WO2019200743A1 - 森林火灾预测方法、装置、计算机设备和存储介质 - Google Patents

森林火灾预测方法、装置、计算机设备和存储介质 Download PDF

Info

Publication number
WO2019200743A1
WO2019200743A1 PCT/CN2018/095485 CN2018095485W WO2019200743A1 WO 2019200743 A1 WO2019200743 A1 WO 2019200743A1 CN 2018095485 W CN2018095485 W CN 2018095485W WO 2019200743 A1 WO2019200743 A1 WO 2019200743A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
fire
forest
forest fire
prediction model
Prior art date
Application number
PCT/CN2018/095485
Other languages
English (en)
French (fr)
Inventor
王义文
王健宗
肖京
Original Assignee
平安科技（深圳）有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Publication of WO2019200743A1 publication Critical patent/WO2019200743A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply

Definitions

  • the present application relates to the field of computer technology, and in particular, to a forest fire prediction method, apparatus, computer equipment and storage medium.
  • the existing fire prediction is basically based on the natural parameters such as air temperature, air humidity, precipitation, continuous drought days, wind, etc., but the accuracy of the prediction results is low.
  • the Australian Fire Insurance Forecast System was established after more than 800 field ignition experiments. Although it has a solid field test basis, it is only suitable for a single type of combustibles. Therefore, it is an urgent problem to provide a forest fire prediction method with higher predictability and better accuracy.
  • the main object of the present application is to provide a forest fire prediction method, apparatus, computer apparatus and storage medium that are more practical and accurate than the prior art.
  • a forest fire prediction method which includes:
  • the fire risk coefficient of the forest area is determined according to the operation result.
  • the application also provides a forest fire prediction device, comprising:
  • An operation unit configured to input the specified data into a preset forest fire prediction model based on a ConvLSTM model
  • a determining unit configured to determine a fire risk coefficient of the forest area according to the operation result.
  • the application further provides a computer device comprising a memory and a processor, the memory storing computer readable instructions, the processor executing the computer readable instructions to implement the steps of any of the methods described above.
  • the present application also provides a computer non-transitory readable storage medium having stored thereon computer readable instructions that, when executed by a processor, implement the steps of any of the methods described above.
  • the forest fire prediction method, device, computer equipment and storage medium of the present application make full use of the time series modeling capability and space-time characteristics of the ConvLSTM model to establish a fire prediction model that takes both time series and spatial sequences into account, thereby more accurately predicting forest fires. .
  • FIG. 1 is a schematic flow chart of a forest fire prediction method according to an embodiment of the present application.
  • FIG. 2 is a schematic diagram of an internal structure of a ConvLSTM according to an embodiment of the present application
  • FIG. 3 is a schematic structural diagram of a code prediction ConvLSTM network according to an embodiment of the present application.
  • FIG. 4 is a schematic block diagram showing the structure of a forest fire prediction apparatus according to an embodiment of the present application.
  • FIG. 5 is a schematic block diagram showing the structure of an arithmetic unit according to an embodiment of the present application.
  • FIG. 6 is a schematic block diagram showing the structure of an arithmetic unit according to an embodiment of the present application.
  • FIG. 7 is a schematic block diagram showing the structure of a forest fire prediction apparatus according to an embodiment of the present application.
  • FIG. 8 is a schematic block diagram showing the structure of a computer device according to an embodiment of the present application.
  • an embodiment of the present application provides a forest fire prediction method, including the following steps:
  • the forest area is a region in which plants such as trees are grown, which are generally rare in humans, and areas such as trees and other plants, such as virgin forests in the Daxinganling Mountains, forests in a certain area on Changbai, etc. .
  • the above designated data refers to data that may reflect the fire situation in the forest area, including: fire point information data corresponding to the above forest area, fire image data, fire date data, burn scar data during the fire season, fire alarm record data, and terrain data. And terrain vector data, etc. In other embodiments, weather data or the like may also be included.
  • the above ConvLSTM model not only has the timing modeling capability of the LSTM, but also can describe the local features like the CNN, and can be said to have both spatiotemporal characteristics.
  • the internal structure of ConvLSTM is shown in Figure 2.
  • the working principle can be expressed by the following formula:
  • the fire risk coefficient refers to a coefficient of fire in the forest area, that is, a probability of occurrence of a forest fire, and is generally a forest fire risk rating.
  • the step S2 of inputting the specified data into the preset forest fire prediction model based on the ConvLSTM model includes:
  • the fire point information data, the fire image data, the ignition date data, the burn scar data in the fire season, and the on-site fire alarm record data are related to the date of occurrence, that is, the nature of the time series, so the Analysis of timing.
  • the above terrain data and terrain vector data are generally spatial data, such as height, area, etc., so the terrain data and the terrain vector data are spatially sequenced.
  • the fire point information data, the fire image data, the fire date data, the burn scar data in the fire season, the on-site fire alarm record data, the terrain data, and the terrain vector data are classified.
  • Data with time information is divided into one category (fire point information data, fire image data, fire date data, burn scar data in the fire season, and live fire alarm data), and data with spatial information is divided into one category (topographic data and terrain).
  • Vector data and then input the corresponding data into the above forest fire prediction model for calculation, because the forest fire prediction model is based on the ConvLSTM model, which can analyze the timing of the data with time information.
  • the analysis of spatial sequence aspects of data with spatial information is common knowledge in the art for data input into the ConvLSTM model, and is not described here.
  • the fire point information data is obtained by analyzing fire mask data of the forest area.
  • a mask is a template for an image filter that is often used to process remotely sensed images.
  • the fire mask data is an image filter template for extracting a fire point. The specific process is: when extracting the ground object information of the satellite remote sensing image, pixel filtering is performed on the image through an n*n matrix. Then display the fire point information we need.
  • the fire mask data set MOD14A1 is divided into 10 levels, as shown in the following table:
  • Unprocessed cell 1 Unprocessed pixel 2
  • Unprocessed cell 3 Waters 4 cloud 5 Bare land without fire
  • Extract the pixel values of the attribute values 8 and 9, superimpose each of them with the vegetation coverage map of the study area, remove the pseudo forest fire point information, and then compare with the vegetation information of the corresponding time period, and change the image of the vegetation before and after the fire date.
  • the element is determined to be a fire point, and finally the 8+9 data set is used to determine the fire point information.
  • the above fire image metadata and ignition date data are obtained by analyzing the fired data set.
  • the above-mentioned burned land refers to the land in the forest that has not yet been raised with new forests after being burned by fire.
  • the data set of the above-mentioned fire burnt land refers to a set of fire burned land in the above forest area, and each fire burnt land in the set of fire burned land records fire image data and fire date data.
  • the pixel represents the area of the ground.
  • the altitude of remote sensing satellites is generally between 4,000 and 600 kilometers, and the image resolution is generally between 1 and 1 meter.
  • a pixel is equivalent to a point on a TV screen (a television is an image frame composed of several dots), which is equivalent to a pixel on a computer display screen, which is equivalent to one of a group of people holding different swatches to draw a picture.
  • a television is an image frame composed of several dots
  • the resolution is 1 km
  • one pixel represents an area of 1 km X 1 km on the ground, that is, 1 square kilometer
  • the resolution is 30 meters
  • one pixel represents an area of 30 m ⁇ 30 m on the ground
  • the resolution is 1 meter
  • the above-mentioned fire pixel is a unit indicating the area of the fire. If a fire pixel is 1 square meter, how many fire pixels can be used to describe the area of the fire.
  • the above-mentioned fire date data refers to date data when a fire occurs in a fire.
  • the burn scar data for the above fire season was obtained by analyzing the vegetation index data of the above forest area.
  • Vegetation index is an indicator of vegetation distribution density and growth status, and is positively correlated with vegetation cover.
  • the above-mentioned burn scar data in the fire season refers to the trace left by the forest fire, that is, the vegetation index on the trace area is much lower than the vegetation index around it, but over time, the trace area will continue to recover.
  • the growth state of the vegetation that is, the vegetation index of the trace area is increased, so the burn scar data of the fire season can be obtained through the analysis of the vegetation index data of the forest area.
  • the above vegetation index data is based on the spectral characteristics of the vegetation, combining the visible and near-infrared bands of the satellite to form various vegetation indices.
  • Vegetation index according to the spectral characteristics of the vegetation, combines the visible and near-infrared bands of the satellite to form various vegetation indices. Vegetation index is a simple, effective and empirical measure of the state of vegetation on the surface. In the field of remote sensing applications, vegetation index has been widely used to qualitatively and quantitatively evaluate vegetation cover and its growth vigor. Because the vegetation spectrum is a complex mixed reaction of vegetation, soil brightness, environmental impact, shadow, soil color and humidity, and is affected by atmospheric space-time phase changes, the vegetation index does not have a universal value, and its research often indicates different results. . This index increases rapidly with increasing biomass. When the vegetation is at medium and low coverage, the index increases rapidly with the increase of coverage. When it reaches a certain coverage, the growth is slow, so it is suitable for the dynamic monitoring of the early and middle growth stages of vegetation.
  • the vegetation index specifically includes:
  • NDVI Normalized Vegetation Index
  • ⁇ NIR and ⁇ RED represent the reflectances of the near-infrared band and the red band, respectively.
  • the NDV1 value is between -1 and 1.
  • EVI Enhanced Vegetation Index
  • ⁇ NIR , ⁇ RED , and ⁇ BLUE represent the reflectances of the near-infrared, infrared, and blue bands, respectively.
  • ⁇ NIR and ⁇ RED represent the reflectance of the near-infrared band and the infrared band, respectively.
  • ⁇ NIR and ⁇ RED represent the reflectance of the near-infrared band and the infrared band, respectively.
  • the difference vegetation index is extremely sensitive to changes in soil background and is beneficial to the monitoring of vegetation ecological environment. Therefore, it is also called the Environmental Vegetation Index (EVI).
  • EVI Environmental Vegetation Index
  • L is a soil regulation coefficient, which is related to vegetation concentration and is determined by actual regional conditions to reduce the sensitivity of vegetation index to different soil reflection changes.
  • SAVI NDVI; for medium vegetation coverage, the value of L is generally close to 0.5.
  • the multiplication factor (1+L) is mainly used to ensure that the final SAVI value is between -1 and 1.
  • the index can reduce the impact of the soil background, but may lose part of the vegetation signal, making the vegetation index low.
  • the above-mentioned on-site fire alarm record data is the fire alarm record data of the fire brigade in the above forest area, which generally includes time, place, and cause of fire.
  • the above terrain data is generally obtained by analyzing the satellite remote sensing image of the above forest area, and specifically includes the data shown in the following table:
  • Terrain vector data is a vector data set of basic geographic elements on existing topographic maps, including 9 data sets of water system, residential land and facilities, traffic, pipelines, realms and political districts, landforms and soils, vegetation, place names and annotations, and Save spatial relationships between features and related attribute information.
  • the terrain vector data includes elements such as water system, pipeline, realm and political area, landform and soil quality, vegetation, place name and annotation, and spatial relationship between elements.
  • the fire point information data, the fire image data, the fire date data, the burn scar data of the fire season, and the on-site fire alarm record data are time-series analyzed in the forest fire prediction model, and the terrain data is And before the step S21 of performing spatial sequence analysis in the forest fire prediction model, the terrain vector data includes:
  • the fire alarm record generally records the cause of the fire of the forest fire, such as a fire caused by drying, a fire caused by lightning, and a fire caused by humans.
  • the fire alarm record is issued on August 1, 2016.
  • the cause of the fire is caused by the burning of straw by the farmers.
  • the semantic analysis method can be used to analyze the meaning of the fire alarm record as artificial burning straw and cause fire, and then record it as artificial.
  • Arson record data can be used to analyze the meaning of the fire alarm record as artificial burning straw and cause fire, and then record it as artificial.
  • the record data corresponding to the arson is cleared, and the alarm data of the natural fire is retained.
  • the human factor is an uncontrollable random factor, and the removal of the alarm data can improve the accuracy of the forest fire prediction.
  • the method before the step S2 of inputting the specified data into the preset forest fire prediction model based on the ConvLSTM model, the method includes:
  • the normalization process described above refers to transforming a dimensioned expression into a dimensionless expression and becoming a scalar.
  • the different data is converted into the same expressed scalar to facilitate calculation.
  • the method before the step S2 of inputting the specified data into the preset forest fire prediction model based on the ConvLSTM model, the method includes:
  • S203 Call the forest fire prediction model corresponding to a geographic location of the forest area according to a geographic location of the forest area, where the forest fire prediction model includes multiple, and different forest fire prediction models are for different geographic locations. And a forest fire prediction model based on the ConvLSTM model.
  • the geographical location is a specific longitude and latitude
  • the environmental characteristics of the forest area can be determined by the latitude and longitude of the forest area, for example, the climate of the south and north of the People's Republic of China is different, if used
  • the forest fires are predicted by the same forest fire model, and the predicted results may be different.
  • two or more forest fire prediction models based on the ConvLSTM model are set, and different forest fire prediction models are based on The specified data of the forest areas in different geographical locations are obtained through training and learning, that is, the relevant modeling data of different geographical locations are respectively collected by the same modeling method, and then modeled.
  • the corresponding fire prediction model is called for different regions to obtain more accurate forest fire prediction results.
  • the method before the step S2 of inputting the specified data into the preset forest fire prediction model based on the ConvLSTM model, the method includes:
  • S205 Call the forest fire prediction model corresponding to the topographical attribute of the forest area according to the topographical attribute of the forest area, wherein the forest fire prediction model includes multiple, and different forest fire prediction models are for different landform attributes. And a forest fire prediction model based on the ConvLSTM model.
  • the above-mentioned landform that is, the general name of various forms of the earth's surface
  • terrain can also be referred to as terrain.
  • the surface morphology is diverse and the causes are not the same. It is the result of the combined effects of internal and external geological processes on the earth's crust.
  • the internal geological action caused the fluctuation of the surface, controlled the distribution of the land and sea, the geographical arrangement of the mountains, plateaus, basins and plains, and determined the structural framework of the landform.
  • the geological role of external forces flowing water, wind power, solar radiant energy, atmospheric and biological growth and activities
  • Geomorphological properties are the specific forms of landforms, such as mountains, plateaus, and basins.
  • the forest fire prediction model corresponding to the forest area may be invoked by using the forest area.
  • the landform attribute of the forest area of Daxing'anling is different from the landform attribute of the forest area of Xishuangbanna, if the same forest fire model is used for forest fire. It is predicted that there may be differences in the predicted results.
  • two or more forest fire prediction models for different landform attributes are set, and different forest fire prediction models are trained based on specified data of forest regions with different landforms. Learning, that is, through the same modeling method, the relevant modeling data of different geomorphological attributes are separately collected and then modeled. In the forest fire alarm forecast, the corresponding fire prediction model is called for different landforms to obtain more accurate forest fire prediction results.
  • the acquired modeling data is divided into a training set and a test set, and then the sample data corresponding to the training set is input into the ConvLSTM model for training.
  • the fire point information data and the fire image corresponding to each historical fire risk coefficient data are used.
  • Time series analysis of metadata, fire date data and burn scar data in the fire season, spatial data analysis of terrain data and terrain vector data, and finally a forest fire prediction model based on ConvLSTM model, namely input fire point information data and fire image Metadata, fire date data, burn scar data for fire seasons, topographic data, and terrain vector data are used in the forest fire prediction model, which outputs a fire risk factor.
  • the validity of the forest fire prediction model is verified by the test set. If the verification is passed, the forest fire prediction model can be applied in practice, that is, forest fire is predicted.
  • climate data such as season, air temperature, air humidity, sunny day, rainy day and the like when capturing historical remote sensing images of various satellites are also acquired.
  • the data is mainly time-series. analysis.
  • forest fire prediction is made, the climate data is obtained and then input into the forest fire prediction model.
  • the forest fire prediction method of the present embodiment makes full use of the time series modeling capability and space-time characteristics of the ConvLSTM model to establish a fire prediction model that takes both time series and spatial sequences into account, thereby more accurately predicting forest fires.
  • an embodiment of the present application further provides a forest fire prediction apparatus, including the steps of:
  • the obtaining unit 10 is configured to acquire specified data of a forest area to be predicted of a forest fire;
  • the operation unit 20 is configured to input the specified data into a preset forest fire prediction model based on the ConvLSTM model;
  • the determining unit 30 is configured to determine a fire risk coefficient of the forest area according to the operation result.
  • the forest area is a region in which plants such as trees are grown, and is generally a region where humans are scarce, and plants such as trees, such as virgin forests in the Daxinganling Mountains and forests in a certain region on Changbai.
  • the above designated data refers to data that may reflect the fire situation in the forest area, including: fire point information data corresponding to the above forest area, fire image data, fire date data, burn scar data during the fire season, fire alarm record data, and terrain data. And terrain vector data, etc. In other embodiments, weather data or the like may also be included.
  • the ConvLSTM model not only has the LSTM timing modeling capability, but also can describe local features like CNN, and can be said to have both spatiotemporal characteristics.
  • the internal structure of the ConvLSTM model is shown in Figure 2.
  • the working principle can be expressed by the following formula:
  • the fire risk coefficient refers to a coefficient of fire in the forest area, that is, a probability of occurrence of a forest fire, generally a forest fire risk rating.
  • the operation unit 20 includes:
  • the operation module 21 is configured to perform time series analysis on the fire point information data, fire image data, fire date data, burn scar data in the fire season, and on-site fire alarm record data in the forest fire prediction model, and perform terrain data. And the terrain vector data is subjected to spatial sequence analysis in the forest fire prediction model.
  • the operation module 21, the fire point information data, the fire image data, the ignition date data, the burn scar data in the fire season, and the on-site fire alarm record data are related to the date of occurrence, that is, the nature of the time series, so the timing is performed.
  • the above terrain data and terrain vector data are generally spatial data, such as height, area, etc., so the terrain data and the terrain vector data are spatially sequenced.
  • the fire point information data, the fire image data, the fire date data, the burn scar data in the fire season, the on-site fire alarm record data, the terrain data, and the terrain vector data are classified.
  • Data with time information is divided into one category (fire point information data, fire image data, fire date data, burn scar data in the fire season, and live fire alarm data), and data with spatial information is divided into one category (topographic data and terrain).
  • Vector data and then input the corresponding data into the above forest fire prediction model for calculation, because the forest fire prediction model is based on the ConvLSTM model, which can analyze the timing of the data with time information.
  • the analysis of spatial sequence aspects of data with spatial information is common knowledge in the art for data input into the ConvLSTM model, and is not described here.
  • the fire point information data is obtained by analyzing fire mask data of the forest area.
  • a mask is a template for an image filter that is often used to process remotely sensed images.
  • the fire mask data is an image filter template for extracting a fire point. The specific process is: when extracting the ground object information of the satellite remote sensing image, pixel filtering is performed on the image through an n*n matrix. Then display the fire point information we need.
  • the fire mask data set MOD14A1 is divided into 10 levels, as shown in the following table:
  • Unprocessed cell 1 Unprocessed pixel 2
  • Unprocessed pixel 3 Waters 4 cloud 5 Bare land without fire 6
  • Unknown pixel 7 Low confidence detection fire point
  • Extract the pixel values of the attribute values 8 and 9, superimpose each of them with the vegetation coverage map of the study area, remove the pseudo forest fire point information, and then compare with the vegetation information of the corresponding time period, and change the image of the vegetation before and after the fire date.
  • the element is determined to be a fire point, and finally the 8+9 data set is used to determine the fire point information.
  • the above fire image metadata and ignition date data are obtained by analyzing the fired data set.
  • the above-mentioned burned land refers to the land in the forest that has not yet been raised with new forests after being burned by fire.
  • the data set of the above-mentioned fire burnt land refers to a set of fire burned land in the above forest area, and each fire burnt land in the set of fire burned land records fire image data and fire date data.
  • the pixel represents the area of the ground.
  • the altitude of remote sensing satellites is generally between 4,000 and 600 kilometers, and the image resolution is generally between 1 and 1 meter.
  • a pixel is equivalent to a point on a TV screen (a television is an image frame composed of several dots), which is equivalent to a pixel on a computer display screen, which is equivalent to one of a group of people holding different swatches to draw a picture.
  • a television is an image frame composed of several dots
  • the resolution is 1 km
  • one pixel represents an area of 1 km X 1 km on the ground, that is, 1 square kilometer
  • the resolution is 30 meters
  • one pixel represents an area of 30 m ⁇ 30 m on the ground
  • the resolution is 1 meter
  • the above-mentioned fire pixel is a unit indicating the area of the fire. If a fire pixel is 1 square meter, how many fire pixels can be used to describe the area of the fire.
  • the above-mentioned fire date data refers to date data when a fire occurs in a fire.
  • the burn scar data for the above fire season was obtained by analyzing the vegetation index data of the above forest area.
  • Vegetation index is an indicator of vegetation distribution density and growth status, and is positively correlated with vegetation cover.
  • the above-mentioned burn scar data in the fire season refers to the trace left by the forest fire, that is, the vegetation index on the trace area is much lower than the vegetation index around it, but over time, the trace area will continue to recover.
  • the growth state of the vegetation that is, the vegetation index of the trace area is increased, so the burn scar data of the fire season can be obtained through the analysis of the vegetation index data of the forest area.
  • the above vegetation index data is based on the spectral characteristics of the vegetation, combining the visible and near-infrared bands of the satellite to form various vegetation indices.
  • Vegetation index according to the spectral characteristics of the vegetation, combines the visible and near-infrared bands of the satellite to form various vegetation indices. Vegetation index is a simple, effective and empirical measure of the state of vegetation on the surface. In the field of remote sensing applications, vegetation index has been widely used to qualitatively and quantitatively evaluate vegetation cover and its growth vigor. Because the vegetation spectrum is a complex mixed reaction of vegetation, soil brightness, environmental impact, shadow, soil color and humidity, and is affected by atmospheric space-time phase changes, the vegetation index does not have a universal value, and its research often indicates different results. . This index increases rapidly with increasing biomass. When the vegetation is at medium and low coverage, the index increases rapidly with the increase of coverage. When it reaches a certain coverage, the growth is slow, so it is suitable for the dynamic monitoring of the early and middle growth stages of vegetation.
  • the vegetation index specifically includes:
  • NDVI Normalized Vegetation Index
  • ⁇ NIR and ⁇ RED represent the reflectances of the near-infrared band and the red band, respectively.
  • the NDV1 value is between -1 and 1.
  • EVI Enhanced Vegetation Index
  • ⁇ NIR , ⁇ RED , and ⁇ BLUE represent the reflectances of the near-infrared, infrared, and blue bands, respectively.
  • ⁇ NIR and ⁇ RED represent the reflectance of the near-infrared band and the infrared band, respectively.
  • ⁇ NIR and ⁇ RED represent the reflectance of the near-infrared band and the infrared band, respectively.
  • the difference vegetation index is extremely sensitive to changes in soil background and is beneficial to the monitoring of vegetation ecological environment. Therefore, it is also called the Environmental Vegetation Index (EVI).
  • EVI Environmental Vegetation Index
  • L is a soil regulation coefficient, which is related to vegetation concentration and is determined by actual regional conditions to reduce the sensitivity of vegetation index to different soil reflection changes.
  • SAVI NDVI; for medium vegetation coverage, the value of L is generally close to 0.5.
  • the multiplication factor (1+L) is mainly used to ensure that the final SAVI value is between -1 and 1.
  • the index can reduce the impact of the soil background, but may lose part of the vegetation signal, making the vegetation index low.
  • the above-mentioned on-site fire alarm record data is the fire alarm record data of the fire brigade in the above forest area, which generally includes time, place, and cause of fire.
  • the above terrain data is generally obtained by analyzing the satellite remote sensing image of the above forest area, and specifically includes the data shown in the following table:
  • Terrain vector data is a vector data set of basic geographic elements on existing topographic maps, including 9 data sets of water system, residential land and facilities, traffic, pipelines, realms and political districts, landforms and soils, vegetation, place names and annotations, and Save spatial relationships between features and related attribute information.
  • the terrain vector data includes elements such as water system, pipeline, boundary and political area, landform and soil quality, vegetation, place name and annotation, and spatial relationship between elements.
  • the operation unit 20 includes:
  • the determining module 22 is configured to determine whether there is artificially arson recorded data in the live fire alarm record data
  • the clearing module 23 is configured to clear the recorded data of the artificial arson from the live fire record data if there is record data of the artificial arson.
  • the cause of the fire of the forest fire such as a fire caused by drying, a fire caused by lightning, and a fire caused by human beings, are recorded in the fire alarm record.
  • the fire alarm record is issued on August 1, 2016.
  • the cause of the fire is caused by the burning of straw by the farmers.
  • the semantic analysis method can be used to analyze the meaning of the fire alarm record as artificial burning straw and cause fire, and then record it as artificial.
  • Arson record data can be used to analyze the meaning of the fire alarm record as artificial burning straw and cause fire, and then record it as artificial.
  • the clearing module 23 clears the record data of the corresponding person for arson, and retains the alarm data of the natural fire.
  • the human factor is an uncontrollable random factor, and the removal of the alarm data can improve the accuracy of the forest fire prediction.
  • the forest fire prediction device includes:
  • the normalization unit 201 is configured to perform normalization processing on the obtained specified data to obtain the standardized specified data.
  • the normalization process described above refers to transforming a dimensioned expression into a dimensionless expression and becoming a scalar.
  • the different data is converted into the same expressed scalar to facilitate calculation.
  • the forest fire prediction device includes:
  • a first acquiring unit configured to acquire a geographic location of the forest area
  • a first calling unit configured to invoke the forest fire prediction model corresponding to a geographic location of the forest area according to a geographic location of the forest area, where the forest fire prediction model includes multiple, different forest fire prediction models It is a forest fire prediction model based on different geographic locations and based on the ConvLSTM model.
  • the geographical position is a specific longitude and latitude, and the environmental characteristics of the forest area can be determined by the latitude and longitude of the forest area, for example, the climate of the south and north of the People's Republic of China is different.
  • the forest fire model is used to predict the forest fire model, there may be differences in the predicted results.
  • two or more forest fire prediction models based on the ConvLSTM model are set, and different forest fire prediction models are used. It is based on the specified data of the forest area in different geographical locations for training and learning, that is, the relevant modeling data of different geographical locations are separately collected by the same modeling method, and then modeled. In the forest fire alarm forecast, the corresponding fire prediction model is called for different regions to obtain more accurate forest fire prediction results.
  • the forest fire prediction device includes:
  • a second acquiring unit configured to acquire a topographical attribute of the forest area
  • a second calling unit configured to invoke the forest fire prediction model corresponding to the topographical attribute of the forest area according to the topographical attribute of the forest area, wherein the forest fire prediction model includes multiple, different forest fire prediction models It is a forest fire prediction model based on different geomorphological attributes and based on the ConvLSTM model.
  • the second acquisition unit and the second invocation unit can also be referred to as terrain.
  • the surface morphology is diverse and the causes are not the same. It is the result of the combined effects of internal and external geological processes on the earth's crust.
  • the internal geological action caused the fluctuation of the surface, controlled the distribution of the land and sea, the geographical arrangement of the mountains, plateaus, basins and plains, and determined the structural framework of the landform.
  • the geological role of external forces flowing water, wind power, solar radiant energy, atmospheric and biological growth and activities
  • Geomorphological properties are the specific forms of landforms, such as mountains, plateaus, and basins.
  • the forest fire prediction model corresponding to the forest area may be invoked by using the forest area.
  • the landform attribute of the forest area of Daxing'anling is different from the landform attribute of the forest area of Xishuangbanna, if the same forest fire model is used for forest fire. It is predicted that there may be differences in the predicted results.
  • two or more forest fire prediction models for different landform attributes are set, and different forest fire prediction models are trained based on specified data of forest regions with different landforms. Learning, that is, through the same modeling method, the relevant modeling data of different geomorphological attributes are separately collected and then modeled. In the forest fire alarm forecast, the corresponding fire prediction model is called for different landforms to obtain more accurate forest fire prediction results.
  • the acquired modeling data is divided into a training set and a test set, and then the sample data corresponding to the training set is input into the ConvLSTM model for training.
  • the fire point information data and the fire image corresponding to each historical fire risk coefficient data are used.
  • Time series analysis of metadata, fire date data and burn scar data in the fire season, spatial data analysis of terrain data and terrain vector data, and finally a forest fire prediction model based on ConvLSTM model, namely input fire point information data and fire image Metadata, fire date data, burn scar data for fire seasons, topographic data, and terrain vector data are used in the forest fire prediction model, which outputs a fire risk factor.
  • the validity of the forest fire prediction model is verified by the test set. If the verification is passed, the forest fire prediction model can be applied in practice, that is, forest fire is predicted.
  • climate data such as season, air temperature, air humidity, sunny day, rainy day and the like when capturing historical remote sensing images of various satellites are also acquired.
  • the data is mainly time-series. analysis.
  • forest fire prediction is made, the climate data is obtained and then input into the forest fire prediction model.
  • the forest fire prediction device of the present embodiment makes full use of the time series modeling capability and space-time characteristics of the ConvLSTM model to establish a fire prediction model that takes both time series and spatial sequences into account, thereby more accurately predicting forest fires.
  • the computer device may be a server, and its internal structure may be as shown in FIG. 8.
  • the computer device includes a processor, memory, network interface, and database connected by a system bus. Among them, the computer designed processor is used to provide calculation and control capabilities.
  • the memory of the computer device includes a non-volatile storage medium, an internal memory.
  • the non-volatile storage medium stores an operating system, computer readable instructions, and a database.
  • the memory provides an environment for the operation of operating systems and computer readable instructions in a non-volatile storage medium.
  • the database of the computer device is used to store data such as a forest fire prediction model based on the ConvLSTM model.
  • the network interface of the computer device is used to communicate with an external terminal via a network connection.
  • the computer readable instructions are executed by a processor to implement the processes of the various method embodiments described above.
  • An embodiment of the present application further provides a computer non-volatile readable storage medium having stored thereon computer readable instructions, which are implemented by a processor to implement the processes of the foregoing method embodiments.

Landscapes

  • Business, Economics & Management (AREA)
  • Engineering & Computer Science (AREA)
  • Economics (AREA)
  • Human Resources & Organizations (AREA)
  • Strategic Management (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • Marketing (AREA)
  • General Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • Tourism & Hospitality (AREA)
  • Public Health (AREA)
  • General Health & Medical Sciences (AREA)
  • Primary Health Care (AREA)
  • Water Supply & Treatment (AREA)
  • Development Economics (AREA)
  • Game Theory and Decision Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

一种森林火灾预测方法、装置、计算机设备和存储介质，其中方法包括：获取待预测森林火灾的森林区域的指定数据；将指定数据输入到预设的基于ConvLSTM模型的森林火灾预测模型进行运算；根据运算结果确定森林区域的火险系数。

Description

森林火灾预测方法、装置、计算机设备和存储介质
本申请要求于2018年4月17日提交中国专利局、申请号为2018103453177，申请名称为“森林火灾预测方法、装置、计算机设备和存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及到计算机技术领域，特别是涉及到一种森林火灾预测方法、装置、计算机设备和存储介质。
背景技术
现有的火灾预测基本是根据森林所处区域的空气温度、空气湿度、降水、连旱天数、风等自然参数进行预测，但是预测结果准确度较低。比如，澳大利亚火险预报系统是经过800多次野外点火实验建立起来的，虽然有坚实的野外场试验基础，但是只适用于单一的可燃物类型等。所以提供一种预测实用性更高和准确度更好的森林火灾预测方法是亟需解决的问题。
技术问题
本申请的主要目的为提供一种实用性和准确性相对于现有技术更高的森林火灾预测方法、装置、计算机设备和存储介质。
技术解决方案
为了实现上述申请目的，本申请提出一种森林火灾预测方法，其特征在于，包括：
获取待预测森林火灾的森林区域的指定数据；
将所述指定数据输入到预设的基于ConvLSTM模型的森林火灾预测模型进行运算；
根据运算结果确定所述森林区域的火险系数。
本申请还提供一种森林火灾预测装置，包括:
获取单元，用于获取待预测森林火灾的森林区域的指定数据；
运算单元，用于将所述指定数据输入到预设的基于ConvLSTM模型的森林火灾预测模型进行运算；
确定单元，用于根据运算结果确定所述森林区域的火险系数。
本申请还提供一种计算机设备，包括存储器和处理器，所述存储器存储有计算机可读指令，所述处理器执行所述计算机可读指令时实现上述任一项所述方法的步骤。
本申请还提供一种计算机非易失性可读存储介质，其上存储有计算机可读指令，所述计算机可读指令被处理器执行时实现上述任一项所述的方法的步骤。
有益效果
本申请的森林火灾预测方法、装置、计算机设备和存储介质，充分利用ConvLSTM模型的时序建模能力以及时空特性，建立一个兼顾时间序列和空间序列的火灾预测模型，从而更加准确的进行森林火灾预测。
附图说明
图1为本申请一实施例的森林火灾预测方法的流程示意图；
图2为本申请一实施例的ConvLSTM的内部结构示意图；
图3为本申请一实施例的编码预测ConvLSTM网络的结构示意图；
图4为本申请一实施例的森林火灾预测装置的结构示意框图；
图5为本申请一实施例的运算单元的结构示意框图；
图6为本申请一实施例的运算单元的结构示意框图；
图7为本申请一实施例的森林火灾预测装置的结构示意框图；
图8为本申请一实施例的计算机设备的结构示意框图。
本发明的最佳实施方式
参照图1，本申请实施例提出一种森林火灾预测方法，包括步骤：
S1、获取待预测森林火灾的森林区域的指定数据；
S2、将所述指定数据输入到预设的基于ConvLSTM模型的森林火灾预测模型进行运算；
S3、根据运算结果确定所述森林区域的火险系数。
如上述步骤S1所述，上述森林区域即为生长有树木等植物的区域，其一般为人类稀少，而树木等植物较多的区域，如大兴安岭的原始森林、长白上的某一区域的森林等。上述指定数据是指可能反映森林区域火灾情况的数据，具体包括：对应上述森林区域的火点信息数据、火像元数据、着火日期数据、火灾季节的烧伤疤痕数据、现场火警记录数据、地形数据以及地形矢量数据等。在其它实施例中，还可以包括气象数据等。
如上述步骤S2所述，上述ConvLSTM模型其不仅具有LSTM的时序建模能力，而且还能像CNN一样刻画局部特征，可以说是同时具备时空特性。ConvLSTM的内部结构如图2所示，其工作原理可以由以下公式表示：
i t＝σ(W xi*X t+W hi*H t-1+W ciо C t-1+b i)
f t＝σ(W xf*X t+W hf*H t-1+W cfо C t-1+b f)
C t＝f tо C t-1+i tо tanh(W xc*X t+W hc*H t-1+b c)
o t＝σ(W xo*X t+W ho*H t-1+W coо C t+b o)
H t＝o tо tanh(C t)
上述各式子中，*表示卷积，空心小圆圈表示矩阵对应元素相乘，又称为Hadamard乘积。值得注意的是，这里的χ,C,H,i,f,o都是三维的tensor(张量)，它们的后两个维度代表行和列的空间信息，可以把ConvLSTM模型想象成是处理二维网格中的特征向量的模型，其可以根据网格中周围点的特征来预测中心网格的特征。编码预测ConvLSTM网络则如图3所示。
如上述步骤S3所述，上述火险系数是指上述森林区域发生火灾的系数，即发生森林火灾的概率， 一般为森林火险等级。
在本实施例中，上述将所述指定数据输入到预设的基于ConvLSTM模型的森林火灾预测模型进行运算的步骤S2，包括：
S21、将所述火点信息数据、火像元数据、着火日期数据、火灾季节的烧伤疤痕数据和现场火警记录数据在所述森林火灾预测模型中进行时间序列分析，将地形数据和地形矢量数据在所述森林火灾预测模型中进行空间序列分析。
如上述步骤S21所述，上述火点信息数据、火像元数据、着火日期数据、火灾季节的烧伤疤痕数据和现场火警记录数据等与发生日期相关，即有时序上的性质，因此将其进行时序方面的分析。上述地形数据和地形矢量数据一般为空间数据，如高度、面积等，所以将地形数据和地形矢量数据进行空间序列分析。在一具体实施例中，首先对上述火点信息数据、火像元数据、着火日期数据、火灾季节的烧伤疤痕数据、现场火警记录数据、地形数据以及地形矢量数据等各数据进行分类，将带有时间信息的数据分成一类(火点信息数据、火像元数据、着火日期数据、火灾季节的烧伤疤痕数据和现场火警记录数据)，带有空间信息的数据分成一类(地形数据和地形矢量数据)，然后分别将对应的数据输入到上述森林火灾预测模型中进行计算，因为森林火灾预测模型是基于ConvLSTM模型的模型，可以针对性地对带有时间信息的数据进行时序方面的分析，对带有空间信息的数据进行空间序列方面的分析，关于数据输入到ConvLSTM模型中进行计算是本领域的公知常识，在此不在赘述。
在一具体实施中，上述火点信息数据是通过解析上述森林区域的火掩膜数据得到的。掩膜是一种图像滤镜的模板，经常用其处理的是遥感图像。本申请中，上述火掩膜数据即为用于提取着火点的图像滤镜模板，具体过程为：当提取卫星遥感图像的着地物信息时，通过一个n*n的矩阵来对图像进行像素过滤，然后将我们需要的火点信息显示出来。火掩膜数据集MOD14A1分为10个级别，如下表所示：
级别 对应具体内容
0 未处理像元
1 未处理像元
2 未处理像元
3 水域
4 云
5 没有火的裸地
6 未知像元
7 低置信度检测火点
8 中置信度检测火点
9 高置信度检测火点
提取属性值8和9的像元，将其各自与研究区域的植被覆盖图进行叠加，去掉伪森林火点信息，然后与对应时段的植被信息进行比对，将着火日期前后植被明显变化的像元确定为火点，最后使用8+9数据集确定火点信息。
上述火像元数据和着火日期数据是通过分析火烧迹地数据集而得到的。上述火烧迹地是指森林中经火灾烧毁后尚未长起新林的土地。上述火烧迹地数据集是指上述森林区域的火烧迹地的集合，火烧迹地的集合中对应每一个火烧迹地，记录有火像元数据和着火日期数据。像元代表地面的面积是多少。遥感卫星的飞行高度一般在4000千米～600千米之间，图像分辨率一般从1千米～1米之间。像元相当于电视屏幕上的一个点(电视是由若干个点组成的图像画面)，相当于计算机显示屏幕上的一个象素，相当于一群举着不同色板拼成画图的人中的一个。当分辨率为1千米时，一个像元代表地面1千米X1千米的面积，即1平方千米；当分辨率为30米时，一个像元代表地面30米×30米的面积；当分辨率为1米时，也就是说，图像上的一个像元相当于地面1米x1米的面积，即1平方米。上述火像元即为表示着火迹地的面积一个单位，若一个火像元为1平方米，则可以通过多少个火像元来描述火烧迹地的面积。上述着火日期数据是指火烧迹地发生火灾时的日期数据。
上述火灾季节的烧伤疤痕数据是通过分析上述森林区域的植被指数数据得到的。植被指数是植被分布密度和生长状态的指示因子，与植被覆盖呈正相关关系。而上述的火灾季节的烧伤疤痕数据是指森林火灾留下的痕迹，即该痕迹区域上的植被指数要远低于其周围的植被指数，但是随着时间的过度，上述痕迹区域会不断的恢复植被的生长状态，即痕迹区域的植被指数升高，所以可以通过森林区域的植被指数数据分析得到火灾季节的烧伤疤痕数据。上述植被指数数据是根据植被的光谱特性，将卫星可见光和近红外波段进行组合,形成了各种植被指数。具体的：植被指数，根据植被的光谱特性，将卫星可见光和近红外波段进行组合,形成了各种植被指数。植被指数是对地表植被状况的简单、有效和经验的度量。在遥感应用领域，植被指数已广泛用来定性和定量评价植被覆盖及其生长活力。由于植被光谱表现为植被、土壤亮度、环境影响、阴影、土壤颜色和湿度复杂混合反应，而且受大气空间-时相变化的影响，因此植被指数没有一个普遍的值，其研究经常表明不同的结果。该指数随生物量的增加而迅速增大。在植被处于中、低覆盖度时，该指数随覆盖度的增加而迅速增大，当达到一定覆盖度后增长缓慢，所以适用于植被早、中期生长阶段的动态监测。植被指数具体包括：
归一化植被指数(NDVI)：
Figure PCTCN2018095485-appb-000001
其中，ρ NIR和ρ RED分别代表近红外波段和红光波段的反射率。NDV1值介于-1和1之间。
增强型植被指数(EVI)：
Figure PCTCN2018095485-appb-000002
ρ NIR、ρ RED、ρ BLUE分别代表近红外波段、红外波段、蓝光波段的反射率。
比值植被指数(RVI)：
Figure PCTCN2018095485-appb-000003
ρ NIR和ρ RED分别代表近红外波段、红外波段的反射率。
差值植被指数(DVI):
DVI＝ρ NIR-ρ RED
ρ NIR和ρRED分别代表近红外波段、红外波段的反射率。差值植被指数对土壤背景的变化极为敏感，有利于对植被生态环境的监测，因此又被称为环境植被指数(EVI)。
土壤调整植被指数(SAVI)：
Figure PCTCN2018095485-appb-000004
其中，L是一个土壤调节系数，该系数与植被浓度有关，由实际区域条件确定，用来减小植被指数对不同土壤反射变化的敏感性。当L＝0时，SAVI＝NDVI；对于中等植被覆盖区，L的值一般接近于0.5。乘法因子(1+L)主要是用来保证最后的SAVI值介于-1和1之间。该指数能够降低土壤背景的影响，但可能丢失部分植被信号，使植被指数偏低。
上述现场火警记录数据即为上述森林区域的消防队的出警记录数据，一般包括时间、地点、着火原因等。
上述地形数据，一般是通过对上述森林区域的卫星遥感图像解析得到，其具体包括如下表所示数据：
Figure PCTCN2018095485-appb-000005
上述地形矢量数据(Digital Line Graphic，简称DLG)同样是通过对上述森林区域的卫星遥感图像 解析得到。地形矢量数据是现有地形图上基础地理要素的矢量数据集，包括水系、居民地及设施、交通、管线、境界与政区、地貌与土质、植被、地名及注记9个数据集，且保存要素间空间关系和相关的属性信息。在本实施例中，地形矢量数据包括的要素为水系、管线、境界与政区、地貌与土质、植被、地名及注记，以及要素间的空间关系等。
本实施例中，上述将所述火点信息数据、火像元数据、着火日期数据、火灾季节的烧伤疤痕数据和现场火警记录数据在所述森林火灾预测模型中进行时间序列分析，将地形数据和地形矢量数据在所述森林火灾预测模型中进行空间序列分析的步骤S21之前，包括：
S22、判断所述现场火警记录数据中是否存在人为放火的记录数据；
S23、如果存在人为放火的记录数据，则从所述现场火警记录数据中将所述人为放火的记录数据清除。
如上述步骤S22所述，火警记录中般会记录有森林火灾的着火原因，如干燥引起的火灾、雷电引起的火灾，以及人为引起的火灾等。比如，火警记录为2016年8月1日**地出警，火灾原因是农户烧秸秆而引起，此时可以通过语义分析方法解析火警记录的含义为人为的烧秸秆而引起火灾，进而记为人为放火的记录数据。
如上述步骤S23所述，将上述对应人为放火的记录数据清除，保留下自然火灾的出警数据，人为因素是不可控的随机因素，将其剔除可以提高森林火灾预测的准确性。
在本实施例中，上述将所述指定数据输入到预设的基于ConvLSTM模型的森林火灾预测模型进行运算的步骤S2之前，包括：
S201、将获取的所述指定数据进行归一化处理，得到标准化的所述指定数据。
如上述步骤S201所述，上述的归一化处理是指将有量纲的表达式，经过变换，化为无量纲的表达式，成为标量。本实施例中，即为将各不同的数据转换为相同表达的标量，以方便计算。
在一具体实施例中，上述将所述指定数据输入到预设的基于ConvLSTM模型的森林火灾预测模型进行运算的步骤S2之前，包括：
S202、获取所述森林区域的地理位置；
S203、根据所述森林区域的地理位置调用对应所述森林区域的地理位置的所述森林火灾预测模型，其中，所述森林火灾预测模型包括多个，不同的森林火灾预测模型是针对不同地理位置，并基于ConvLSTM模型建立的森林火灾预测模型。
如上述步骤S202和S203所述，上述地理位置即为具体的经度和纬度，可以通过森林区域所处的经纬度确定森林区域的环境特点，比如，中华人民共和国的南方与北方的气候不同，如果使用同一个森林火灾模型进行森林火灾进行预测，其预测的结果可能存在差异，本实施例中，设置两个或多个以基于ConvLSTM模型建立的森林火灾预测模型，不同的森林火灾预测模型，是基于不同的地理位置的森林区 域的指定数据进行训练学习得到的，即通过同样的建模方法分别采集不同的地理位置的相关的建模数据，然后进行建模。在森林火警预测时，针对不同的地区，调用对应的火灾预测模型，以得到更为准确的森林火灾预测结果。
在另一具体实施例中，上述将所述指定数据输入到预设的基于ConvLSTM模型的森林火灾预测模型进行运算的步骤S2之前，包括：
S204、获取所述森林区域的地貌属性；
S205、根据所述森林区域的地貌属性调用对应所述森林区域的地貌属性的所述森林火灾预测模型，其中，所述森林火灾预测模型包括多个，不同的森林火灾预测模型是针对不同地貌属性，并基于ConvLSTM模型建立的森林火灾预测模型。
如上述步骤S204和S205所述，上述地貌即地球表面各种形态的总称，也能称为地形。地表形态是多种多样的，成因也不尽相同，是内、外力地质作用对地壳综合作用的结果。内力地质作用造成了地表的起伏，控制了海陆分布的轮廊及山地、高原、盆地和平原的地域配置，决定了地貌的构造格架。而外营力(流水、风力、太阳辐射能、大气和生物的生长和活动)地质作用，通过多种方式，对地壳表层物质不断进行风化、剥蚀、搬运和堆积，从而形成了现代地面的各种形态。地貌属性即为地貌的具体形态，如山地、高原、盆地等。本实施例，可以通过森林区域地貌属性特点调用与其对应的森林火灾预测模型使用，比如，大兴安岭的森林区域的地貌属性与西双版纳的森林区域的地貌属性不同，如果使用同一个森林火灾模型进行森林火灾预测，其预测的结果可能存在差异，本实施例中，设置两个或多个针对不同地貌属性森林火灾预测模型，不同的森林火灾预测模型，是基于不同的地貌的森林区域的指定数据进行训练学习得到的，即通过同样的建模方法分别采集不同的地貌属性的相关的建模数据，然后进行建模。在森林火警预测时，针对不同的地貌，调用对应的火灾预测模型，以得到更为准确的森林火灾预测结果。
本实施例中，建立上述森林火灾预测模型的过程如下：
提取指定森林区域内的，指定历史时间段内的历史卫星遥感图像，并分别对每一张历史卫星遥感图像进行解析，以得到对应的地形数据、地形矢量数据、植被指数数据和火掩膜数据。然后通过火掩膜数据得出对应的火点信息；通过植被指数数据得出火灾季节的烧伤疤痕数据。还需要获取该指定森林区域的上述指定历史时间段内的现场火警记录数据和火烧迹地数据集；其中，通过火烧迹地数据集解析出各火烧迹地的火像元数据和着火日期数据；还会获取上述指定历史时间段内的各时间的历史火险系数数据。
将获取的建模数据分为训练集和测试集，然后通过训练集对应的样本数据输入到ConvLSTM模型中进行训练，在训练过程中，将各历史火险系数数据对应的火点信息数据、火像元数据、着火日期数据和火灾季节的烧伤疤痕数据进行时间序列分析，地形数据和地形矢量数据进行空间序列分析，最终得到一个基于ConvLSTM模型的森林火灾预测模型，即输入火点信息数据、火像元数据、着火日期数据、火灾季节的烧伤疤痕数据、地形数据和地形矢量数据到森林火灾预测模型中，该森林火灾预测模型会输 出一个火险系数。然后通过测试集验证森林火灾预测模型的有效性，若验证通过，则可以将该森林火灾预测模型进行实际应用，即预测森林火灾。
本实施例中，在建立模型时，还获取拍摄各历史卫星遥感图像时的气候数据，如季节、空气温度、空气湿度、晴天、雨天等数据，在建模时，这些数据主要进行时间序列的分析。在进行森林火灾预测时，对应获取气候数据，然后输入到森林火灾预测模型中即可。
本实施例的森林火灾预测方法，充分利用ConvLSTM模型的时序建模能力以及时空特性，建立一个兼顾时间序列和空间序列的火灾预测模型，从而更加准确的进行森林火灾预测。
参照图4，本申请实施例还提供一种森林火灾预测装置，包括步骤：
获取单元10，用于获取待预测森林火灾的森林区域的指定数据；
运算单元20，用于将所述指定数据输入到预设的基于ConvLSTM模型的森林火灾预测模型进行运算；
确定单元30，用于根据运算结果确定所述森林区域的火险系数。
如上述获取单元10，上述森林区域即为生长有树木等植物的区域，其一般为人类稀少，而树木等植物较多的区域，如大兴安岭的原始森林、长白上的某一区域的森林等。上述指定数据是指可能反映森林区域火灾情况的数据，具体包括：对应上述森林区域的火点信息数据、火像元数据、着火日期数据、火灾季节的烧伤疤痕数据、现场火警记录数据、地形数据以及地形矢量数据等。在其它实施例中，还可以包括气象数据等。
如上述运算单元20，上述ConvLSTM模型其不仅具有LSTM的时序建模能力，而且还能像CNN一样刻画局部特征，可以说是同时具备时空特性。ConvLSTM模型的内部结构如图2所示，其工作原理可以由以下公式表示：
i t＝σ(W xi*X t+W hi*H t-1+W ciо C t-1+b i)
f t＝σ(W xf*X t+W hf*H t-1+W cfо C t-1+b f)
C t＝f tо C t-1+i tо tanh(W xc*X t+W hc*H t-1+b c)
o t＝σ(W xo*X t+W ho*H t-1+W coо C t+b o)
H t＝o tо tanh(C t)
上述各式子中，*表示卷积，空心小圆圈表示矩阵对应元素相乘，又称为Hadamard乘积。值得注意的是，这里的χ,C,H,i,f,o都是三维的tensor，它们的后两个维度代表行和列的空间信息，可以把ConvLSTM模型想象成是处理二维网格中的特征向量的模型，其可以根据网格中周围点的特征来预测中心网格的特征。编码预测ConvLSTM网络则如图3所示。
如上述确定单元30，上述火险系数是指上述森林区域发生火灾的系数，即发生森林火灾的概率，一般为森林火险等级。
参照图5，在本实施例中，上述运算单元20，包括：
运算模块21，用于将所述火点信息数据、火像元数据、着火日期数据、火灾季节的烧伤疤痕数据和现场火警记录数据在所述森林火灾预测模型中进行时间序列分析，将地形数据和地形矢量数据在所述森林火灾预测模型中进行空间序列分析。
如上述运算模块21，上述火点信息数据、火像元数据、着火日期数据、火灾季节的烧伤疤痕数据和现场火警记录数据等与发生日期相关，即有时序上的性质，因此将其进行时序方面的分析。上述地形数据和地形矢量数据一般为空间数据，如高度、面积等，所以将地形数据和地形矢量数据进行空间序列分析。在一具体实施例中，首先对上述火点信息数据、火像元数据、着火日期数据、火灾季节的烧伤疤痕数据、现场火警记录数据、地形数据以及地形矢量数据等各数据进行分类，将带有时间信息的数据分成一类(火点信息数据、火像元数据、着火日期数据、火灾季节的烧伤疤痕数据和现场火警记录数据)，带有空间信息的数据分成一类(地形数据和地形矢量数据)，然后分别将对应的数据输入到上述森林火灾预测模型中进行计算，因为森林火灾预测模型是基于ConvLSTM模型的模型，可以针对性地对带有时间信息的数据进行时序方面的分析，对带有空间信息的数据进行空间序列方面的分析，关于数据输入到ConvLSTM模型中进行计算是本领域的公知常识，在此不在赘述。
在一具体实施中，上述火点信息数据是通过解析上述森林区域的火掩膜数据得到的。掩膜是一种图像滤镜的模板，经常用其处理的是遥感图像。本申请中，上述火掩膜数据即为用于提取着火点的图像滤镜模板，具体过程为：当提取卫星遥感图像的着地物信息时，通过一个n*n的矩阵来对图像进行像素过滤，然后将我们需要的火点信息显示出来。火掩膜数据集MOD14A1分为10个级别，如下表所示：
级别 对应具体内容
0 未处理像元
1 未处理像元
2 未处理像元
3 水域
4 云
5 没有火的裸地
6 未知像元
7 低置信度检测火点
8 中置信度检测火点
9 高置信度检测火点
提取属性值8和9的像元，将其各自与研究区域的植被覆盖图进行叠加，去掉伪森林火点信息，然后与对应时段的植被信息进行比对，将着火日期前后植被明显变化的像元确定为火点，最后使用8+9数据集确定火点信息。
上述火像元数据和着火日期数据是通过分析火烧迹地数据集而得到的。上述火烧迹地是指森林中经火灾烧毁后尚未长起新林的土地。上述火烧迹地数据集是指上述森林区域的火烧迹地的集合，火烧迹地的集合中对应每一个火烧迹地，记录有火像元数据和着火日期数据。像元代表地面的面积是多少。遥感卫星的飞行高度一般在4000千米～600千米之间，图像分辨率一般从1千米～1米之间。像元相当于电视屏幕上的一个点(电视是由若干个点组成的图像画面)，相当于计算机显示屏幕上的一个象素，相当于一群举着不同色板拼成画图的人中的一个。当分辨率为1千米时，一个像元代表地面1千米X1千米的面积，即1平方千米；当分辨率为30米时，一个像元代表地面30米×30米的面积；当分辨率为1米时，也就是说，图像上的一个像元相当于地面1米x1米的面积，即1平方米。上述火像元即为表示着火迹地的面积一个单位，若一个火像元为1平方米，则可以通过多少个火像元来描述火烧迹地的面积。上述着火日期数据是指火烧迹地发生火灾时的日期数据。
上述火灾季节的烧伤疤痕数据是通过分析上述森林区域的植被指数数据得到的。植被指数是植被分布密度和生长状态的指示因子，与植被覆盖呈正相关关系。而上述的火灾季节的烧伤疤痕数据是指森林火灾留下的痕迹，即该痕迹区域上的植被指数要远低于其周围的植被指数，但是随着时间的过度，上述痕迹区域会不断的恢复植被的生长状态，即痕迹区域的植被指数升高，所以可以通过森林区域的植被指数数据分析得到火灾季节的烧伤疤痕数据。上述植被指数数据是根据植被的光谱特性，将卫星可见光和近红外波段进行组合,形成了各种植被指数。具体的：植被指数，根据植被的光谱特性，将卫星可见光和近红外波段进行组合,形成了各种植被指数。植被指数是对地表植被状况的简单、有效和经验的度量。在遥感应用领域，植被指数已广泛用来定性和定量评价植被覆盖及其生长活力。由于植被光谱表现为植被、土壤亮度、环境影响、阴影、土壤颜色和湿度复杂混合反应，而且受大气空间-时相变化的影响，因此植被指数没有一个普遍的值，其研究经常表明不同的结果。该指数随生物量的增加而迅速增大。在植被处于中、低覆盖度时，该指数随覆盖度的增加而迅速增大，当达到一定覆盖度后增长缓慢，所以适用于植被早、中期生长阶段的动态监测。植被指数具体包括：
归一化植被指数(NDVI)：
Figure PCTCN2018095485-appb-000006
其中，ρ NIR和ρ RED分别代表近红外波段和红光波段的反射率。NDV1值介于-1和1之间。
增强型植被指数(EVI)：
Figure PCTCN2018095485-appb-000007
ρ NIR、ρ RED、ρ BLUE分别代表近红外波段、红外波段、蓝光波段的反射率。
比值植被指数(RVI)：
Figure PCTCN2018095485-appb-000008
ρ NIR和ρ RED分别代表近红外波段、红外波段的反射率。
差值植被指数(DVI):
DVI＝ρ NIR-ρ RED
ρ NIR和ρ RED分别代表近红外波段、红外波段的反射率。差值植被指数对土壤背景的变化极为敏感，有利于对植被生态环境的监测，因此又被称为环境植被指数(EVI)。
土壤调整植被指数(SAVI)：
Figure PCTCN2018095485-appb-000009
其中，L是一个土壤调节系数，该系数与植被浓度有关，由实际区域条件确定，用来减小植被指数对不同土壤反射变化的敏感性。当L＝0时，SAVI＝NDVI；对于中等植被覆盖区，L的值一般接近于0.5。乘法因子(1+L)主要是用来保证最后的SAVI值介于-1和1之间。该指数能够降低土壤背景的影响，但可能丢失部分植被信号，使植被指数偏低。
上述现场火警记录数据即为上述森林区域的消防队的出警记录数据，一般包括时间、地点、着火原因等。
上述地形数据，一般是通过对上述森林区域的卫星遥感图像解析得到，其具体包括如下表所示数据：
Figure PCTCN2018095485-appb-000010
上述地形矢量数据(Digital Line Graphic，简称DLG)同样是通过对上述森林区域的卫星遥感图像解析得到。地形矢量数据是现有地形图上基础地理要素的矢量数据集，包括水系、居民地及设施、交通、管线、境界与政区、地貌与土质、植被、地名及注记9个数据集，且保存要素间空间关系和相关的属性信息。在本实施例中，地形矢量数据包括的要素为水系、管线、境界与政区、地貌与土质、植被、地名 及注记，以及要素间的空间关系等。
参照图6，本实施例中，上述运算单元20，包括：
判断模块22，用于判断所述现场火警记录数据中是否存在人为放火的记录数据；
清除模块23，用于如果存在人为放火的记录数据，则从所述现场火警记录数据中将所述人为放火的记录数据清除。
如上述判断模块22，火警记录中般会记录有森林火灾的着火原因，如干燥引起的火灾、雷电引起的火灾，以及人为引起的火灾等。比如，火警记录为2016年8月1日**地出警，火灾原因是农户烧秸秆而引起，此时可以通过语义分析方法解析火警记录的含义为人为的烧秸秆而引起火灾，进而记为人为放火的记录数据。
如上述清除模块23，将上述对应人为放火的记录数据清除，保留下自然火灾的出警数据，人为因素是不可控的随机因素，将其剔除可以提高森林火灾预测的准确性。
参照图7，在本实施例中，上述森林火灾预测装置，包括：
归一化单元201，用于将获取的所述指定数据进行归一化处理，得到标准化的所述指定数据。
如上述归一化单元201，上述的归一化处理是指将有量纲的表达式，经过变换，化为无量纲的表达式，成为标量。本实施例中，即为将各不同的数据转换为相同表达的标量，以方便计算。
在一具体实施例中，上述森林火灾预测装置，包括：
第一获取单元，用于获取所述森林区域的地理位置；
第一调用单元，用于根据所述森林区域的地理位置调用对应所述森林区域的地理位置的所述森林火灾预测模型，其中，所述森林火灾预测模型包括多个，不同的森林火灾预测模型是针对不同地理位置，并基于ConvLSTM模型建立的森林火灾预测模型。
如上述第一获取单元和第一调用单元，上述地理位置即为具体的经度和纬度，可以通过森林区域所处的经纬度确定森林区域的环境特点，比如，中华人民共和国的南方与北方的气候不同，如果使用同一个模型进行森林火灾模型进行预测，其预测的结果可能存在差异，本实施例中，设置两个或多个以基于ConvLSTM模型建立的森林火灾预测模型，不同的森林火灾预测模型，是基于不同的地理位置的森林区域的指定数据进行训练学习得到的，即通过同样的建模方法分别采集不同的地理位置的相关的建模数据，然后进行建模。在森林火警预测时，针对不同的地区，调用对应的火灾预测模型，以得到更为准确的森林火灾预测结果。
在另一具体实施例中，上述森林火灾预测装置，包括：
第二获取单元，用于获取所述森林区域的地貌属性；
第二调用单元，用于根据所述森林区域的地貌属性调用对应所述森林区域的地貌属性的所述森林火灾预测模型，其中，所述森林火灾预测模型包括多个，不同的森林火灾预测模型是针对不同地貌属性， 并基于ConvLSTM模型建立的森林火灾预测模型。
如上述第二获取单元和第二调用单元，上述地貌即地球表面各种形态的总称，也能称为地形。地表形态是多种多样的，成因也不尽相同，是内、外力地质作用对地壳综合作用的结果。内力地质作用造成了地表的起伏，控制了海陆分布的轮廊及山地、高原、盆地和平原的地域配置，决定了地貌的构造格架。而外营力(流水、风力、太阳辐射能、大气和生物的生长和活动)地质作用，通过多种方式，对地壳表层物质不断进行风化、剥蚀、搬运和堆积，从而形成了现代地面的各种形态。地貌属性即为地貌的具体形态，如山地、高原、盆地等。本实施例，可以通过森林区域地貌属性特点调用与其对应的森林火灾预测模型使用，比如，大兴安岭的森林区域的地貌属性与西双版纳的森林区域的地貌属性不同，如果使用同一个森林火灾模型进行森林火灾预测，其预测的结果可能存在差异，本实施例中，设置两个或多个针对不同地貌属性森林火灾预测模型，不同的森林火灾预测模型，是基于不同的地貌的森林区域的指定数据进行训练学习得到的，即通过同样的建模方法分别采集不同的地貌属性的相关的建模数据，然后进行建模。在森林火警预测时，针对不同的地貌，调用对应的火灾预测模型，以得到更为准确的森林火灾预测结果。
本实施例中，建立上述森林火灾预测模型的过程如下：
提取指定森林区域内的，指定历史时间段内的历史卫星遥感图像，并分别对每一张历史卫星遥感图像进行解析，以得到对应的地形数据、地形矢量数据、植被指数数据和火掩膜数据。然后通过火掩膜数据得出对应的火点信息；通过植被指数数据得出火灾季节的烧伤疤痕数据。还需要获取该指定森林区域的上述指定历史时间段内的现场火警记录数据和火烧迹地数据集；其中，通过火烧迹地数据集解析出各火烧迹地的火像元数据和着火日期数据；还会获取上述指定历史时间段内的各时间的历史火险系数数据。
将获取的建模数据分为训练集和测试集，然后通过训练集对应的样本数据输入到ConvLSTM模型中进行训练，在训练过程中，将各历史火险系数数据对应的火点信息数据、火像元数据、着火日期数据和火灾季节的烧伤疤痕数据进行时间序列分析，地形数据和地形矢量数据进行空间序列分析，最终得到一个基于ConvLSTM模型的森林火灾预测模型，即输入火点信息数据、火像元数据、着火日期数据、火灾季节的烧伤疤痕数据、地形数据和地形矢量数据到森林火灾预测模型中，该森林火灾预测模型会输出一个火险系数。然后通过测试集验证森林火灾预测模型的有效性，若验证通过，则可以将该森林火灾预测模型进行实际应用，即预测森林火灾。
本实施例中，在建立模型时，还获取拍摄各历史卫星遥感图像时的气候数据，如季节、空气温度、空气湿度、晴天、雨天等数据，在建模时，这些数据主要进行时间序列的分析。在进行森林火灾预测时，对应获取气候数据，然后输入到森林火灾预测模型中即可。
本实施例的森林火灾预测装置，充分利用ConvLSTM模型的时序建模能力以及时空特性，建立一个兼顾时间序列和空间序列的火灾预测模型，从而更加准确的进行森林火灾预测。
参照图8，本申请实施例中还提供一种计算机设备，该计算机设备可以是服务器，其内部结构可以如图8所示。该计算机设备包括通过系统总线连接的处理器、存储器、网络接口和数据库。其中，该计算机设计的处理器用于提供计算和控制能力。该计算机设备的存储器包括非易失性存储介质、内存储器。该非易失性存储介质存储有操作系统、计算机可读指令和数据库。该内存器为非易失性存储介质中的操作系统和计算机可读指令的运行提供环境。该计算机设备的数据库用于存储基于ConvLSTM模型的森林火灾预测模型等数据。该计算机设备的网络接口用于与外部的终端通过网络连接通信。该计算机可读指令被处理器执行时以实现上述各方法实施例的流程。
本申请一实施例还提供一种计算机非易失性可读存储介质，其上存储有计算机可读指令，计算机可读指令被处理器执行时实现上述各方法实施例的流程。
以上所述仅为本申请的优选实施例，并非因此限制本申请的专利范围，凡是利用本申请说明书及附图内容所作的等效结构或等效流程变换，或直接或间接运用在其他相关的技术领域，均同理包括在本申请的专利保护范围内。

Claims (20)

  1. 一种森林火灾预测方法，其特征在于，包括：
    获取待预测森林火灾的森林区域的指定数据；
    将所述指定数据输入到预设的基于ConvLSTM模型的森林火灾预测模型进行运算；
    根据运算结果确定所述森林区域的火险系数。
  2. 根据权利要求1所述的森林火灾预测方法，其特征在于，所述指定数据包括对应所述森林区域的火点信息数据、火像元数据、着火日期数据、火灾季节的烧伤疤痕数据、现场火警记录数据、地形数据以及地形矢量数据。
  3. 根据权利要求2所述的森林火灾预测方法，其特征在于，所述将所述指定数据输入到预设的基于ConvLSTM模型的森林火灾预测模型进行运算的步骤，包括：
    将所述火点信息数据、火像元数据、着火日期数据、火灾季节的烧伤疤痕数据和现场火警记录数据在所述森林火灾预测模型中进行时间序列分析，将地形数据和地形矢量数据在所述森林火灾预测模型中进行空间序列分析。
  4. 根据权利要求2的森林火灾预测方法，其特征在于，所述将所述火点信息数据、火像元数据、着火日期数据、火灾季节的烧伤疤痕数据和现场火警记录数据在所述森林火灾预测模型中进行时间序列分析，将地形数据和地形矢量数据在所述森林火灾预测模型中进行空间序列分析的步骤之前，包括：
    判断所述现场火警记录数据中是否存在人为放火的记录数据；
    如果存在人为放火的记录数据，则从所述现场火警记录数据中将所述人为放火的记录数据清除。
  5. 根据权利要求1所述的森林火灾预测方法，其特征在于，所述将所述指定数据输入到预设的基于ConvLSTM模型的森林火灾预测模型进行运算的步骤之前，包括：
    将获取的所述指定数据进行归一化处理，得到标准化的所述指定数据。
  6. 根据权利要求1所述的森林火灾预测方法，其特征在于，所述将所述指定数据输入到预设的基于ConvLSTM模型的森林火灾预测模型进行运算的步骤之前，包括：
    获取所述森林区域的地理位置；
    根据所述森林区域的地理位置调用对应所述森林区域的地理位置的所述森林火灾预测模型，其中，所述森林火灾预测模型包括多个，不同的森林火灾预测模型是针对不同地理位置，并基于ConvLSTM模型建立的森林火灾预测模型。
  7. 根据权利要求1所述的森林火灾预测方法，其特征在于，所述将所述指定数据输入到预设的基于ConvLSTM模型的森林火灾预测模型进行运算的步骤之前，包括：
    获取所述森林区域的地貌属性；
    根据所述森林区域的地貌属性调用对应所述森林区域的地貌属性的所述森林火灾预测模型，其中，所述森林火灾预测模型包括多个，不同的森林火灾预测模型是针对不同地貌属性，并基于ConvLSTM模型建立的森林火灾预测模型。
  8. 一种森林火灾预测装置，其特征在于，包括:
    获取单元，用于获取待预测森林火灾的森林区域的指定数据；
    运算单元，用于将所述指定数据输入到预设的基于ConvLSTM模型的森林火灾预测模型进行运算；
    确定单元，用于根据运算结果确定所述森林区域的火险系数。
  9. 根据权利要求8所述的森林火灾预测装置，其特征在于，所述指定数据包括对应所述森林区域的火点信息数据、火像元数据、着火日期数据、火灾季节的烧伤疤痕数据、现场火警记录数据、地形数据以及地形矢量数据。
  10. 根据权利要求9所述的森林火灾预测装置，其特征在于，所述运算单元，包括：
    运算模块，用于将所述火点信息数据、火像元数据、着火日期数据、火灾季节的烧伤疤痕数据和现场火警记录数据在所述森林火灾预测模型中进行时间序列分析，将地形数据和地形矢量数据在所述森林火灾预测模型中进行空间序列分析。
  11. 根据权利要求9的森林火灾预测装置，其特征在于，所述运算单元，包括：
    判断模块，用于判断所述现场火警记录数据中是否存在人为放火的记录数据；
    清除模块，用于如果存在人为放火的记录数据，则从所述现场火警记录数据中将所述人为放火的记录数据清除。
  12. 根据权利要求8所述的森林火灾预测装置，其特征在于，所述森林火灾预测装置，包括：
    归一化单元，用于将获取的所述指定数据进行归一化处理，得到标准化的所述指定数据。
  13. 根据权利要求8所述的森林火灾预测装置，其特征在于，所述森林火灾预测装置，包括：
    第一获取单元，用于获取所述森林区域的地理位置；
    第一调用单元，用于根据所述森林区域的地理位置调用对应所述森林区域的地理位置的所述森林火灾预测模型，其中，所述森林火灾预测模型包括多个，不同的森林火灾预测模型是针对不同地理位置，并基于ConvLSTM模型建立的森林火灾预测模型。
  14. 根据权利要求8所述的森林火灾预测装置，其特征在于，所述森林火灾预测装置，包括：
    第二获取单元，用于获取所述森林区域的地貌属性；
    第二调用单元，用于根据所述森林区域的地貌属性调用对应所述森林区域的地貌属性的所述森林火灾预测模型，其中，所述森林火灾预测模型包括多个，不同的森林火灾预测模型是针对不同地貌属性，并基于ConvLSTM模型建立的森林火灾预测模型。
  15. 一种计算机设备，包括存储器和处理器，所述存储器存储有计算机可读指令，其特征在于，所述处理器执行所述计算机可读指令时实现森林火灾预测方法，该森林火灾预测方法包括：
    获取待预测森林火灾的森林区域的指定数据；
    将所述指定数据输入到预设的基于ConvLSTM模型的森林火灾预测模型进行运算；
    根据运算结果确定所述森林区域的火险系数。
  16. 根据权利要求15所述的计算机设备，其特征在于，所述指定数据包括对应所述森林区域的火点信息数据、火像元数据、着火日期数据、火灾季节的烧伤疤痕数据、现场火警记录数据、地形数据以及地形矢量数据。
  17. 根据权利要求16所述的计算机设备，其特征在于，所述将所述指定数据输入到预设的基于ConvLSTM模型的森林火灾预测模型进行运算的步骤，包括：
    将所述火点信息数据、火像元数据、着火日期数据、火灾季节的烧伤疤痕数据和现场火警记录数据在所述森林火灾预测模型中进行时间序列分析，将地形数据和地形矢量数据在所述森林火灾预测模型中进行空间序列分析。
  18. 根据权利要求16的计算机设备，其特征在于，所述将所述火点信息数据、火像元数据、着火日期数据、火灾季节的烧伤疤痕数据和现场火警记录数据在所述森林火灾预测模型中进行时间序列分析，将地形数据和地形矢量数据在所述森林火灾预测模型中进行空间序列分析的步骤之前，包括：
    判断所述现场火警记录数据中是否存在人为放火的记录数据；
    如果存在人为放火的记录数据，则从所述现场火警记录数据中将所述人为放火的记录数据清除。
  19. 一种计算机非易失性可读存储介质，其上存储有计算机可读指令，其特征在于，所述计算机可读指令被处理器执行时实现森林火灾预测方法，该森林火灾预测方法包括：
    获取待预测森林火灾的森林区域的指定数据；
    将所述指定数据输入到预设的基于ConvLSTM模型的森林火灾预测模型进行运算；
    根据运算结果确定所述森林区域的火险系数。
  20. 根据权利要求19所述的计算机非易失性可读存储介质，其特征在于，所述指定数据包括对应所述森林区域的火点信息数据、火像元数据、着火日期数据、火灾季节的烧伤疤痕数据、现场火警记录数据、地形数据以及地形矢量数据。
PCT/CN2018/095485 2018-04-17 2018-07-12 森林火灾预测方法、装置、计算机设备和存储介质 WO2019200743A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201810345317.7 2018-04-17
CN201810345317.7A CN108717582A (zh) 2018-04-17 2018-04-17 森林火灾预测方法、装置、计算机设备和存储介质

Publications (1)

Publication Number Publication Date
WO2019200743A1 true WO2019200743A1 (zh) 2019-10-24

Family

ID=63899097

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/095485 WO2019200743A1 (zh) 2018-04-17 2018-07-12 森林火灾预测方法、装置、计算机设备和存储介质

Country Status (2)

Country Link
CN (1) CN108717582A (zh)
WO (1) WO2019200743A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2022085870A1 (ko) * 2020-10-23 2022-04-28 대한민국(산림청 국립산림과학원장) 산불 위험 지수 제공 장치 및 방법
KR20220102330A (ko) * 2021-01-13 2022-07-20 대한민국(산림청 국립산림과학원장) 산불 위험 예측 모델에 의한 산불 위험 지수 제공 장치 및 방법
CN117113854A (zh) * 2023-10-18 2023-11-24 珠江水利委员会珠江水利科学研究院 一种基于ConvLSTM和三维数值模拟的咸潮预报方法

Families Citing this family (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110119682A (zh) * 2019-04-04 2019-08-13 北京理工雷科电子信息技术有限公司 一种红外遥感图像火点识别方法
CN110220554A (zh) * 2019-06-05 2019-09-10 北京百度网讯科技有限公司 用于输出信息的方法和装置
CN111523277B (zh) * 2020-05-19 2022-08-16 珠海格力电器股份有限公司 火灾预测方法、存储装置、处理器
CN111625999A (zh) * 2020-05-29 2020-09-04 中南林业科技大学 一种基于深度学习技术的森林火灾预警模型及系统
CN113408776B (zh) * 2020-12-21 2023-03-28 电子科技大学 一种基于时间维特征增强的川西野火风险预警方法
CN112950880A (zh) * 2021-01-26 2021-06-11 特斯联科技集团有限公司 一种基于大数据的火灾预警方法及系统
CN113362553A (zh) * 2021-06-08 2021-09-07 内蒙古工业大学 基于多源异构数据的草原火灾监测预警方法及装置
CN114005237B (zh) * 2021-10-30 2023-03-28 南京林业大学 一种基于热成像分析技术的林火识别方法、设备及计算机存储介质
CN113903136B (zh) * 2021-12-13 2022-02-22 环球数科集团有限公司 一种森林火灾应急态势建模分析系统
CN114298392B (zh) * 2021-12-23 2023-05-12 电子科技大学 一种面向森林火灾的火线经纬度预测方法
CN117575320B (zh) * 2023-11-29 2024-04-30 中国科学院空天信息创新研究院 一种基于图卷积神经网络的森林火灾风险预测方法

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2003073128A1 (en) * 2001-05-30 2003-09-04 Instituto Superior Técnico Lidar system controlled by computer for smoke identification applied, in particular, to early stage forest fire detection
CN1529291A (zh) * 2003-10-16 2004-09-15 华南理工大学 森林火灾智能监测报警装置
US20160034812A1 (en) * 2014-07-31 2016-02-04 Qualcomm Incorporated Long short-term memory using a spiking neural network
CN105551168A (zh) * 2015-12-03 2016-05-04 中国科学院遥感与数字地球研究所 一种农田火灾遥感监测预警的方法及系统

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102663082A (zh) * 2012-04-06 2012-09-12 昆明理工大学 基于数据挖掘的森林火灾预测方法
JP5801277B2 (ja) * 2012-11-07 2015-10-28 株式会社パスコ 林野火災の予測装置及び林野火災の予測プログラム
CN103605899B (zh) * 2013-11-29 2017-05-24 民政部国家减灾中心 一种区域火灾风险估计方法
CN104615848A (zh) * 2014-12-26 2015-05-13 中国南方电网有限责任公司 一种集森林火险预报和地面火点监测于一体的综合应用方法

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2003073128A1 (en) * 2001-05-30 2003-09-04 Instituto Superior Técnico Lidar system controlled by computer for smoke identification applied, in particular, to early stage forest fire detection
CN1529291A (zh) * 2003-10-16 2004-09-15 华南理工大学 森林火灾智能监测报警装置
US20160034812A1 (en) * 2014-07-31 2016-02-04 Qualcomm Incorporated Long short-term memory using a spiking neural network
CN105551168A (zh) * 2015-12-03 2016-05-04 中国科学院遥感与数字地球研究所 一种农田火灾遥感监测预警的方法及系统

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2022085870A1 (ko) * 2020-10-23 2022-04-28 대한민국(산림청 국립산림과학원장) 산불 위험 지수 제공 장치 및 방법
KR20220102330A (ko) * 2021-01-13 2022-07-20 대한민국(산림청 국립산림과학원장) 산불 위험 예측 모델에 의한 산불 위험 지수 제공 장치 및 방법
KR102496878B1 (ko) * 2021-01-13 2023-02-07 대한민국 산불 위험 예측 모델에 의한 산불 위험 지수 제공 장치 및 방법
CN117113854A (zh) * 2023-10-18 2023-11-24 珠江水利委员会珠江水利科学研究院 一种基于ConvLSTM和三维数值模拟的咸潮预报方法
CN117113854B (zh) * 2023-10-18 2024-02-02 珠江水利委员会珠江水利科学研究院 一种基于ConvLSTM和三维数值模拟的咸潮预报方法

Also Published As

Publication number Publication date
CN108717582A (zh) 2018-10-30

Similar Documents

Publication Publication Date Title
WO2019200743A1 (zh) 森林火灾预测方法、装置、计算机设备和存储介质
Aboelnour et al. Application of remote sensing techniques and geographic information systems to analyze land surface temperature in response to land use/land cover change in Greater Cairo Region, Egypt
Dutta et al. Impervious surface growth and its inter-relationship with vegetation cover and land surface temperature in peri-urban areas of Delhi
Serneels et al. Land cover changes around a major east African wildlife reserve: the Mara Ecosystem (Kenya)
Roy et al. Monitoring and prediction of land use/land cover change using the integration of Markov chain model and cellular automation in the Southeastern Tertiary Hilly Area of Bangladesh
Ares et al. Detection of process‐related changes in plant patterns at extended spatial scales during early dryland desertification
Pu et al. Oakwood crown closure estimation by unmixing Landsat TM data
Singh Geospatial technique for land use/land cover mapping using multi-temporal satellite images: A case study of Samastipur District (India)
Polinova et al. Reconstructing pre-fire vegetation condition in the wildland urban interface (WUI) using artificial neural network
Philippe et al. Assessing Forest cover change and deforestation hot-spots in the north Kivu Province, DR-Congo using remote sensing and GIS
Marinho et al. Temporal record and spatial distribution of fire foci in State of Minas Gerais, Brazil
Erlandsson et al. An artificial intelligence approach to remotely assess pale lichen biomass
Spröhnle et al. Earth observation-based multi-scale impact assessment of internally displaced person (IDP) camps on wood resources in Zalingei, Darfur
Esam et al. Land use and land cover changes of west tahta region, sohag governorate, upper egypt
Liu et al. Open-air grape classification and its application in parcel-level risk assessment of late frost in the eastern Helan Mountains
Williams et al. Mapping the spatio-temporal evolution of irrigation in the Coastal Plain of Georgia, USA
Brundrett et al. A monitoring toolkit for banksia woodlands: comparison of different scale methods to measure recovery of vegetation after fire
Igulu et al. Monitoring impervious surface area dynamics to assess urbanisation of a catchment: Msimbazi River Valley Dar es Salaam, 1989-2015
Pathmanandakumar A study of land cover change detection in Oddusuddan DS Division of Mullaitivu District in Sri Lanka based on GIS and RS technology
Ikiel et al. Analysis of the changes in vegetation of the Sapanca Lake Basin (in Turkey) using multitemporal satellite data
Subramani et al. Land use and land cover change detection and urban sprawl analysis of Panamarathupatti Lake, Salem
Prior et al. Multi‐decadal stability of woody cover in a mesic eucalypt savanna in the Australian monsoon tropics
Al-Doski et al. Monitoring land cover changes in Halabja City, Iraq
Freyberg et al. Idaho Wildfires: Assessing Drought and Fire Conditions, Trends, and Susceptibility to Inform State Mitigation Efforts and Bolster Monitoring Protocol in North-Central Idaho
Ara et al. Urban Area Growth Monitoring in Sylhet City Using Remote Sensing and Geographic Information System from 2002 to 2017

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18915249

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18915249

Country of ref document: EP

Kind code of ref document: A1