A method for quickly identifying and extracting fire-burned land based on time-series remote sensing vegetation index

Abstract

A kind of fire-burned trace land quick identification extraction method based on time sequence remote sensing vegetation index belongs to the cross field of electronic information technology and resource and environment technology, including the following steps: step one, obtains and pre-processes historical and latest remote sensing image, calculates spectrum index sensitive to fire-burn and constructs time sequence curve;Step two, extracts historical synchronous data, carries out trend fitting using non-parametric Theil-Sens method, obtains the prediction model of each pixel;Step three, calculates the residual between the actual value and predicted value of spectrum index of latest image and previous effective image, and further calculates residual change amount;Step four, carries out threshold binaryzation processing to the residual change amount image of each spectrum index, takes intersection to multi-index binary result and fuses iteration with historical monitoring result, finally determines the range of fire-burned trace land.The present application significantly improves the timeliness and accuracy of fire-burned trace land identification, and can provide effective technical support for forest fire prevention, ecological assessment and emergency management.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of electronic information technology and resource and environmental technology, and relates to a method for rapid identification and extraction of burned areas. It utilizes iteratively updated time-series remote sensing data, calculates the historical trend of surface vegetation changes during the same period, and analyzes the residual changes based on the latest remote sensing data to achieve rapid monitoring of burned areas. Background Technology

[0002] Wildfires are a key driver of global ecosystem dynamics and atmospheric composition changes. They are runaway combustion phenomena occurring on vegetated surfaces, caused by natural factors (such as lightning strikes) or human activities. Satellite remote sensing imagery, through periodic observations of the Earth's surface, acquires electromagnetic wave signals reflected or emitted by objects on the Earth's surface, generating image data. The charred black soil left after vegetation burns has unique reflective characteristics. Based on this characteristic, a time-series remote sensing observation dataset is constructed by arranging the data according to the acquisition time. By investigating and detecting the change points in the time-series curve represented by each pixel, information such as the extent and duration of the fire can be identified, enabling comprehensive monitoring and management of wildfires. Summary of the Invention

[0003] To overcome the shortcomings of existing technologies, this invention provides a rapid identification and extraction method for burned areas based on time-series remote sensing vegetation indices. By integrating multiple time-series spectral indices with trend prediction residual analysis, the timeliness and accuracy of burned area identification are significantly improved, providing effective technical support for forest fire prevention, ecological assessment and emergency management.

[0004] The technical solution adopted by this invention to solve its technical problem is: A rapid identification and extraction method for burned areas based on time-series remote sensing vegetation indices includes the following steps: Step 1: Acquire and preprocess historical and latest remote sensing images, calculate spectral indices sensitive to fire, and construct time-series curves. Step 2: Extract historical data from the same period and use the non-parametric Theil-Sens method to fit trends and obtain the prediction model for each pixel; Step 3: Calculate the residuals between the actual and predicted values ​​of the spectral index of the latest image and the previous effective image, and further calculate the change in residuals; Step 4: Threshold binarization is performed on the residual change images of each spectral index. The intersection of the multi-index binary results is taken and fused with the historical monitoring results for iteration to finally determine the range of the burned area.

[0005] Further, in step one, remote sensing image data is acquired, including historical and recent images. The image data is preprocessed, and spectral indices sensitive to the burned area are calculated from the preprocessed long-term remote sensing image data. These spectral indices include MIRBI, NBR2, and SRI (SWIR1 / SWIR2). The resulting vegetation index time-series curve for each pixel is recorded as follows: spectral indices in pixels The exponent value at time t ,in, Number the spectral index type. t represents pixel coordinates and t represents the timestamp.

[0006] Furthermore, in step two, for the preprocessed latest remote sensing image, historical images from the same period and the current time are extracted based on the image's acquisition time. same month The spectral index time series data for one month, preprocessed image data, and observations from the previous two months for each pixel in the current image were used to fit the pixels using the nonparametric Theil-Sens (TS) method. Trend fitting was performed on the historical time series data of the k-th spectral index to obtain the fitting slope. and fitting intercept The fitting formula is: ; in, For pixels First spectral indices in time The predicted value, where t is a relative quantity with respect to a given start time.

[0007] Furthermore, in step three, the latest remote sensing image (time) is acquired. ) and previous effective remote sensing images (time) Earlier than (And no cloud cover), calculate pixels First The residuals between the actual and predicted values ​​of each spectral index: ; ; in, For pixels First Spectral indices at the current time The actual observed value; Pixels obtained based on historical time series trend models First spectral indices in time The predicted value; For pixels First Each spectral index in the early effective time The actual observed value; For pixels obtained based on the trend model First spectral indices in time The predicted value; For pixels First The residual value of each spectral index at the current time; For pixels First The residual values ​​of each spectral index during the early effective time period; For pixels First The residual change of a spectral index between the current time and the previous effective time is used to characterize the degree of abnormal change of the pixel's spectral features relative to the historical trend. The residual difference image is composed of the residual changes, wherein Indicates the type of spectral index. This represents the time corresponding to the residual difference image. Then, the change in residual is calculated: ; The residual difference image is defined as follows: , k represents the k-th index, and t represents the time of the current residual image.

[0008] Furthermore, in step four, for each spectral index... ,Will According to preset threshold Perform binarization; take the intersection of the binary results of all spectral indices, i.e. The newly appearing burned area in the current image is obtained, and the newly appearing burned area is merged with the burned area obtained from historical monitoring until the area range no longer increases, and the final fire site identification result is obtained.

[0009] Preferably, the binarization process is as follows: when The pixel was determined to be burned and recorded as follows: Otherwise, it is 0. It is a universal threshold determined based on expert experience.

[0010] The beneficial effects of this invention are mainly reflected in the following aspects: by integrating multi-time series spectral indices and trend prediction residual analysis, the timeliness and accuracy of fire-marked area identification are significantly improved, which can provide effective technical support for forest fire prevention, ecological assessment and emergency management. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of a method for rapid identification and extraction of burned areas based on time-series remote sensing vegetation indices.

[0012] Figure 2 HLS data coverage band data map.

[0013] Figure 3 This is a composite image of visible light before and after the fire in the burned area.

[0014] Figure 4 Fit the time series curves and trends of historical remote sensing data from the same period.

[0015] Figure 5 This is a time series of spectral indices from the first image after the wildfire and previous observations.

[0016] Figure 6 The time series of spectral index residuals is calculated by fitting the historical remote sensing observation trend of opportunities.

[0017] Figure 7 The residual difference of the burned area image.

[0018] Figure 8 This is a schematic diagram illustrating the final extraction effect of the burned area.

[0019] Figure 9 This is a schematic diagram of a rapid identification and extraction system for burned areas based on time-series remote sensing vegetation indices.

[0020] Figure 10 This is a schematic diagram of an electronic device provided according to an embodiment of the present invention. Detailed Implementation

[0021] The present invention will now be further described with reference to the accompanying drawings.

[0022] Reference Figures 1-8 A method for rapid identification and extraction of burned areas based on time-series remote sensing vegetation indices includes the following steps: Step 1: Data Acquisition and Preprocessing: Acquire and preprocess historical and latest remote sensing images, calculate spectral indices sensitive to fire, and construct time-series curves. The process is as follows: Remote sensing image data, including historical and recent images, is acquired. The image data is preprocessed, and spectral indices sensitive to burned areas are calculated from the preprocessed long-term remote sensing image data. These spectral indices include MIRBI, NBR2, and SRI (SWIR1 / SWIR2). The resulting vegetation index time-series curve for each pixel is denoted as the... spectral indices in pixels The exponent value at time t ,in, Number the spectral index type. t represents pixel coordinates and t represents the timestamp.

[0023] Step 2: Historical Trend Fitting: Extract historical data from the same period and use the non-parametric Theil-Sens method to fit the trend, obtaining the prediction model for each pixel. The process is as follows: For the latest preprocessed remote sensing image, extract historical images from the same period and the current time based on the image's acquisition time. same month The spectral index time series data for one month, preprocessed image data, and observations from the previous two months for each pixel in the current image were used to fit the pixels using the nonparametric Theil-Sens (TS) method. Trend fitting was performed on the historical time series data of the k-th spectral index to obtain the fitting slope. and fitting intercept The fitting formula is: ; in, For pixels First spectral indices in time The predicted value, where t is a relative quantity with respect to a given start time.

[0024] Step 3: Residual Calculation and Variation Analysis: Calculate the residuals between the actual and predicted values ​​of the spectral indices of the latest image and previous effective images, and further calculate the variation in the residuals. The process is as follows: Acquire the latest remote sensing imagery (time) ) and previous effective remote sensing images (time) Earlier than (And no cloud cover), calculate pixels The residual between the actual and predicted values ​​of the k-th spectral index: ; ; in, For pixels First Spectral indices at the current time The actual observed value; Pixels obtained based on historical time series trend models First spectral indices in time The predicted value; For pixels First Each spectral index in the early effective time The actual observed value; For pixels obtained based on the trend model First spectral indices in time The predicted value; For pixels First The residual value of each spectral index at the current time; For pixels First The residual values ​​of each spectral index during the early effective time period; For pixels First The residual change of a spectral index between the current time and the previous effective time is used to characterize the degree of abnormal change of the pixel's spectral features relative to the historical trend. The residual difference image is composed of the residual changes, wherein Indicates the type of spectral index. This indicates the time corresponding to the residual difference image.

[0025] Then calculate the change in residuals: ; The residual difference image is defined as follows: , k represents the k-th index, and t represents the time of the current residual image.

[0026] Step 4: Region Extraction and Result Fusion: Threshold binarization is performed on the residual change images of each spectral index. The intersection of the multi-index binary results is taken and iteratively fused with historical monitoring results to finally determine the extent of the burned area. The process is as follows: For each spectral index ,Will According to preset threshold Perform binarization; take the intersection of the binary results of all spectral indices, i.e. The newly appearing burned area in the current image is obtained, and the newly appearing burned area is merged with the burned area obtained from historical monitoring until the area range no longer increases, and the final fire trace identification result is obtained. The binarization process is as follows: when It was determined at the time that the pixel was burned. It is a universal threshold determined based on expert experience; denoted as Otherwise, it is 0.

[0027] The implementation process of the application example of this invention is as follows: Step S1: Acquire remote sensing images of the wildfire area and perform preprocessing, as follows: 1.1 Rapid identification and extraction of burned areas were achieved using coordinated Landsat and Sentinel-2 (HLS) remote sensing imagery data. HLS, developed and maintained by NASA's Landsat Science Team, aims to integrate Landsat-8 / 9 and Sentinel-2A / 2B observational data. By unifying and coordinating the original Landsat OLI / TIRS and Sentinel-2 MSI data, two products, HLSL30 (based on Landsat) and HLSS30 (based on Sentinel-2), were generated, both providing surface reflectance data with a spatial resolution of 30 meters. Through preprocessing steps such as spectral response function matching, bandpass adjustment, resampling, and bidirectional reflectance distribution function (BRDF) normalization, HLS achieved radiometric consistency among multiple sensor sources. The band information of the HLS data is as follows: Figure 2 As shown.

[0028] 1.2. Perform remote sensing image preprocessing: Use the Fmask band provided by the HLS dataset to mask the clouds and their shadow-covered areas in the remote sensing images to remove invalid observations and noise in the images, so as to improve the reliability of subsequent analysis; merge the HLSL30 and HLSS30 datasets after cloud removal processing, and sort them in order according to the timestamp t to construct the remote sensing image time series.

[0029] 1.3 Spectral Index Calculation: In this embodiment, three indices—MIRBI, NBR2, and SRI—are selected to identify and extract burned areas. The calculation formulas for these three spectral indices are as follows: ; ; ; in, The reflectance value is the first band of shortwave infrared, used to characterize the spectral response characteristics of target ground objects in the first band of shortwave infrared. The reflectance value is the reflectance value of the second band of shortwave infrared, used to characterize the spectral response characteristics of the target area under the second band of shortwave infrared. The reflectance value is the surface reflectance after radiometric calibration and atmospheric correction, and its value ranges from 0 to 1. MIRBI is a feature index constructed based on the shortwave infrared band, used to characterize the comprehensive spectral variation characteristics of the target area under the shortwave infrared band. NBR2 is a feature index constructed by the normalized ratio relationship between the first band and the second band of shortwave infrared, used to reduce the influence of changes in illumination conditions on spectral information. SRI (SWIR1 / SWIR2) is a feature index constructed by the ratio relationship between the first band and the second band of shortwave infrared, used to enhance the differences between different land cover types in the shortwave infrared band.

[0030] Step S2: Vegetation exhibits similar phenological stages within the same time frame across different years. Burnt areas are located using historical modeling and near-real-time anomaly detection. The process is as follows: 2.1 Based on the acquisition time of the latest remote sensing image, extract the temporal trend of spectral index in historical images from the same period (such as the previous year).

[0031] Surface vegetation growth often exhibits cyclical patterns. In the absence of disturbance, the current time and historical spectral index time series should show similar trends. By modeling historical trends, we can obtain the potential trends of land cover changes at the current stage.

[0032] In trend analysis, the first step is to extract historical images from the same period and compare them with the current time. same month The dataset consists of 2 months of spectral index time-series data, preprocessed image data, and observations from the previous two months for each pixel in the current image. A nonparametric Theil-Sens (TS) fitting method is used to fit the pixels. Trend fitting was performed on the historical time series data of the k-th spectral index to obtain the fitting slope. and fitting intercept The fitting formula is: ; in, For pixels First spectral indices in time The predicted value; such as Figure 2 , Figure 3 The image shows the spectral index variation trend of the currently burned area after a wildfire, visualized using a composite visible light image, compared to historical images from the same period. TS fitting is a nonparametric fitting method insensitive to outliers. It works by selecting any two "time-observation" pairs from a given time series, assuming... and By calculating the slope of the straight line determined by these two sets of data points, we can initially capture the trend of data change between these two time points. The formula is as follows: ; in, For pixels First The slope of the trend change of a spectral index is used to characterize the direction and rate of change of the spectral index of the pixel over time in the historical time series. For pixels First spectral indices in time Observed values; For pixels First spectral indices in time Observed values; and These are the observation times of the corresponding spectral indices, and satisfy the following conditions: The slope is calculated by analyzing any two sets of "time-spectral index observations" in the historical time series, and is used to reflect the changing trend of the spectral index at different time scales.

[0033] Because time series data contains numerous data points, to comprehensively and accurately describe the overall trend, it is necessary to traverse all possible combinations of "time-observation" pairs and calculate the corresponding slopes. Subsequently, these slopes are statistically analyzed, typically using the median as a representative value to determine the overall trend slope of the time series. For example, if there are n different combinations of "time-observation" pairs, the calculated... slope values The final determined trend slope For this The median of the slope values. Simultaneously, combining all data points, the fit intercept is calculated using methods such as linear regression. Thus, the complete fitting equation is obtained.

[0034] 2.2 Compare the latest image observations with the previous valid observations, and extract the burned areas according to the given threshold; In areas where vegetation cover has been burned by wildfires, the leaf area index decreases due to leaf burning, specifically reflected in the ground reflectance signal as a significant reduction in near-infrared (NIR) reflectance. Simultaneously, the reduction in black residue and canopy moisture after the fire manifests as an increase in short-wave infrared (SWIR) reflectance. Therefore, the spectral indices composed of these bands show abnormal increases or decreases compared to healthy vegetation areas, and gradually recover over a period of time.

[0035] For acquiring the latest remote sensing imagery (time) ) and previous effective remote sensing images (time) Earlier than And there is no cloud cover), targeting each pixel in the image. We focus on the first Analyze each spectral index. (Pixel...) In time The actual observed value of the k-th spectral index is denoted as Based on the prediction model built from historical data from the same period, the predicted value of the pixel at the same time and with the same spectral index was obtained. Subtracting the two values ​​gives the residual value at this point. It can be expressed by the formula: ; The same principle applies to time. , pixels First The actual value of each spectral index is The corresponding predicted value is The corresponding residual value is: ; Further calculation of residual change reveals the characteristics of spectral index variation over time. Time and Pixels at any given time , No. The residuals of each spectral index are subtracted to obtain the change in residual. The calculation formula is: ; Define the SI residual difference image before and after the fire as ,in Indicates the first Different spectral indices have varying sensitivities and characteristic information in reflecting the impact of surface vegetation and wildfires. This indicates the time corresponding to the current residual image. The SI residual difference images before and after the fire from MIRBI, NBR2, and SRI are respectively... , and .

[0036] For the residual differences before and after the fire at three different indices, respectively, use... , and Three thresholds are used to binarize the residual difference image, and the final range of the burned area is obtained by taking the intersection. It should be noted that the thresholds given here are obtained by fitting a Gaussian mixture model to samples from burned and normal areas in different vegetation cover zones. The specific results of burning area identification and extraction will vary in different regions and can be improved by regional adaptive adjustment.

[0037] Step S3: Implement rapid detection of burned areas based on the changing trend of vegetation index in time series remote sensing. The process is as follows: 3.1. Using historical remote sensing image time series for modeling and anomaly detection in current remote sensing observations can effectively improve the timeliness of fire site detection. The intersection of multiple index detection results can effectively increase the reliability of the extraction results. 3.2 The rapid detection system for burned areas updates the burned area through time-series iterations and visualizes the final detection results through a GIS platform, which can help decision-makers quickly calculate ecological and economic losses.

[0038] This embodiment also provides a rapid identification and extraction system for burned areas based on time-series remote sensing vegetation indices, such as... Figure 7 As shown, the system used to implement the above-mentioned method for rapid identification and extraction of burned areas based on time-series remote sensing vegetation indices includes: The time series construction module is used to acquire long-term time series remote sensing images, calculate multispectral indices, and construct time series. The trend fitting module is used to extract historical data from the same period, analyze its changing trends, and record the fitting parameters. The residual analysis module is used to calculate the residuals and residual changes between the latest image and the previous valid image; The region extraction module is used to binarize the residual change image according to a priori threshold, take the intersection of multiple exponents, and fuse it with historical results to obtain the final burned region.

[0039] This embodiment also provides a computing device, including a processor and a memory; the memory stores computer execution instructions; when the processor calls the computer execution instructions, it executes the method for rapid identification and extraction of burned areas based on time-series remote sensing vegetation indices.

[0040] The embodiments described in this specification are merely examples of implementations of the inventive concept and are for illustrative purposes only. The scope of protection of this invention should not be considered limited to the specific forms described in these embodiments; rather, it extends to equivalent technical means conceived by those skilled in the art based on the inventive concept.

Claims

1. A method for rapid identification and extraction of burned areas based on time-series remote sensing vegetation indices, characterized in that, The method includes the following steps: Step 1: Acquire and preprocess historical and latest remote sensing images, calculate spectral indices sensitive to fire, and construct time-series curves; Step 2: Extract historical data from the same period and use the non-parametric Theil-Sens fitting method to fit the trend and obtain the prediction model for each pixel. Step 3: Calculate the residuals between the actual and predicted values ​​of the spectral index of the latest image and the previous effective image, and further calculate the change in residuals; Step 4: Threshold binarization is performed on the residual change images of each spectral index. The intersection of the multi-index binary results is taken and fused with the historical monitoring results for iteration to finally determine the range of the burned area.

2. The method for rapid identification and extraction of burned areas based on time-series remote sensing vegetation indices as described in claim 1, characterized in that, In step one, remote sensing image data is acquired, including historical and recent images. The image data is then preprocessed, and spectral indices sensitive to the burned areas are calculated from the preprocessed long-term remote sensing image data. These spectral indices include MIRBI, NBR2, and SRI. The resulting vegetation index time-series curve for each pixel is recorded as follows: spectral indices in pixels The exponent value at time t is ,in, Number the spectral index type. t represents pixel coordinates, and t represents the timestamp.

3. The method for rapid identification and extraction of burned areas based on time-series remote sensing vegetation indices as described in claim 1, characterized in that, In step two, for the preprocessed latest remote sensing image, historical images from the same period and the current time are extracted based on the image's acquisition time. same month The spectral index time series data for one month, preprocessed image data, and observations from the previous two months for each pixel in the current image were used to fit the pixels using a nonparametric TS fitting method. Trend fitting was performed on the historical time series data of the k-th spectral index to obtain the fitting slope. and fitting intercept .

4. The method for rapid identification and extraction of burned areas based on time-series remote sensing vegetation indices as described in claim 1, characterized in that, In step three, the latest remote sensing image and previously valid remote sensing image are acquired, and the pixel count is calculated. The residual between the actual value and the predicted value of the k-th spectral index is calculated, and then the change in residual is calculated.

5. The method for rapid identification and extraction of burned areas based on time-series remote sensing vegetation indices as described in claim 1, characterized in that, In step four, for each spectral index ,Will According to preset threshold Perform binarization; take the intersection of the binary results of all spectral indices, i.e. The newly appearing burned area in the current image is obtained, and the newly appearing burned area is merged with the burned area obtained from historical monitoring until the area range no longer increases, and the final fire site identification result is obtained.

6. The method for rapid identification and extraction of burned areas based on time-series remote sensing vegetation indices as described in claim 5, characterized in that, The binarization process is as follows: when The pixel was determined to be burned and recorded as follows: Otherwise, it is 0. It is a universal threshold determined based on expert experience.

Images