CN108230310A - A kind of method that non-fire space-time data is extracted based on semivariable function - Google Patents

Abstract

A kind of method that non-fire space-time data is extracted based on semivariable function, the present invention determines the position of all fire pixels in the data of combustion zone and the moment occurs first, and judge the vegetation pattern of all fire pixels, then the data of all fire pixels and the non-fire pel data that surrounding is identical with fire pixel vegetation pattern are extracted, spatial Correlation Analysis is carried out to it using semivariable function, and establish corresponding fire buffering area by Cheng Zuowei buffer distances are become, then the data of fire pixel correlation phase are extracted again, the process of fire buffering area is established in repetition, the data for finally selecting not falling within the pixel in all buffering areas are as non-fire data.The present invention can eliminate the subjective impact that human factor is brought, otherness caused by the factors such as geographical location, vegetation pattern can be eliminated again, the accidental sexual factor of fire generation is also eliminated simultaneously, the comparative of non-fire data and fire data is improved, is conducive to improve the precision of the analog result of training pattern.

Description

A method of extracting non-fire spatio-temporal data based on semivariogram

technical field

The invention particularly relates to a method for extracting non-fire spatio-temporal data based on a semivariogram.

Background technique

Forest fire is a very common and extremely destructive natural disaster. There are more than 200,000 forest fires in the world every year on average, and the burned forest area accounts for more than 1‰ of the total forest area in the world. On average, more than 10,000 forest fires occur in China every year, burning hundreds of thousands to millions of hectares of forests, accounting for about 5-8 ‰ of the country's forest area. Forest fires not only kill and injure forest trees, directly reduce forest area, but also seriously damage forest structure and forest environment, leading to the loss of forest ecosystem, forest biomass decline, and even human and livestock casualties.

It is very meaningful and necessary to predict the level of forest fire danger. Remote sensing technology can realize large-scale, high-spatial-resolution forest fire danger level prediction, which has certain scientific implications for the allocation of forest fire protection resources and personnel mobilization. Guiding significance. At present, the research on the prediction model of fire danger level has been relatively in-depth. The training model generally needs to input climate parameters, terrain parameters and vegetation parameters, among which the vegetation parameters can reflect the physical characteristics of vegetation (such as type, coverage, etc.). The training model requires comparative data, that is, the input data of the model includes not only the data of the fire (fire data), but also the data of the non-fire (non-fire data). The quality of the input data has an important impact on the final simulation effect of the trained model.

At present, the method of obtaining fire data and non-fire data based on remote sensing technology is mainly through the following steps: first, map the fire burning area according to the fire point extraction algorithm or directly use related fire products, and then extract the fire image according to the mapping results or fire products In this way, the fire data can be obtained; while the non-fire data is usually centered on the fire pixel, and the buffer radius is determined based on experience to establish a buffer zone, and a non-fire pixel is randomly extracted from the area outside the buffer zone. The data is used as the contrast data of the fire pixel. Although this method of non-fire data extraction is simple to operate, there are many problems: First, the determination of the buffer radius of fire pixels is based on experience, and there are subjective effects brought by human factors. At the same time, the size of the buffer radius varies with the research area. , vegetation types, etc., which means that the buffer radius of each fire pixel is different. If the buffer radius is set too small, there will be a gap between the extracted non-fire data and the corresponding fire data. Spatial correlation, thus lacking contrast, affects the simulation effect of the training model and cannot achieve the purpose of prediction well; if the buffer radius is set too large, the area of non-fire data extraction will be too small. Second, the method of extracting non-fire data only considers the spatial correlation and does not consider the temporal correlation. The fire data are close, so the non-fire data extracted by the above method still lacks comparison. To sum up, how to extract non-fire data accurately, conveniently and efficiently has become an urgent technical problem in this field.

Contents of the invention

In view of the above, the object of the present invention is to: aim at the problems of poor contrast in the existing non-fire data extraction methods, and provide a method for combining semivariogram fitting with spherical models and multi-temporal features of remote sensing images to extract non-fire data. The method of fire spatio-temporal data, the present invention can not only eliminate the subjective influence brought by human factors, but also eliminate the differences caused by factors such as geographical location and vegetation type, and improve the contrast between non-fire data and fire data, which is beneficial to improve The accuracy of the simulation results for the trained model.

The technical scheme provided by the invention is as follows:

A method for extracting non-fire data based on semivariogram, characterized in that it comprises the steps of:

Step 1: Determine the location and occurrence time of the fire pixel in the burning area data;

Step 2: Determine the vegetation type of the fire pixel;

Step 3: Extract the data of the fire pixel and the surrounding non-fire pixel data of the same vegetation type as the fire pixel;

Step 4: Use the semivariogram to analyze the spatial correlation of the fire pixels, use the variable range as the buffer distance of the fire pixels and establish the corresponding fire buffer;

Step 5: Extract the data of fire pixel related time phase;

Step 6: Repeat step 4 to establish the fire buffer zone of the time phase related to the fire pixel;

Step 7: Select the pixels that do not fall into the fire buffer zone obtained in steps 4 and 6 and extract their data as non-fire data.

Further, the range in the present invention is obtained by fitting the semivariogram value using a spherical model.

Furthermore, the establishment of the buffer zone in the present invention is specifically based on the fire pixel as the center and the range as the radius to form the buffer zone.

Further, the relevant phases of the fire pixel in the present invention are specifically: taking a single fire pixel as the research object, and the moment when the pixel is on fire as the target phase, then the adjacent phases of the target phase and the other phases The phase corresponding to the year and the target time is the relevant phase.

Further, the burning area data in the present invention is specifically remote sensing image data.

The concept of the present invention is described in detail below: in order to consider the spatial correlation between fire pixels and non-fire pixels, it is often necessary to establish a buffer area for fire pixels. However, there are often multiple fire pixels in the remote sensing image of fire distribution, so it is necessary to find out the buffers of all the fire pixels, and make the comparison data of the fire pixels not fall in any buffer. Due to the differences between vegetation types, the present invention introduces vegetation type factors in the process of establishing the buffer zone, and uses the spherical model to fit the variation trend of the semivariogram value, and uses the range as the threshold distance of spatial correlation to establish a fire image. At the same time, the time correlation of fire pixels in different phases is considered, avoiding the extraction of non-fire pixel data close to the fire pixel data, thereby improving the difference between fire data and non-fire data Finally, the purpose of improving the simulation accuracy of the fire danger level prediction model is achieved.

Compared with the beneficial effects of the prior art the present invention is:

The present invention introduces the vegetation type factor, uses the variable range of the semivariogram as the buffer radius to establish the buffer zone of the fire pixel, thereby not only eliminating the subjective influence brought by human factors, but also eliminating the differences caused by different geographical locations and vegetation types. etc.; at the same time, the present invention takes into account the time correlation of fire pixels in different time phases, thereby eliminating non-fire data that are not sufficiently different from fire data; therefore, the present invention can enhance the fire data The comparison with non-fire data improves the simulation accuracy of the fire danger level prediction model.

Description of drawings

Figure 1 is a schematic diagram of the fire distribution image and the fire pixel buffer radius in Mile County, Yunnan Province.

Fig. 2 is a schematic diagram of the semivariogram value change of the normalized difference vegetation index (Normalized Difference Vegetation Index, NDVI).

Fig. 3 is a schematic diagram of the semivariogram value change of fuel moisture content (Fuel Moisture Content, FMC).

Figure 4 is a schematic diagram of the buffer superposition of different indicator parameters.

Figure 5 is a schematic diagram of the buffer overlay of different fire pixels.

Fig. 6 is a schematic diagram of multi-temporal buffer superposition.

Detailed ways

The principle and characteristics of the present invention are further described below in conjunction with the accompanying drawings and specific embodiments of the description:

The occurrence and development of fire is a very complex process involving many influencing factors. Therefore, it is necessary to select different and representative fire risk assessment indicators according to different time, space scales and research purposes when evaluating fire danger levels. At present, commonly used fire risk indicators can be roughly divided into topographical conditions, weather environment, and vegetation factors, among which vegetation factors are one of the main factors affecting the maturity of vegetation fires.

The present invention selects two important factors - fuel moisture content (Fuel Moisture Content, FMC) and normalized difference vegetation index (Normalized Difference Vegetation Index, NDVI) as the training data of the fire danger level prediction model. FMC refers to the ratio obtained by dividing the difference between the wet weight and dry weight of the plants in the sampling unit by the dry weight, which reflects the water content of the leaves of the unit vegetation, and its level directly affects the level of fire danger. It is a very important indicator. The fire induction factor can be obtained through the inversion of the vegetation radiative transfer model in the prior art. NDVI is a common vegetation index, which can reflect the growth status and vegetation coverage of vegetation. It can be obtained by calculating the first and second bands of remote sensing satellite reflectance products (MOD09A1). The following will take these two vegetation parameters as an example to describe the implementation of the present invention in detail:

Example:

A method for extracting non-fire data based on a semivariogram, comprising the steps of:

Step 1: Data preparation;

This embodiment uses Mile County, Yunnan Province in January 2016 to describe in detail:

This embodiment uses the remote sensing satellite product (MCD64A1) to provide fire distribution images, and this product records the time and place of fire occurrences every month; due to the differences between vegetation types, different vegetation types should be processed separately. This embodiment uses The land cover product (MCD12Q1) provided by the remote sensing satellite is used to judge the vegetation type of each fire pixel.

Step 2: Spatial decorrelation;

According to the statistics of MCD64A1 fire distribution products, in January 2016, a total of 46 pixels were on fire in Mile County. In this embodiment, the buffer radius is calculated for each pixel according to FMC and NDVI, and then the fire buffer zone is established. As shown in Figure 1, the left image is the image of fire distribution in Mile County, Yunnan Province, and the right image is a schematic diagram of the buffer radius of a certain fire pixel. The dark squares in the circle represent the fire-occurring pixels, the light-colored squares in the circle and the light-colored squares outside the circle represent the fire-free pixels, and the area occupied by the circle represents the buffer zone. The cells inside the circle are in the buffer zone, so they cannot be selected as non-fire cells, and only the cells outside the buffer range can be selected as non-fire cells. In the following, only the operation of creating a buffer zone for one of the fire pixels will be described in detail, and the processing of the rest of the fire pixels is the same.

The present invention mainly carries out spatial decorrelation processing based on semivariogram:

The semivariogram function, also known as the semivariogram function, is a key function in the study of soil variability in geostatistics, and it is also a continuous function used to describe the spatial continuous variation of soil properties. It can reflect the relationship between different distance observations of soil properties. The basic formula is shown in formula (1) below.

Among them, R _(h) is the semivariogram, N is the number of pixels h away from the fire point pixel, Z _(x) is the pixel value of the fire point pixel, and Z _(xi+h) is the distance from the fire point The cell value at cell h.

In this example, a spherical model is used to fit the semivariogram values. The spherical model is also known as the Matron model. It is well known in the art that the distance at which the model first appears horizontal is the range, and the sample positions separated by distances closer to the range are spatially autocorrelated, while the sample positions farther than the range are not spatially autocorrelated, Its basic formula is shown in formula (2).

Among them, Y _h is the independent variable, h is the dependent variable, C ₀ , C, and a are the model parameters, and the parameter a represents the variable range.

Calculate the semivariogram values of FMC and NDVI as the distance increases according to formula (1). The change form of the semivariogram value of NDVI as the distance increases is shown in Figure 2. The semivariogram value of FMC presents as the distance increases. The variation form is shown in Fig. 3; then the model parameter a (range) is obtained by fitting the formula (2), so that the variation range of NDVI is 33 pixel distances, and the FMC variation range is 35 pixel distances. At this time, the fire pixel has two buffer radii, and the larger buffer radius is selected to establish the fire buffer zone of the fire pixel, that is, the superposition diagram of the fire buffer zone is obtained as shown in Figure 4. For the same fire pixel, The inner circle is the buffer of NDVI, and the outer circle is the buffer of FMC. Similarly, repeat the above process for the rest of the fire pixels, and then establish the corresponding fire buffer zone, then you can get the superposition diagram of the fire buffer zone as shown in Figure 5, and the selection of non-fire pixels cannot fall in any buffer zone .

Step 3: Time decorrelation;

In actual situations, the occurrence of a fire has a strong chance. For a certain pixel, there is no fire in this phase, and it is not in the buffer zone of any fire pixel in this phase, but its data shows that the fire danger level of this pixel is very high, probably because of the A fire broke out in the adjacent time phase or the corresponding time phase in other years. If the fire occurrences of the pixel in the corresponding time phases of adjacent phases and other years are not considered, it is possible to select this pixel as a non-fire pixel, thereby reducing the difference between fire and non-fire data.

The present invention is based on the multi-temporal features of remote sensing images, takes months as the time scale, takes January 2016 as the target phase, and sets December 2015, February 2016, and 2001 to 2017 (MCD64A1 products started in November 2000 ) every January as the relevant time phase, extract the fire distribution images of the above relevant time phases, and then repeat the operation of step 2 to establish the fire buffer zone of the fire pixels in each of the above relevant time phases. As shown in Figure 6, the fire buffer zone obtained in step 2 and the fire buffer zone obtained in step 3 are superimposed to obtain an image, in which figure (a) is a schematic diagram of the buffer zone of the target phase, and figure (b) is the previous phase of the target phase Schematic diagram of the buffer zone in the time phase (i.e. December 2015), figure (c) is a schematic diagram of the buffer zone in the next phase of the target phase (ie February 2016), and figure (d) is the buffer zone in January 2001 Schematic diagram, Figure (e) is a schematic diagram of the buffer zone in January 2002, Figure (f) is a schematic diagram of the buffer zone in January of the nth year, Figure (g) is the target phase and all phases related to the target phase Schematic overlay of the buffer.

Step 4: Non-fire data extraction;

In Figure 6, 46 pixels that did not fall into the fire buffer zone were randomly selected as the comparison data of the fire data to ensure the extraction quality of the non-fire data.

The embodiments of the present invention have been described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementations, and the above-mentioned specific implementations are only illustrative, rather than restrictive. Under the enlightenment of the invention, many forms can also be made without departing from the gist of the present invention and the scope of protection of the claims, and these all belong to the protection of the present invention.

Claims (5)

1. A kind of 1. method that non-fire data is extracted based on semivariable function, which is characterized in that include the following steps：

   Step 1：It determines the position of fire pixel in the data of combustion zone and the moment occurs；

   Step 2：Determine the vegetation pattern of fire pixel；

   Step 3：Extract the data of fire pixel and the non-fire pel data that surrounding is identical with fire pixel vegetation pattern；

   Step 4：Spatial Correlation Analysis is carried out to fire pixel using semivariable function, the buffering of Cheng Zuowei fire pixels will be become Distance simultaneously establishes corresponding fire buffering area；

   Step 5：Extract the data of fire pixel correlation phase；

   Step 6：Repeat the fire buffering area that step 4 establishes phase related to fire pixel；

   Step 7：It chooses the pixel for not falling within step 4 and step 6 gained fire buffering area and extracts its data as non-fire number According to.
2. A kind of 2. method that non-fire data is extracted based on semivariable function according to claim 1, which is characterized in that institute It is specially remote sensing image data to state combustion zone data in step 1.
3. 3. a kind of method that non-fire data is extracted based on semivariable function according to claim 1 or 2, feature are existed In it is to be fitted semivariable function value using spherical model to obtain to become journey in the step 4.
4. 4. a kind of method that non-fire data is extracted based on semivariable function according to claim 1 or 2, feature are existed In the foundation of fire buffering area is specifically using fire pixel as the center of circle in the step 4, and buffering is formed to become Cheng Zuowei radiuses Area.
5. 5. a kind of method that non-fire data is extracted based on semivariable function according to claim 1 or 2, feature are existed In the related phase of fire pixel is specifically in the step 6：Using single fire pixel as research object, pixel is occurred As target phase at the time of fire, corresponding phase is phase when the adjacent phase of target phase and remaining time are with target Close phase.