systems, fire points are directly displayed as the unit. Due to the potentially large number of fire points, this can lead to slow system loading and updates, which is detrimental to the system's interactive experience. Summary of the Invention

[0004] To achieve the above objectives, the present invention provides a post-processing method based on fire point identification results from remote sensing images, comprising the following steps: S1. Use the sliding time window method to obtain the set of fire points at the start of the T+0 period mission. The fire points include coordinates (geospatial coordinates of the center point coordinates of the thermal anomaly pixels in the infrared remote sensing image), generation time TG, and solar flare angle SG, etc. S2. Use the fire scene aggregation algorithm to aggregate the fire point set to form a fire scene set; S3. Use the fire scene filtering method to filter the fire scene set to obtain the filtered fire scene set; S4. In the time dimension, the spatiotemporal aggregation algorithm is used to aggregate the filtered fire scene set to obtain the set of fire events in period T+0. S5. The T-1 fire event and the T+0 fire event are spliced ​​together to obtain the updated fire event.

[0005] Preferably, the start time of the T+0 period task is t0, and the width of the sliding time window is D days, that is, the start time of the window is t0-D days, and the end time is t0; The fire point includes auxiliary attributes, such as coordinates, generation time (TG), and solar flare angle (SG), but is not limited to these. The coordinates correspond to the geospatial coordinates of the center point coordinates of the thermal anomaly pixel in the infrared remote sensing image, and the spatial resolution of the infrared remote sensing image is SR.

[0006] Preferably, in step S1, the process of obtaining the fire point set is as follows: S1.1. By setting the start time of the T+0 period task as t0 as the starting point of the current processing cycle, the width of the sliding time window D days is determined, such as the past 1 hour, the past 6 hours, or the past 24 hours. The longer the window width, the more fire point information can be captured, but it may also contain more noise and false alarms. The step size of each sliding of the time window is determined, which is a fixed time interval (sliding once every 5 minutes or every 10 minutes) or a dynamically adjusted step size. S1.2 Collect all infrared remote sensing image data within the time window [t0-D, t0]. Within the current sliding time window, traverse the infrared remote sensing image data and use the fire point recognition algorithm to identify thermal anomaly pixels. S1.3 For each identified fire point, extract its geospatial coordinates, geospatial coordinates and solar flare angle as auxiliary attributes. Among them, the geospatial coordinates of the fire point are determined according to the center point coordinates of the thermal anomaly pixel, and the generation time TG of the fire point is recorded, that is, the specific time when the fire point is identified. Taking into account the influence of the sun's position and angle on the fire point identification, the solar flare angle SG is calculated. S1.4. All identified fire points and their associated attributes are grouped into a fire point set and stored in the fire point set in order of their generation time. The fire generation time is the moment when the fire point is generated. S1.5. Based on the infrared remote sensing image data source, determine the fire point generation frequency as Ni times per day, and determine the daily execution frequency M of the task, ensuring that M is not less than the maximum frequency among all data sources, i.e. , Generate the frequency of corresponding fire points for the i-th type of infrared remote sensing image data source; S1.6. Based on the task execution frequency M and the window width D, dynamically adjust the sliding time window to ensure that the latest fire point data is obtained at each time point.

[0007] Preferably, the fire field is a set of fire points that satisfy the nearest neighbor condition at the time a certain fire point is generated; The nearest neighbor condition is as follows: for any fire point within the fire area, there is at least one other fire point within the same fire area such that its spatial distance from that fire point is less than a fire point distance threshold, wherein the fire point distance threshold is greater than... .

[0008] Preferably, in step S2, the formation process of the fire cluster is as follows: S2.1 Perform preprocessing operations on the fire point set, including data cleaning and data standardization steps; S2.2 Aggregate all fire points according to the generation time of each fire point in the set to generate M*D fire point subsets; S2.3. For each fire point in the subset of fire points, generate a circular buffer with radius R centered on its coordinates, where... SR represents the spatial resolution of infrared remote sensing images; S2.4 Merge the circular buffers in each fire point subset to form a fire field set. Each fire field contains auxiliary attributes, including fire field center point coordinates, fire field boundary polygon, fire field generation time, solar flare angle 1 / 4 quantile, number of fire points contained, and minimum bounding rectangle of the fire field, but not limited to these. Among them, the fire field center point coordinates are the geographic center coordinates of the fire field, the fire field boundary polygon is determined by the boundary of the merged buffers and is used to define the boundary of the fire field, the fire field generation time is the earliest time of fire formation, the solar flare angle 1 / 4 quantile is the 1 / 4 quantile of the solar flare angle of all fire points in the fire field, which is used as a representative value of the solar flare angle of the fire field, the number of fire points contained is the total number of fire points in the fire field, and the minimum bounding rectangle of the fire field refers to the smallest rectangle that surrounds the boundary of the fire field, which is used to simplify the geometric representation and calculation of the fire field.

[0009] Preferably, in step S3, the process of obtaining the filtered fire scene set is as follows: S3.1 Perform attribute analysis on each fire in the fire set, including auxiliary attributes such as the coordinates of the fire center point, the boundary polygon, the generation time, the 1 / 4 quantile of the solar flare angle, the number of fire points contained, and the minimum bounding rectangle of the fire. S3.2 Filtering solar flare angles: Compare the 1 / 4 quantile of the solar flare angle with a set solar flare angle threshold A. Mark fires with solar flare angles less than threshold A as "non-fires," which helps to eliminate false fire reports caused by solar flares. ; S3.3 Filter the number of fire points. Count the number of fire points contained in a fire scene and mark fire scenes with a number of fire points less than a set threshold B as "non-fire scenes". This helps to exclude small-scale fire point sets that may be caused by non-fire factors. ; S3.4. Conduct a comprehensive analysis of various ancillary attributes and environmental factors of the fire scene, perform a final assessment and screening of the fire scene, determine whether the fire scene is a real fire event, and output a filtered set of fire scenes. The filtered set of fire scenes is considered to be a real fire event.

[0010] Preferably, the fire event is a time series set composed of fire sites that satisfy the spatiotemporal correlation conditions in the time dimension, and the spatiotemporal correlation conditions are determined by specific spatial aggregation algorithms and time aggregation algorithms; The conditions for the existence of a fire event are as follows: if a fire event exists, then any time window with a width of D0 days is taken within the time interval from the start to the end of the fire event, and there are no less than K0 fire sites within the window; otherwise, the fire event does not exist. The termination condition for the fire event is as follows: In the T+0 period task described in S1, if the difference between the fire generation time of the last fire scene in a certain fire event and the start time t0 of the task is greater than or equal to D0 days, then the fire event is considered to have ended in the T+j period task. .

[0011] Preferably, in step S4, the process of obtaining the set of fire events in period T+0 is as follows: S4.1. Use a spatial aggregation algorithm to process the filtered fire scene set to obtain a fire scene time series set, wherein each fire scene time series satisfies the following condition: for any filtered fire scene in the sequence, there exists at least one other filtered fire scene in the same sequence whose boundary polygon intersects with the boundary polygon of the arbitrary filtered fire scene. S4.2. Based on the sliding time window and generation frequency, each fire time series is expanded into a one-dimensional discrete sequence of length D*M. In this discrete sequence, the generation time of each fire is assigned a value of 1, and the other discrete time points are assigned a value of 0. S4.3. Traverse and process all the aforementioned one-dimensional discrete sequences to obtain a set of fire events. Each one-dimensional discrete sequence, after being processed by a time aggregation algorithm, generates L fire events. The time aggregation algorithm is a density-based clustering algorithm, including, but not limited to, the DBSCAN algorithm, the OPTICS algorithm, and the DPC algorithm; S4.4 Each fire event includes auxiliary attributes, such as event number, event status, fire sites involved in the event, event start time, event end time, cumulative impact range of the event, and event address, but not limited to these. Among them, the event number is unique, the event status has only two types: "ended" and "ongoing", the fire sites involved in the event are: all filtered fire sites involved in the fire event up to the start time t0 of the task in S1, the event start time is: the minimum value of the fire site generation time among the fire site attributes involved in the event, the event end time is: the maximum value of the fire site generation time among the fire site attributes involved in the event, and this value is less than or equal to the start time t0 of the task in S1, the cumulative impact range of the event is: the outer boundary of all the fire sites involved in the event superimposed, and the event address is: the geographical coordinates and corresponding province, city, county, street, etc., corresponding to the center of the cumulative impact range of the event.

[0012] Preferably, in step S5, the process of obtaining the updated fire event is as follows: S5.1 Query the fire events in period T-1 in the database storing fire events according to the query conditions, wherein the query conditions are: the event end time is within the range of t0-D-D1 to t0-D1, where... The unit is days; S5.2 Match fire events in period T-1 and fire events in period T+0, find all the same fire events and update them; S5.3 Match the fire events of period T-1 and period T+0, find all newly added fire events and add them to the database. Among them, the newly added fire events must meet the following conditions: they do not exist in the fire event set of period T-1, and only exist in the fire event set of period T+0. Iterate through all the newly added fire events and add them to the database. S5.4 Match T-1 and T+0 fire events, identify all fire events that need to be terminated directly, and update them. Specific steps include: S5.41. Identify all fire events that need to be terminated directly. The fire events that need to be terminated directly must meet the following conditions: they exist only in the fire event set of period T-1 and do not exist in the fire event set of period T+0. S5.42. Traverse all the fire events that need to be terminated directly, and perform update operations on the auxiliary attributes of the third fire event to be updated in each database. The third fire event to be updated has the same event number as the fire event that needs to be terminated directly. The update operation is to update the event status to "terminated".

[0013] Preferably, in S5.2, the specific steps for identifying and updating all the same fire events include: S5.21. Find all the same fire events, wherein the same fire event satisfies the following conditions: it exists in both the set of fire events in period T-1 and the set of fire events in period T+0, and the fire sites contained in the two events have an intersection. The part that exists in the set of fire events in period T-1 is called the same fire event in period T-1, and the part that exists in the set of fire events in period T+0 is called the same fire event in period T+0. Assign the event number of the same fire event in period T-1 to the same fire event in period T+0. S5.22. Find all fire events with new fire sites, wherein the fire events with new fire sites are included in the set of the same fire events and satisfy the following conditions: at least one updated fire site exists, exists in the fire sites included in the event of the same fire event in period T+0, and does not exist in the fire sites included in the event of the same fire event in period T-1. Traverse the set of fire events with new fire sites, and perform update operations on the subordinate attributes of the first fire event to be updated in each database. The first fire event to be updated has the same event number as the fire events with new fire sites. The update operation includes: updating the event status to "ongoing", updating the event end time to the event end time of the same fire event in period T+0; appending the updated fire site to the fire sites included in the event, and recalculating and updating the cumulative impact range of the event based on the updated fire sites included in the event. S5.23. Find all fire events without new fire sites, wherein the fire events without new fire sites are included in the set of the same fire events and are the complement of the set of fire events with new fire sites. Traverse the set of fire events without new fire sites and perform update operations on the subordinate attributes of the second fire event to be updated in each database. The second fire event to be updated has the same event number as the fire events without new fire sites. The update operation is: according to the time aggregation algorithm result described in S4, update the event status to "continued" or "ended".

[0014] This invention provides a post-processing method based on fire point identification results from remote sensing images. It has the following beneficial effects: I. This post-processing method based on remote sensing image fire point identification results fully utilizes the spatiotemporal correlation between fire points, aggregating geographically adjacent fire points in the same time phase into "fire fields". Then, it uses statistical information of similar fire point sets to filter the fire fields, eliminating most "non-fire fields" caused by solar flare angles, etc., significantly reducing the false alarm rate of fire point identification. Furthermore, it uses a spatiotemporal aggregation algorithm to aggregate overlapping "fire fields" in different time phases into "fire events", further eliminating most occasional "non-fire fields", further reducing the false alarm rate of fire point identification.

[0015] Second, this post-processing method based on the fire point identification results of remote sensing images scientifically and intuitively organizes and aggregates the fire point data, giving each fire event a unique number, realizing the cataloging of fire events, and updating the results of each fire event in near real-time. This enables the unique location of the fire event while continuously providing early warning, tracking, trend analysis, and disaster analysis.

[0016] Third, the post-processing method based on the fire point identification results of remote sensing images further aggregates the fire point results into "fire scene" and "fire event" and uses them as the display unit on the homepage, which improves the data loading and update speed of the visualization system and enhances the interactive experience of the system; it can also perform multi-dimensional statistics on fires: the granularity of the display can be rolled up to "fire event" or drilled down to "fire scene" and "fire event", which enhances the three-dimensionality of the data displayed by the system. Attached Figure Description

[0017] Figure 1 This is a flowchart of a post-processing method based on fire point identification results from remote sensing images according to the present invention; Figure 2 This is a schematic diagram of the method steps of the present invention. Detailed Implementation

[0018] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments. The embodiments of the present invention are given for illustrative and descriptive purposes only, and are not intended to be exhaustive or to limit the invention to the forms disclosed. Many modifications and variations will be apparent to those skilled in the art. The embodiments were chosen and described to better illustrate the principles and practical application of the invention, and to enable those skilled in the art to understand the invention and design various embodiments with various modifications suitable for a particular purpose.

[0019] First embodiment, such as Figure 1 , Figure 2 As shown, the present invention provides a technical solution: a post-processing method based on fire point identification results from remote sensing images, comprising the following steps: S1. Use the sliding time window method to obtain the set of fire points at the start of the T+0 period mission. The fire points include coordinates (geospatial coordinates of the center point coordinates of the thermal anomaly pixels in the infrared remote sensing image), generation time TG, and solar flare angle SG, etc. The start time of the T+0 period mission is t0, and the width of the sliding time window is D days. That is, the start time of the window is t0-D days and the end time is t0. The fire points include auxiliary attributes, including coordinates, generation time TG, and solar flare angle SG, but not limited to these. The coordinates correspond to the geospatial coordinates of the center point coordinates of the thermal anomaly pixels in the infrared remote sensing image. The spatial resolution of the infrared remote sensing image is SR. Furthermore, the process of obtaining the fire point set is as follows: By setting the start time of the T+0 period task as t0 as the starting point of the current processing cycle, the width of the sliding time window, D days, is determined, such as the past 1 hour, past 6 hours, or past 24 hours. A longer window width allows for the capture of more fire point information, but may also include more noise and false alarms. The step size for each sliding of the time window is determined, either as a fixed time interval (sliding every 5 minutes or 10 minutes) or a dynamically adjusted step size. All infrared remote sensing image data within the time window [t0-D, t0] are collected. Within the current sliding time window, the infrared remote sensing image data is traversed, and thermal anomaly pixels are identified using a fire point recognition algorithm. For each identified fire point... For each fire point, its geospatial coordinates, geospatial coordinates, and solar flare angle are extracted. Specifically, the geospatial coordinates of the fire point are determined based on the center point coordinates of the thermal anomaly pixels, and the generation time TG (the specific time when the fire point was identified) is recorded. Considering the influence of solar position and angle on fire point identification, the solar flare angle SG is calculated. All identified fire points and their associated attributes are grouped into a fire point set and stored in the fire point set in chronological order of generation time. The fire generation time is the moment the fire point is generated. Based on the infrared remote sensing image data source, the fire point generation frequency is determined to be Ni times per day. The daily execution frequency M of the task is determined to be no less than the maximum frequency among all data sources. , The frequency of generating corresponding fire points for the i-th type of infrared remote sensing image data source is determined by dynamically adjusting the sliding time window based on the task execution frequency M and the window width D, ensuring that the latest fire point data is acquired at each time point. To ensure that there are enough fire points within the sliding time window to determine the existence or termination of a fire event, the following conditions must be met: ; S2. Use a fire clustering algorithm to aggregate the fire point set into a fire field set. A fire field is a set of fire points that satisfy the nearest neighbor condition at the time a fire point is generated. The nearest neighbor condition is: for any fire point in a fire field, there is at least one other fire point in the same fire field whose spatial distance to that fire point is less than a fire point distance threshold. The fire point distance threshold is greater than... ; Furthermore, the formation process of a fire cluster is as follows: Preprocessing operations are performed on the fire point set, including data cleaning and data standardization. Data cleaning removes duplicate fire point data to ensure each fire point is unique within the set. The integrity of the fire point data is checked, including geospatial coordinates, generation time, solar flare angle, and other auxiliary attributes. Geospatial coordinates are standardized to ensure coordinate system consistency across different data sources. Generation times are formatted for subsequent time series analysis. All fire points are aggregated according to their generation time, generating M*D fire point subsets. For each fire point in a subset, a circular buffer with radius R is generated centered on its coordinates. SR represents the spatial resolution of the infrared remote sensing image. The circular buffers in each subset of fire points are merged to form a fire field set. Each fire field contains auxiliary attributes, including the coordinates of the fire field center point, the fire field boundary polygon, the fire field generation time, the 1 / 4 quantile of the solar flare angle, the number of fire points contained, and the minimum bounding rectangle of the fire field, but not limited to these. Among them, the coordinates of the fire field center point are the geographic center coordinates of the fire field, the fire field boundary polygon is determined by the boundary of the merged buffer and is used to define the boundary of the fire field, the fire field generation time is the earliest time when the fire field was formed, the 1 / 4 quantile of the solar flare angle is the 1 / 4 quantile of the solar flare angle of all fire points in the fire field, which serves as a representative value of the solar flare angle of the fire field, the number of fire points contained is the total number of fire points in the fire field, and the minimum bounding rectangle of the fire field refers to the smallest rectangle that encloses the boundary of the fire field, which is used to simplify the geometric representation and calculation of the fire field. S3. A fire scene filtering method is used to filter the fire scene set, resulting in a filtered fire scene set. Attribute analysis is performed on each fire scene in the set, including auxiliary attributes such as the fire scene center point coordinates, boundary polygon, generation time, 1 / 4 quantile of solar flare angle, number of contained fire points, and minimum bounding rectangle of the fire scene. Solar flare angle is filtered by comparing the 1 / 4 quantile of solar flare angle with a set solar flare angle threshold A. Fire scenes with solar flare angles less than threshold A are marked as "non-fire scenes," which helps to eliminate false fire points caused by solar flares. The system filters for the number of fire points, counts the number of fire points contained in a fire scene, and marks fire scenes with fewer than a set threshold B as "non-fire scenes." This helps to exclude small-scale fire scene sets that may be caused by non-fire factors. By comprehensively analyzing various ancillary attributes and environmental factors of the fire scene, the fire scene is finally assessed and screened to determine whether the fire scene is a real fire event, and a filtered set of fire scenes is output. The filtered set of fire scenes is considered to be a real fire event. S4. In the time dimension, a spatiotemporal aggregation algorithm is used to aggregate the filtered fire scene set to obtain the set of fire events in period T+0. A fire event is a time series set composed of fire scenes that satisfy the spatiotemporal correlation conditions in the time dimension. The spatiotemporal correlation conditions are determined by the specific spatial aggregation algorithm and the temporal aggregation algorithm. The existence condition of a fire event is: if a fire event exists, then within the time interval from the start to the end of the fire event, an arbitrary time window with a width of D0 days is selected, and there are no less than K0 fire scenes within this window; otherwise, the fire event does not exist. The termination condition of a fire event is: in the task of period S1T+0, if the difference between the fire scene generation time of the last fire scene of a certain fire event and the start time t0 of the task is greater than or equal to D0 days, then the fire event is considered to have ended in the task of period T+j. ; S5. The T-1 fire event and the T+0 fire event are spliced ​​together to obtain the updated fire event.

[0020] The second embodiment is based on the first embodiment; please refer to [link / reference]. Figure 1 , Figure 2 As shown, in S4, the process of obtaining the set of fire events in period T+0 is as follows: The filtered fire scene set is processed using a spatial aggregation algorithm to obtain a set of fire scene time series. Each fire scene time series satisfies the following condition: for any filtered fire scene within the series, there exists at least one other filtered fire scene within the same series whose boundary polygon intersects with the boundary polygon of the first filtered fire scene. Each fire scene time series is expanded into a one-dimensional discrete sequence of length D*M according to a sliding time window and generation frequency. In this discrete sequence, each fire scene's corresponding generation time is assigned a value of 1, and the remaining discrete time points are assigned a value of 0. All one-dimensional discrete sequences are traversed and processed to obtain a set of fire events. Each one-dimensional discrete sequence, after processing by the time aggregation algorithm, generates L fire events. The time aggregation algorithm is a density-based clustering algorithm, including DBSCAN, OPTICS, and DPC algorithms, but not limited to these. Each fire event contains auxiliary attributes, including event number, event status, fire fields included in the event, event start time, event end time, cumulative impact range of the event, and event address, but not limited to these. Among them, the event number is unique, the event status has only two types: "ended" and "ongoing", the fire fields included in the event are: all filtered fire fields included in the fire event up to the start time t0 of the task in period S1, the event start time is: the minimum value of the fire field generation time among the fire field attributes of the event, the event end time is: the maximum value of the fire field generation time among the fire field attributes of the event, and this value is less than or equal to the start time t0 of the task in period S1, the cumulative impact range of the event is: the outer boundary of all fire fields included in the event superimposed, and the event address is: the geographical coordinates and corresponding province, city, county, street, etc., corresponding to the center of the cumulative impact range of the event. In S5, the updated process for obtaining fire events is as follows: Search the database storing fire events for fire events in period T-1 based on the following criteria: the event's end time falls within the range of t0-D-D1 to t0-D1. The unit is days; Matching fire events in period T-1 and period T+0, identifying and updating all identical fire events, the specific steps include: Identifying all identical fire events, where an identical fire event satisfies the following conditions: it exists simultaneously in both the set of fire events in period T-1 and the set of fire events in period T+0, and the fire sites contained in the events of both periods intersect. The portion existing in the set of fire events in period T-1 is called the identical fire event in period T-1, and the portion existing in the set of fire events in period T+0 is called the identical fire event in period T+0. Assign the event number of the identical fire event in period T-1 to the identical fire event in period T+0. Identifying all fire events with new fire sites, where a fire event with a new fire site is included in the set of identical fire events and satisfies the following conditions: at least one updated fire site exists in the fire sites contained in the events of the identical fire events in period T+0, but does not exist in the fire sites contained in the events of the identical fire events in period T-1. Traversing the sets of events with new fire sites... The set of fire events is used to update the attributes of the first fire event to be updated in each database. The first fire event to be updated has the same event number as the fire event with new fires. The update operation includes: updating the event status to "ongoing" and updating the event end time to the event end time of the same fire event in period T+0; adding the updated fire to the fires included in the event; recalculating and updating the cumulative impact range of the event based on the fires included in the updated event; finding all fire events without new fires, which are included in the same set of fire events and are the complement of the set of fire events with new fires; traversing the set of fire events without new fires, and updating the attributes of the second fire event to be updated in each database. The second fire event to be updated has the same event number as the fire event without new fires. The update operation is: updating the event status to "ongoing" or "ended" based on the result of the S4 time aggregation algorithm. Match fire events in period T-1 and fire events in period T+0, find all newly added fire events and add them to the database. Among them, the newly added fire events must meet the following conditions: they do not exist in the fire event set in period T-1, and they only exist in the fire event set in period T+0. Iterate through all newly added fire events and add them to the database. Matching fire events in period T-1 and period T+0, identifying and updating all fire events that need to be terminated directly, the specific steps include: identifying all fire events that need to be terminated directly, which must meet the following conditions: exist only in the set of fire events in period T-1 and not in the set of fire events in period T+0; iterating through all fire events that need to be terminated directly; updating the attribute of the third fire event to be updated in each database, where the event number of the third fire event to be updated is the same as that of the fire event that needs to be terminated directly; and updating the event status to "terminated".

[0021] Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art and related fields based on the embodiments of the present invention without inventive effort should fall within the scope of protection of the present invention. Structures, devices, and operating methods not specifically described and explained in the present invention, unless otherwise specified or limited, shall be implemented according to conventional means in the art.

Claims

1. A post-processing method based on fire point identification results from remote sensing images, characterized in that, Includes the following steps: S1. Use the sliding time window method to obtain the set of fire points at the start of the T+0 period task. The fire points include the coordinates, generation time TG, and solar flare angle SG as secondary attributes. S2. Use the fire scene aggregation algorithm to aggregate the fire point set to form a fire scene set; S3. Use the fire scene filtering method to filter the fire scene set to obtain the filtered fire scene set; S4. In the time dimension, the spatiotemporal aggregation algorithm is used to aggregate the filtered fire scene set to obtain the set of fire events in period T+0. S5. The T-1 fire event and the T+0 fire event are spliced ​​together to obtain the updated fire event.

2. The post-processing method based on fire point identification results of remote sensing images according to claim 1, characterized in that: The start time of the T+0 period task is t0, and the width of the sliding time window is D days, that is, the start time of the window is t0-D days, and the end time is t0. The fire point includes auxiliary attributes, including coordinates, generation time (TG), and solar flare angle (SG). The coordinates correspond to the geospatial coordinates of the center point coordinates of the thermal anomaly pixel in the infrared remote sensing image, and the spatial resolution of the infrared remote sensing image is SR.

3. The post-processing method based on fire point identification results from remote sensing images according to claim 2, characterized in that: In S1, the process of obtaining the fire point set is as follows: S1.

1. By setting the start time of the T+0 period task to t0 as the starting point of the current processing cycle, the width of the sliding time window D days is determined, and the step size of each sliding of the time window is determined. S1.2 Collect all infrared remote sensing image data within the time window [t0-D, t0]. Within the current sliding time window, traverse the infrared remote sensing image data and use the fire point recognition algorithm to identify thermal anomaly pixels. S1.3 For each identified fire point, extract its geospatial coordinates, geospatial coordinates and solar flare angle as auxiliary attributes. Among them, the geospatial coordinates of the fire point are determined according to the center point coordinates of the thermal anomaly pixel, and the generation time TG of the fire point is recorded, that is, the specific time when the fire point is identified. Taking into account the influence of the sun's position and angle on the fire point identification, the solar flare angle SG is calculated. S1.

4. All identified fire points and their associated attributes are grouped into a fire point set and stored in the fire point set in order of their generation time. The fire generation time is the moment when the fire point is generated. S1.

5. Based on the infrared remote sensing image data source, determine the fire point generation frequency as Ni times per day, and determine the daily execution frequency M of the task, ensuring that M is not less than the maximum frequency among all data sources, i.e. , Generate the frequency of corresponding fire points for the i-th type of infrared remote sensing image data source; S1.

6. Based on the task execution frequency M and the window width D, dynamically adjust the sliding time window to ensure that the latest fire point data is obtained at each time point.

4. The post-processing method based on fire point identification results from remote sensing images according to claim 3, characterized in that: The fire field is the set of fire points that satisfy the nearest neighbor condition at the time a certain fire point is generated; The nearest neighbor condition is as follows: for any fire point within the fire area, there is at least one other fire point within the same fire area such that its spatial distance from that fire point is less than a fire point distance threshold, wherein the fire point distance threshold is greater than... .

5. The post-processing method based on fire point identification results from remote sensing images according to claim 4, characterized in that: In S2, the formation process of the fire field cluster is as follows: S2.1 Perform preprocessing operations on the fire point set, including data cleaning and data standardization steps; S2.2 Aggregate all fire points according to the generation time of each fire point in the set to generate M*D fire point subsets; S2.

3. For each fire point in the subset of fire points, generate a circular buffer with radius R centered on its coordinates, where... SR represents the spatial resolution of infrared remote sensing images; S2.4 Merge the circular buffers in each fire point subset to form a fire field set. Each fire field contains auxiliary attributes, including the coordinates of the fire field center point, the fire field boundary polygon, the fire field generation time, the 1 / 4 quantile of the solar flare angle, the number of fire points contained within the fire field, and the minimum bounding rectangle of the fire field. Among these, the coordinates of the fire field center point are the geographic center coordinates of the fire field; the fire field boundary polygon is determined by the boundary of the merged buffers and is used to define the boundary of the fire field; the fire field generation time is the earliest time the fire field was formed; the 1 / 4 quantile of the solar flare angle is the 1 / 4 quantile of the solar flare angle of all fire points in the fire field, serving as a representative value of the solar flare angle of the fire field; the number of fire points contained within the fire field is the total number of fire points in the fire field; and the minimum bounding rectangle of the fire field refers to the smallest rectangle enclosing the boundary of the fire field, used to simplify the geometric representation and calculation of the fire field.

6. The post-processing method based on fire point identification results of remote sensing images according to claim 5, characterized in that: In step S3, the process of obtaining the filtered fire scene set is as follows: S3.1 Perform attribute analysis on each fire in the fire set, including fire center point coordinates, boundary polygon, generation time, solar flare angle 1 / 4 quantile, number of fire points, and auxiliary attributes of the minimum bounding rectangle of the fire. S3.2 Filtering solar flare angles: Compare the 1 / 4 quantile of the solar flare angle with a set solar flare angle threshold A. Mark fires with solar flare angles less than threshold A as "non-fires," which helps to eliminate false fire reports caused by solar flares. ; S3.3 Filter the number of fire points. Count the number of fire points contained in a fire scene and mark fire scenes with fewer than a set threshold B as "non-fire scenes". This helps to exclude small-scale fire point sets that may be caused by non-fire factors. ; S3.

4. Conduct a comprehensive analysis of various ancillary attributes and environmental factors of the fire scene, perform a final assessment and screening of the fire scene, determine whether the fire scene is a real fire event, and output a filtered set of fire scenes. The filtered set of fire scenes is considered to be a real fire event.

7. The post-processing method based on fire point identification results of remote sensing images according to claim 6, characterized in that: The fire event is a time series set of fire locations that meet the spatiotemporal correlation conditions in the time dimension. The spatiotemporal correlation conditions are determined by specific spatial aggregation algorithms and time aggregation algorithms. The conditions for the existence of a fire event are as follows: if a fire event exists, then any time window with a width of D0 days is taken within the time interval from the start to the end of the fire event, and there are no less than K0 fire sites within the window; otherwise, the fire event does not exist. The termination condition for the fire event is as follows: In the T+0 period task described in S1, if the difference between the fire generation time of the last fire scene in a certain fire event and the start time t0 of the task is greater than or equal to D0 days, then the fire event is considered to have ended in the T+j period task. .

8. The post-processing method based on fire point identification results of remote sensing images according to claim 7, characterized in that: In S4, the process of obtaining the set of fire events in period T+0 is as follows: S4.

1. Use a spatial aggregation algorithm to process the filtered fire scene set to obtain a fire scene time series set, wherein each fire scene time series satisfies the following condition: for any filtered fire scene in the sequence, there exists at least one other filtered fire scene in the same sequence whose boundary polygon intersects with the boundary polygon of the arbitrary filtered fire scene. S4.

2. Based on the sliding time window and generation frequency, each fire time series is expanded into a one-dimensional discrete sequence of length D*M. In this discrete sequence, the generation time of each fire is assigned a value of 1, and the other discrete time points are assigned a value of 0. S4.

3. Traverse and process all the aforementioned one-dimensional discrete sequences to obtain a set of fire events. Each one-dimensional discrete sequence, after being processed by a time aggregation algorithm, generates L fire events. The time aggregation algorithm is a density-based clustering algorithm, including the DBSCAN algorithm, the OPTICS algorithm, and the DPC algorithm. S4.4 Each fire event includes auxiliary attributes, including event number, event status, fire sites included in the event, event start time, event end time, cumulative impact range of the event, and event address. The event number is unique. The event status has only two options: "Ended" and "Ongoing." The fire sites included in the event are all filtered fire sites included in the fire event up to the start time t0 of the task period mentioned in S1. The event start time is the minimum fire generation time among the fire site attributes of the fire event. The event end time is the maximum fire generation time among the fire site attributes of the fire event, and this value is less than or equal to the start time t0 of the task period mentioned in S1. The cumulative impact range of the event is the outer boundary of all fire sites included in the event superimposed. The event address is the geographical coordinates and corresponding province, city, county, and street information corresponding to the center of the cumulative impact range of the event.

9. The post-processing method based on fire point identification results of remote sensing images according to claim 8, characterized in that: In step S5, the process for obtaining the updated fire event is as follows: S5.1 Query the fire events in period T-1 in the database storing fire events according to the query conditions, wherein the query conditions are: the event end time is within the range of t0-D-D1 to t0-D1, where... The unit is days; S5.2 Match fire events in period T-1 and fire events in period T+0, find all the same fire events and update them; S5.3 Match the fire events of period T-1 and period T+0, find all newly added fire events and add them to the database. Among them, the newly added fire events must meet the following conditions: they do not exist in the fire event set of period T-1, and only exist in the fire event set of period T+0. Iterate through all the newly added fire events and add them to the database. S5.4 Match T-1 and T+0 fire events, identify all fire events that need to be terminated directly, and update them. Specific steps include: S5.

41. Identify all fire events that need to be terminated directly. The fire events that need to be terminated directly must meet the following conditions: they exist only in the fire event set of period T-1 and do not exist in the fire event set of period T+0. S5.

42. Traverse all the fire events that need to be terminated directly, and perform an update operation on the auxiliary attributes of the third fire event to be updated in each database. The third fire event to be updated has the same event number as the fire event that needs to be terminated directly. The update operation is to update the event status to "terminated".

10. The post-processing method based on fire point identification results of remote sensing images according to claim 9, characterized in that: In S5.2, the specific steps for identifying and updating all the same fire events include: S5.

21. Find all the same fire events, wherein the same fire event satisfies the following conditions: it exists in both the set of fire events in period T-1 and the set of fire events in period T+0, and the fire sites contained in the two events have an intersection. The part that exists in the set of fire events in period T-1 is called the same fire event in period T-1, and the part that exists in the set of fire events in period T+0 is called the same fire event in period T+0. Assign the event number of the same fire event in period T-1 to the same fire event in period T+0. S5.

22. Find all fire events with new fire sites, wherein the fire events with new fire sites are included in the set of the same fire events and satisfy the following conditions: at least one updated fire site exists, exists in the fire sites included in the same fire event in period T+0, and does not exist in the fire sites included in the same fire event in period T-1. Traverse the set of fire events with new fire sites, and perform update operations on the subordinate attributes of the first fire event to be updated in each database. The first fire event to be updated has the same event number as the fire events with new fire sites. The update operation includes: updating the event status to "ongoing", updating the event end time to the event end time of the same fire event in period T+0; appending the updated fire site to the fire sites included in the event, and recalculating and updating the cumulative impact range of the event based on the updated fire sites included in the event. S5.

23. Find all fire events without new fire sites, wherein the fire events without new fire sites are included in the set of the same fire events and are the complement of the set of fire events with new fire sites. Traverse the set of fire events without new fire sites and perform update operations on the subordinate attributes of the second fire event to be updated in each database. The second fire event to be updated has the same event number as the fire events without new fire sites. The update operation is: according to the time aggregation algorithm result described in S4, update the event status to "continued" or "ended".

Images