A method for extracting fire-burned land based on a cloud computing platform and a random forest algorithm

Abstract

This invention relates to the technical field of burned area extraction methods, specifically to a burned area extraction method based on a cloud computing platform and a random forest algorithm, comprising the following steps: Step 1) Remote sensing data acquisition, preprocessing, and construction of a time-series image dataset; Step 2) Construction of a burned area sample library; Step 3) Segmentation of sample point data; Step 4) Feature vector extraction and establishment of a time-series feature vector image set; Step 5) Construction and optimization of a random forest model; Step 6) Accuracy evaluation of burned area extraction results; Step 7) Burned area mapping. This invention applies the random forest algorithm to the burned area extraction method, achieving year-by-year, high-precision, and large-scale burned area extraction. It enables dynamic monitoring of burned areas over time, allowing for rapid acquisition of high-quality synthetic remote sensing data, significantly improving user research and work efficiency, and ultimately achieving rapid and automatic extraction of large-scale, long-term burned areas.

Description

Technical Field

[0001] This invention relates to the technical field of methods for extracting burned areas, specifically to a method for extracting burned areas based on a cloud computing platform and a random forest algorithm. Background Technology

[0002] Forest fires are a major type of forest disturbance, directly altering the community composition, age structure, energy flow, and nutrient cycling of forest ecosystems, thus significantly impacting regional carbon cycles, biodiversity, and global climate change. Therefore, accurately and quickly identifying the extent of fires and extracting burned areas provides crucial decision-making support for post-fire recovery strategies, forest quality improvement, and sustainable development. Traditional field survey methods are heavily influenced by weather, topography, and other environmental conditions, and consume significant time and effort from researchers. With the rapid development of remote sensing technology and the increasing availability of data sources, fire disturbance information extraction has progressed from traditional single-temporal to multi-temporal methods, and from single-source data to multi-source data fusion. Using remote sensing technology for mapping and analyzing fire-affected areas allows for efficient and rapid assessment of post-disaster losses and continuous monitoring and evaluation of vegetation recovery on a large scale. However, current methods for extracting burned areas still suffer from inaccurate data, low precision, and small scale. Summary of the Invention

[0003] To address the aforementioned problems, this invention proposes a method for extracting burned areas based on a cloud computing platform and a random forest algorithm. Using long-term, year-by-year Landsat imagery and field survey data, the random forest algorithm is applied to the burned area extraction method, achieving year-by-year, high-precision, and large-scale extraction of burned areas. It enables dynamic monitoring of burned areas over time and, based on the Google Earth Engine cloud computing platform, allows for the rapid acquisition of high-quality synthetic remote sensing data, significantly improving user research and work efficiency. Ultimately, it achieves rapid and automatic extraction of large-scale, long-term burned areas.

[0004] To achieve the above objectives, the present invention adopts the following technical solution:

[0005] A method for extracting burned areas based on a cloud computing platform and a random forest algorithm includes the following steps:

[0006] Step 1) Remote sensing data acquisition, preprocessing, and construction of time-series image datasets;

[0007] Step 2) Construct a sample library of burned areas;

[0008] Step 3) Divide the sample data;

[0009] Step 4) Feature vector extraction and establishment of a time-series feature vector image set;

[0010] Step 5) Construct a random forest model and optimize the model;

[0011] Step 6) Evaluation of the accuracy of the results of the fire-damaged area extraction;

[0012] Step 7) Mapping the burned area.

[0013] Preferably, the specific operations for remote sensing data acquisition and preprocessing in step 1) are as follows:

[0014] S1. Based on the Google Earth Engine platform, acquire Landsat series imagery, including Landsat5, Landsat7 and Landsat8 surface reflectance imagery products.

[0015] S2. The acquired images are processed as follows: invalid observations such as clouds and shadows are removed from the images; terrain shadows are removed from the images; strip errors in the Landsat 7 images are removed; and the reflectance of the OLI and TM / ETM+ sensors is normalized using a statistical coordination function between spectral values.

[0016] Preferably, the specific operation in step 1) to construct the time series image dataset is as follows: based on the median image synthesis method provided by the Google Earth Engine platform, high-quality images are synthesized year by year from the preprocessed image data to obtain a long-term high-quality image dataset.

[0017] Preferably, step 2) specifically involves: based on the Google Earth Engine cloud platform, according to the time series image dataset obtained in step 1), manually selecting sample points for the burned areas, and merging them with field survey sample points to generate a burned area sample library.

[0018] Preferably, step 3) specifically involves: dividing the burned area sample database obtained in step 2) into a training dataset and a validation dataset; randomly allocating all sample data to one of the two datasets in a 70%:30% ratio; subsetting the training dataset, where 80% of the data is used to train a separate classification algorithm and 20% of the data is used to optimize the model; and using the validation dataset to evaluate the classification results of the subsequent final sample data.

[0019] Preferably, step 4) specifically involves: based on the Google Earth Engine platform, calculating the feature vector of the image, (1) spectral index feature vector, normalized normalized vegetation index NDVI, enhanced vegetation index EVI, area burned index BAI, improved normalized normalized water index MNDWI, normalized fire index NBR and normalized fire index NBR2; (2) KT transform main components, brightness TCB, greenness TCG and humidity TCW; finally, a time series feature vector dataset is established, and the calculation formulas for the relevant indices are as follows:

[0020] NDVI = (NIR - RED) / (NIR + RED)

[0021] EVI=2.5*(NIR-RED) / (NIR+6RED-7.5BLUE+1)

[0022] BAI = 1 / [(0.1-RED)] 2 +(0.06-NIR) 2 ]

[0023] MNDWI=(GREEN-SWIR2) / (GREEN+SWIR2)

[0024] NBR = (NIR - SWIR²) / (NIR + SWIR²)

[0025] NBR2=(SWIR1-SWIR2) / (SWIR1+SWIR2)

[0026] TCB=(0.3029*BLUE)+(0.2786*GREEN)+(0.4733*RED)

[0027] +(0.5599*NIR)+(0.508*SWIR1)+(0.1872*SWIR2)

[0028] TCG=(-0.2941*BLUE)-(0.243*GREEN)-(0.5424*RED)

[0029] +(0.7276*NIR)+(0.0713*SWIR1)-(0.1608*SWIR2)

[0030] TCW=(0.1511*BLUE)+(0.1973*GREEN)+(0.3283*RED)

[0031] +(0.3407*NIR)-(0.7117*SWIR1)-(0.4559*SWIR2)

[0032] Preferably, step 5) specifically involves: building a random forest model on the Google Earth Engine cloud platform, setting the number of growth trees and the number of feature variables input when splitting nodes, adjusting the settings of the above two parameters by calculating the OOB misclassification rate to obtain the optimal model, and performing supervised classification based on the results of the 9 feature vectors in step 4 to obtain the spatial distribution information of the burned areas.

[0033] Preferably, step 6) specifically involves: using the validation dataset from step 3) to evaluate the accuracy of the model classification results, and assessing the accuracy of the fire-affected area extraction results by calculating user accuracy, producer accuracy, overall accuracy, and the Kappa coefficient.

[0034] Preferably, step 7) specifically involves downloading the output results of the random forest model to the local processor and importing them into ArcGIS for mapping processing.

[0035] Due to the adoption of the above technical solution, the beneficial effects of the present invention are as follows: Based on long-term, year-by-year Landsat image data and field survey sample data, the present invention applies the random forest algorithm to the method of fire scar extraction, realizing year-by-year, high-precision, and large-scale fire scar extraction, dynamically monitoring fire scars over time, and simultaneously realizing the rapid acquisition of high-quality synthetic remote sensing data based on the Google Earth Engine cloud computing platform, significantly improving the research and work efficiency of users, and ultimately achieving rapid and automatic extraction of large-scale, long-term fire scars. Attached Figure Description

[0036] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only for the present invention and protect some embodiments. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 This is a flowchart of the present invention;

[0038] Figure 2 This is a spatial distribution map of the burned area in Embodiment 2 of the present invention. Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. 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.

[0040] Example 1

[0041] like Figures 1-2 As shown, a method for extracting burned areas based on a cloud computing platform and a random forest algorithm includes the following steps:

[0042] Step 1) Remote sensing data acquisition, preprocessing, and construction of time-series image datasets:

[0043] S1. Based on the Google Earth Engine platform, acquire Landsat series imagery, including Landsat5, Landsat7 and Landsat8 surface reflectance imagery products.

[0044] S2. The acquired images are processed as follows: invalid observations such as clouds and shadows are removed from the images; terrain shadows are removed from the images; strip errors in the Landsat 7 images are removed; and the reflectance of the OLI and TM / ETM+ sensors is normalized using a statistical compatibility function between spectral values.

[0045] S3. Constructing a Time-Series Image Dataset: Based on the median image synthesis method provided by the Google Earth Engine platform, high-quality images are synthesized year by year from the preprocessed image data to obtain a long-term high-quality image dataset.

[0046] Step 2) Construct a sample library of burned areas: Based on the Google Earth Engine cloud platform, according to the time series image dataset obtained in Step 1), manually select sample points for burned areas and merge them with field survey sample points to generate a sample library of burned areas.

[0047] Step 3) Divide the sample data: Divide the burned area sample database obtained in Step 2) into a training dataset and a validation dataset; randomly allocate all sample data to one of the two datasets in a ratio of 70%:30%; subset the training dataset, where 80% of the data is used to train a separate classification algorithm and 20% of the data is used to optimize the model; the validation dataset is used to evaluate the classification results of the final sample data.

[0048] Step 4) Feature vector extraction and establishment of time series feature vector image set: Based on the Google Earth Engine platform, the feature vectors of the images are calculated, (1) spectral index feature vectors, normalized difference vegetation index (NDVI), enhanced vegetation index (EVI), area of ​​fire index (BAI), improved normalized water index (MNDWI), normalized fire index (NBR) and normalized fire index (NBR2); (2) main components of KT transform, brightness (TCB), greenness (TCG) and humidity (TCW); Finally, a time series feature vector dataset is established, and the calculation formulas of the relevant indices are as follows:

[0049] NDVI = (NIR - RED) / (NIR + RED)

[0050] EVI=2.5*(NIR-RED) / (NIR+6RED-7.5BLUE+1)

[0051] BAI = 1 / [(0.1-RED)] 2 +(0.06-NIR) 2 ]

[0052] MNDWI=(GREEN-SWIR2) / (GREEN+SWIR2)

[0053] NBR = (NIR - SWIR²) / (NIR + SWIR²)

[0054] NBR2=(SWIR1-SWIR2) / (SWIR1+SWIR2)

[0055] TCB=(0.3029*BLUE)+(0.2786*GREEN)+(0.4733*RED)

[0056] +(0.5599*NIR)+(0.508*SWIR1)+(0.1872*SWIR2)

[0057] TCG=(-0.2941*BLUE)-(0.243*GREEN)-(0.5424*RED)

[0058] +(0.7276*NIR)+(0.0713*SWIR1)-(0.1608*SWIR2)

[0059] TCW=(0.1511*BLUE)+(0.1973*GREEN)+(0.3283*RED)

[0060] +(0.3407*NIR)-(0.7117*SWIR1)-(0.4559*SWIR2)

[0061] Step 5) Construct and optimize the random forest model: Construct a random forest model on the Google Earth Engine cloud platform, set the number of growing trees and the number of feature variables input when splitting nodes, adjust the settings of the above two parameters by calculating the OOB misclassification rate, obtain the optimal model, and perform supervised classification based on the results of the 9 feature vectors in Step 4 to obtain the spatial distribution information of the burned areas.

[0062] Step 6) Accuracy evaluation of burned area extraction results: The accuracy of the model classification results is evaluated using the validation dataset in Step 3). The accuracy of the burned area extraction results is evaluated by calculating user accuracy, producer accuracy, overall accuracy, and Kappa coefficient.

[0063] Step 7) Mapping of burned areas: Download the output results of the random forest model to the local processor and import them into ArcGIS for mapping processing.

[0064] This invention, based on long-term, year-by-year Landsat imagery and field survey data, applies the random forest algorithm to the method of fire scar extraction, achieving year-by-year, high-precision, and large-scale fire scar extraction. It enables dynamic monitoring of fire scars over time and utilizes the Google Earth Engine cloud computing platform to quickly acquire high-quality synthetic remote sensing data, significantly improving users' research and work efficiency. Ultimately, it achieves rapid and automatic extraction of large-scale, long-term fire scars.

[0065] Example 2

[0066] The spatial distribution information of burned areas in Mohe City and Tahe County from 1986 to 2020 was extracted using the method in Example 1, including the following steps:

[0067] Step 1) Remote sensing data acquisition:

[0068] Based on the Google Earth Engine platform, Landsat series images from 1986 to 2020 were acquired, including Landsat5, Landsat7, and Landsat8 surface reflectance image products. These images were then retrieved on the GEE cloud platform according to the image set IDs “LANDSAT / LT05 / C01 / T1_SR”, “LANDSAT / LE07 / C01 / T1_SR”, and “LANDSAT / LC08 / C01 / T1_SR”, resulting in a total of 2,333 images.

[0069] Step 2) Image preprocessing:

[0070] The acquired images were processed as follows: 1. Invalid observations such as clouds and shadows were removed from the images using the QA band; 2. Topographic shadows were removed from the images using the slope normalization method; 3. Strip errors in Landsat 7 images were removed using the focus statistics method; 4. The reflectance of OLI and TM / ETM+ sensors was normalized using the statistical compatibility function between spectral values.

[0071] Step 3) Construct a time-series image dataset:

[0072] Based on the Google Earth Engine platform, the ee.ImageCollection.median and ee.ImageCollection.clip commands are used to perform median synthesis on the preprocessed image data and crop it according to the study area to obtain a high-quality long-term series image dataset.

[0073] Step 4) Construct a sample library of burned areas:

[0074] Based on the Google Earth Engine cloud platform, according to the time series image dataset obtained in step 1), artificial sample points are selected for the burned areas and merged with the field survey sample points to generate a burned area sample library.

[0075] Step 5) Divide the sample data:

[0076] The sample database of burned areas was divided into two datasets: a training dataset and a validation dataset. All sample data were randomly allocated to one of the two datasets in a 70%:30% ratio. The training dataset was subsetted, with 80% of the data used to train a separate classification algorithm and 20% used to optimize the model. The validation dataset was used to evaluate the classification results of the final sample data.

[0077] Step 6) Feature vector extraction and establishment of a time-series feature vector image set:

[0078] Based on the Google Earth Engine platform, the feature vectors of the images are calculated, mainly including: 1. Spectral index feature vectors: Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), Burn Area Index (BAI), Modified Normalized Difference Water Index (MNDWI), Normalized Burn Index (NBR), and Normalized Burn Index (NBR2); 2. Key components of the KT transform: Luminance (TCB), Greenness (TCG), and Humidity (TCW). Finally, a time-series feature vector dataset is established, and the calculation formulas for the relevant indices are as follows:

[0079] NDVI = (NIR - RED) / (NIR + RED)

[0080] EVI=2.5*(NIR-RED) / (NIR+6RED-7.5BLUE+1)

[0081] BAI = 1 / [(0.1-RED)] 2 +(0.06-NIR) 2 ]

[0082] MNDWI=(GREEN-SWIR2) / (GREEN+SWIR2)

[0083] NBR = (NIR - SWIR²) / (NIR + SWIR²)

[0084] NBR2=(SWIR1-SWIR2) / (SWIR1+SWIR2)

[0085] TCB=(0.3029*BLUE)+(0.2786*GREEN)+(0.4733*RED)

[0086] +(0.5599*NIR)+(0.508*SWIR1)+(0.1872*SWIR2)

[0087] TCG=(-0.2941*BLUE)-(0.243*GREEN)-(0.5424*RED)

[0088] +(0.7276*NIR)+(0.0713*SWIR1)-(0.1608*SWIR2)

[0089] TCW=(0.1511*BLUE)+(0.1973*GREEN)+(0.3283*RED)

[0090] +(0.3407*NIR)-(0.7117*SWIR1)-(0.4559*SWIR2)

[0091] Step 7) Build and optimize the random forest model:

[0092] A random forest model was built on the Google Earth Engine cloud platform. The number of growth trees and the number of feature variables input when splitting nodes were set. The settings of the above two parameters were adjusted by calculating the OOB misclassification rate. The optimal model was obtained when the number of growth trees and the number of feature variables input when splitting nodes were set to 500 and 6 respectively. Supervised classification was performed based on the results of the 9 feature vectors in step 4) to obtain the spatial distribution information of the burned areas.

[0093] Step 8) Evaluation of the accuracy of the results of the fire-damaged area extraction:

[0094] The accuracy of the model classification results was evaluated using the validation dataset. The calculated user accuracy, producer accuracy, overall accuracy, and Kappa coefficient were 94.50%, 88.79%, 90.50%, and 0.81, respectively.

[0095] Step 9) Mapping the burned area:

[0096] The output of the random forest model was downloaded to the local processor and imported into ArcGIS for mapping processing, thereby obtaining the spatial distribution information of burned areas in Mohe City and Tahe County from 1986 to 2020.

[0097] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0098] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for extracting burned areas based on a cloud computing platform and a random forest algorithm, characterized in that, Includes the following steps: Step 1) Remote sensing data acquisition, preprocessing, and construction of time-series image datasets; Step 2) Construct a sample library of burned areas; Step 3) Divide the sample data; Step 4) Feature vector extraction and establishment of a time-series feature vector image set; Step 5) Construct a random forest model and optimize the model; Step 6) Evaluation of the accuracy of the results of the fire-damaged area extraction; Step 7) Mapping the burned area; The specific operation in step 1) to construct the time series image dataset is as follows: Based on the median image synthesis method provided by the Google Earth Engine platform, high-quality images are synthesized year by year from the preprocessed image data to obtain a long-term high-quality image dataset. Step 2) The specific operation is as follows: Based on the Google Earth Engine cloud platform, according to the time series image dataset obtained in Step 1), manually select sample points for the burned areas and merge them with the field survey sample points to generate a burned area sample library. The specific operation of step 3) is as follows: divide the sample database of burned areas obtained in step 2) into a training dataset and a validation dataset; All sample data were randomly allocated to one of two datasets in a 70%:30% ratio; The training dataset is subsetted, with 80% of the data used to train a separate classification algorithm and 20% used to optimize the model; the validation dataset is used to evaluate the classification results of subsequent final sample data. Step 4) The specific operation is as follows: Based on the Google Earth Engine platform, calculate the feature vector of the image, (1) spectral index feature vector, normalized vegetation index NDVI, enhanced vegetation index EVI, area of ​​fire index BAI, improved normalized water index MNDWI, normalized fire index NBR and normalized fire index NBR2; (2) KT transform main components, brightness TCB, greenness TCG and humidity TCW; Finally, establish a time series feature vector dataset, and the calculation formula of the relevant index is as follows: NDVI = (NIR - RED) / (NIR + RED) EVI=2.5*(NIR-RED) / (NIR+6RED-7.5BLUE+1) BAI=1 / [(0.1-RED) 2 +(0.06-NIR) 2 , MNDWI=(GREEN-SWIR2) / (GREEN+SWIR2) NBR = (NIR - SWIR²) / (NIR + SWIR²) NBR2=(SWIR1-SWIR2) / (SWIR1+SWIR2) TCB=(0.3029*BLUE)+(0.2786*GREEN)+(0.4733*RED) +(0.5599*NIR)+(0.508*SWIR1)+(0.1872*SWIR2) TCG=(-0.2941*BLUE)-(0.243*GREEN)-(0.5424*RED) +(0.7276*NIR)+(0.0713*SWIR1)-(0.1608*SWIR2) TCW=(0.1511*BLUE)+(0.1973*GREEN)+(0.3283*RED) +(0.3407*NIR)-(0.7117*SWIR1)-(0.4559*SWIR2) Step 5) The specific operation is as follows: Build a random forest model on the Google Earth Engine cloud platform, set the number of growth trees and the number of feature variables input when splitting nodes, and adjust the settings of the two parameters, the number of growth trees and the number of feature variables input when splitting nodes, by calculating the OOB misclassification rate to obtain the optimal model. Perform supervised classification based on the results of the 9 feature vectors in step 4) to obtain the spatial distribution information of the burned areas.

2. The method for extracting burned areas based on a cloud computing platform and random forest algorithm according to claim 1, characterized in that, The specific operations for remote sensing data acquisition and preprocessing in step 1) are as follows: S1. Based on the Google Earth Engine platform, acquire Landsat series imagery, including Landsat5, Landsat7 and Landsat8 surface reflectance imagery products. S2. The acquired images are processed as follows: invalid observations such as clouds and shadows are removed from the images; terrain shadows are removed from the images; strip errors in the Landsat 7 images are removed; and the reflectance of the OLI and TM / ETM+ sensors is normalized using a statistical coordination function between spectral values.

3. The method for extracting burned areas based on a cloud computing platform and random forest algorithm according to claim 2, characterized in that, Step 6) Specifically, the accuracy of the model classification results is evaluated using the validation dataset from Step 3. The accuracy of the fire trace extraction results is assessed by calculating user accuracy, producer accuracy, overall accuracy, and the Kappa coefficient.

4. The method for extracting burned areas based on a cloud computing platform and random forest algorithm according to claim 3, characterized in that, Step 7) Specifically, download the output results of the random forest model to the local processor and import them into ArcGIS for mapping processing.

Images