CN114492726A - An Inversion Algorithm of Forest Fuel Moisture Content Based on Remote Sensing Data - Google Patents

Abstract

The invention discloses an inversion algorithm for forest fuel moisture content based on remote sensing data, which belongs to the field of deep learning. The present invention adopts the MLP model to establish a connection algorithm between the remote sensing spectral reflectance and the forest canopy vegetation and the moisture content of the surface litter. By preprocessing the remote sensing data, the canopy vegetation is inverted using its spectral reflectance. and the moisture content of surface litter, and the fitting degree of the final model can reach about 0.8. The algorithm also proposes an optimization scheme for optical remote sensing with poor penetration between the canopy and the surface, which also provides a theoretical basis for the large-scale determination of regional canopy and surface litter moisture content by remote sensing estimation.

Description

An Inversion Algorithm of Forest Fuel Moisture Content Based on Remote Sensing Data

technical field

The invention belongs to the field of deep learning, and in particular relates to a forest fuel moisture content inversion algorithm based on remote sensing data.

Background technique

At present, the measurement methods of forest fuel moisture content mainly include equilibrium moisture content method, meteorological element method and remote sensing estimation method. The equilibrium moisture content method comprehensively considers the equilibrium moisture content, the initial moisture content of combustibles, time and time delay factors in an ideal environment, and then predicts the change in moisture content over a period of time through a model. The meteorological element regression method is mainly to establish statistical models of various meteorological factors and the moisture content of combustibles, mainly including fire risk scale model method, comprehensive index method, Rothermel model and BEHAVE model. Remote sensing estimation has been widely used with the development of remote sensing technology. With the rapid development of computers and the advancement of satellite technology, the application direction of remote sensing technology has been extended to the detection of soil and vegetation moisture in the 1970s. The hyperspectral technology that appeared in the 1990s can use optical sensors to obtain spectral data in various regions, and the spectral information is mainly derived from combustibles.

Compared with the other two methods, the remote sensing estimation method has the advantages of low cost and large measurement scale. At present, there are many studies using remote sensing spectroscopy to estimate the moisture content of live combustibles, but in practice, the moisture content of dead combustibles is lower than that of live combustibles, so it has a greater impact on fire occurrence. Therefore, the use of remote sensing technology to estimate the moisture content of canopy vegetation and surface litter is of great significance in fire forecasting. The traditional estimation of the moisture content of combustibles in the region is based on a large amount of artificial measured data. Although this method has high accuracy, it is very inefficient, consumes a lot of manpower and material resources, and causes certain damage to the regional ecology.

Therefore, a new method with convenient acquisition, high timeliness and long detection distance is needed to provide data for the inversion of regional fuel moisture content.

SUMMARY OF THE INVENTION

In view of the above problems existing in the prior art, the purpose of the present invention is to provide an inversion algorithm for forest fuel moisture content based on remote sensing data.

In order to solve the above problems, the technical scheme adopted in the present invention is as follows:

An inversion algorithm for forest fuel moisture content based on remote sensing data, including the following steps:

Step 1: Extract remote sensing data, preprocess the remote sensing data, and cut and filter the quadratic points;

Step 2: Go to the plot points marked in Step 1 for field sampling;

Step 3: Use the MLP deep learning model to invert the moisture content of the canopy combustibles;

Step 4: Use the MLP deep learning model to invert the water content of surface combustibles.

The described steps of preprocessing remote sensing data are:

Step 1: Perform atmospheric correction on the 10m resolution band and the 20m resolution band in turn, and obtain L2A level data of two sets of 10m resolution and two sets of 20m resolution bands respectively;

Step 2: Use the nearest neighbor method to resample the data obtained in step 1 into a 10m resolution band;

Step 3: Perform band synthesis on the data obtained in Step 2 to generate a true color image.

The field sampling in the step 2 is to collect the canopy vegetation and the litter on the surface at the sampling point, and measure and record the longitude and latitude, tree species, temperature, humidity and atmospheric pressure of the sample point, and calculate the water content of all samples by the following formula: Rate:

absolute moisture content

Relative moisture content

Wherein, _WH is the wet weight of combustibles (g), and W _D is the dry weight of combustibles (g).

The canopy water content inversion is to use the MLP deep learning model to select red light, green light, near-infrared and two short-wave infrared in the original data as the input, and directly carry out the inversion of the water content of the canopy combustibles, And perform machine learning with the actual sampling data many times to optimize the model.

The inversion of the water content of the surface combustibles is to process the original data using the bidirectional reflection distribution function to obtain multi-angle remote sensing data, which is publicized as follows:

与

分别指入射方向和观测方向在方位上的角度；where λ is the wavelength, θ _i is the angle between the incident direction of sunlight and the zenith angle, θ _r is the angle between the observation direction and the zenith angle,

and

Refers to the angle of the incident direction and the observation direction in the azimuth, respectively;

Substitute the obtained data into the 4-scale model in the radiative transfer model again, and the reflectance relationship of the 4-scale model is:

R=R _T K _T +R _G K _G +R _ZT K _ZT +R _ZG K _ZG

Among them: R _T represents the reflectivity of the canopy light surface;

K _T represents the probability that the sensor observes the light surface on the ground

R _G represents the reflectivity of the surface light surface;

K _G represents the probability that the sensor observes the light surface on the ground

R _ZT represents the reflectivity of the canopy background surface;

K _ZT represents the probability that the sensor observes the background surface of the canopy

R _ZG represents the reflectivity of the ground background surface;

K _ZG represents the probability that the sensor observes the ground background surface;

Then use the following formula to obtain surface remote sensing data;

In the formula: M is the multiple scattering factor, A, B, C are the relationship between M and K components;

Finally, in the MLP deep learning model, the red light, green light, near-infrared and two short-wave infrared in the obtained surface remote sensing data are selected as the input to invert the water content of the surface combustibles, and multiple times with the actual sampling to obtain The data is used for machine learning and the model is optimized.

Compared with the prior art, the present invention is based on the MLP deep learning model, uses remote sensing data to invert the water content, only uses a small amount of measured data as a test for the accuracy of the model, and uses remote sensing data to carry out a wide range of remote sensing data. Distance detection, convenient acquisition and high timeliness, the present invention also provides a new method for inversion of regional fuel moisture content

Description of drawings

1 is a schematic diagram of the overall flow of the algorithm of the present invention;

Figure 2 is a true color remote sensing image;

Figure 3 shows the distribution of the plot points in the sampling area;

Figure 4 is a structural diagram of the MLP deep learning model;

Figure 5 is a comparison diagram of the training and actual errors of the canopy combustibles moisture content model;

Figure 6 shows the comparison between the predicted value of the canopy MLP model and the actual value:

Figure 7 is a comparison diagram of the training and actual errors of the water content model of the surface combustibles;

Figure 8 is a comparison diagram between the predicted value of the surface MLP model and the actual value;

Figure 9 is an inversion effect diagram of the water content of the canopy combustibles;

Fig. 10 is an inversion effect diagram of the water content of surface combustibles.

Detailed ways

The present invention will be further described below with reference to specific embodiments.

Example 1

The inversion structure of this algorithm is shown in Figure 1.

First, the remote sensing data of the target area is obtained by the Sentinel-2 satellite. The target area of the selected area in this example is Chongli District, Zhangjiakou City, Hebei Province. The remote sensing data of the study area were obtained by the satellite Sentinel-2B at 3:5 on August 25, 2020. Generated on orbit R075 in minutes and 49 seconds, as shown in Figure 1, the data level is L1C data that has completed geometric correction, radiometric calibration and atmospheric apparent reflectance calculations. Data preprocessing operations include atmospheric correction, resampling and cropping. After the Sen2Cor plug-in of SNAP (Sentinel Application Platform) performs atmospheric correction on the 10m resolution band and the 20m resolution band in turn, four sets of L2A level data (two sets for 10m and 20m resolution) can be obtained, and then the data is resampled The resolution is 10m, the resampling method is the nearest neighbor method, and then a true color image (Figure 1) is generated by band synthesis (R:G:B=Band4:Band3:Band2), and then cropped, and finally obtained within the study area. Remote sensing images, and randomly select a number of quadratic points (Region Of Interest, ROI) in the vegetation coverage area for field sampling.

It is necessary to obtain and study the moisture content of the vegetation canopy combustibles and surface litter. In the cropped remote sensing images of the study area, 200 plot points are uniformly and widely distributed and marked on the vector boundary map of Chongli District, such as As shown in Figure 2, the size of each quadratic point is set to 45m × 45m. After locating the plot points on the spot, go to the target area from July 1, 2021 to September 1, 2021 to collect the canopy vegetation and the litter on the surface of the plot points by the direct acquisition method, and measure and record The latitude and longitude, tree species, temperature and humidity and atmospheric pressure of the sample points are obtained.

After taking back all the samples, first weigh the samples and record as the wet weight of combustibles, and then put them into the oven for constant temperature drying. After the weight is constant, measure the dried weight and record as the wet weight of combustibles. , and finally calculate the moisture content of all samples according to formula 1.

Absolute Moisture Content (AMC)

Relative Moisture Content (RMC)

In the formula, W_H is the wet weight of combustibles (g), and W_D is the dry weight of combustibles (g).

The near-infrared band is located in the high reflection area of plants and also in the strong absorption area of water, and the short-wave infrared band is located between the absorption bands of water, and the moisture content of combustibles has a significant correlation with the spectral reflectance of these two bands. The spectral moisture index method mainly calculates the spectral index based on the spectral reflectance and compares it with the measured data to calculate the moisture content of the canopy combustibles. For the moisture content of the surface litter, the remote sensing data needs to be processed according to the radiative transfer model. Firstly, the remote sensing of the ground is initially obtained through satellite remote sensing data, and the moisture content of combustibles is retrieved through spectral reflectance and spectral moisture index.

At the same time, due to the canopy occlusion problem, the canopy occlusion problem is solved by the radiative transfer model, and the correlation between the moisture content of the combustibles and the spectral reflectance is analyzed based on the MLP model.

After the remote sensing data and the data collected in the field are collected, the data is input to the computer for deep learning of the inversion model. The present invention adopts the MLP deep learning model.

The network structure of MLP includes input layer, hidden layer and output layer. It is a commonly used model in deep learning. It learns the characteristics of input data by building a multi-layer neural network. It has strong adaptability and is currently widely used. applied to regression prediction research.

Based on the above model, the acquired L1C-level data is preprocessed by atmospheric correction, resampling, etc., and then processed into L2A-level remote sensing data, and then the target study area is trimmed as the quadratic points for model inversion. Then on the basis of all the bands provided by the remote sensing data, the correlation analysis was carried out, and a total of 5 features in the red (B3), green (B4), near-infrared (B8) and two short-wave infrared (B11, B12) bands were selected. Variables are used as multi-independent variable inputs to the MLP model. According to the spectral characteristics of water content, the most suitable MLP deep learning model is selected. The model structure is shown in Figure 4. Each of the two fully connected layers contains 64 nodes, and uses ReLU (Rectified Linear Unit) as the activation function, and the output layer uses a linear function as the activation function.

Based on the MLP deep learning model, to invert the moisture content of the canopy combustibles, select B3, B4, B8, B11, B12 as the input reflectivity of 5 bands, and select 70% of the data in the sample (140 in total) as training Samples, using Mean Squared Error (MSE) as the loss function, iteratively trained 1000 times. The comparison between the training error and the actual error during the training process is shown in Figure 5. Both can control the error within 1, and the model training effect is better. After the training is completed, select 30% of the data in the sample (60 in total) as the test sample, use the model to predict and compare with the actual value, draw a line graph as shown in Figure 6, and calculate the actual fit (R^2), calculate the result is 0.843.

与观测目标表面的辐射强度

之间的比值，其函数公式如下：In addition to the combustibles in the canopy, it is also necessary to invert the moisture content of the surface litter. direction) reflected irradiance

and the radiation intensity of the observation target surface

The ratio between and its function formula is as follows:

与

分别指入射方向和观测方向在方位上的角度。where λ is the wavelength (nm), θ _i is the angle between the incident direction of sunlight and the zenith angle, θ _r is the angle between the observation direction and the zenith angle,

and

Refers to the angle of the incident direction and the observation direction in the azimuth, respectively.

The reflectance relation of the 4-scale model of the radiative transfer model is:

R=R _T K _T +R _G K _G +R _ZT K _ZT +R _ZG K _ZG (4)

Among them: R _T represents the reflectivity of the canopy light surface;

K _T represents the probability that the sensor observes the light surface on the ground

R _G represents the reflectivity of the surface light surface;

K _G represents the probability that the sensor observes the light surface on the ground

R _ZT represents the reflectivity of the canopy background surface;

K _ZT represents the probability that the sensor observes the background surface of the canopy

R _ZG represents the reflectivity of the ground background surface;

K _ZG represents the probability that the sensor observes the ground background surface.

The relationship between the canopy spectral reflectance R _α and R _β at observation angles α and β is as follows:

In the formula: M is the multiple scattering factor, A, B, C are the relationship between M and K components.

The use of remote sensing data to invert the moisture content of surface combustibles has the problem of canopy occlusion. The spectral reflectance of the vegetation area is jointly determined by factors such as leaves and soil. It is not a plane rigid body, and the radiation can pass through the vegetation canopy and then reappear. After multiple scattering, it finally escapes from the upper layer of vegetation and is received by remote sensing. The remote sensing data obtains a two-dimensional plane model, while the vegetation area is a three-dimensional model. Therefore, in order to obtain the surface reflectance, this paper first uses BRDF to process the original remote sensing data through ENVI to obtain multi-angle remote sensing data, and then substitutes it into the 4-scale model in the radiative transfer model. Based on the measured data and formula 4, the multiple scattering is obtained. factor M and observation probability K, and then obtain surface remote sensing data according to formula 5.

Based on remote sensing data and MLP deep learning model for the inversion of surface litter moisture content, five bands of reflectance, including B3, B4, B8, B11, and B12, were selected as input, and 70% of the data in the sample (140 in total) were selected. ) as the training sample, using the mean squared error as the loss function, and iteratively trained for 1000 times. The comparison chart between the training error and the actual error during the training process is shown in Figure 7. During the training process, the training error is continuously reduced, and the actual error is stabilized and no iteration is performed to prevent overfitting. The actual mean absolute error ended up being 7.69. After the training is completed, 30% of the data in the samples (60 in total) are selected as test samples, and the model is used to predict and compare with the actual value, draw a line graph in Figure 8 and calculate the actual fit (R ² ), the calculation result is 0.448 .

The time of the remote sensing data image selected in this example is August 25, 2020, which is consistent with the time period of the field investigation from July 1, 2021 to September 1, 2021, and the vegetation condition has no obvious change.

Figure 9 is the grayscale map and distribution map of water content in Chongli District based on the input variable of spectral reflectance inversion according to the MLP model. The main tree species in the west of Chongli District are shrubs, and the main tree species in the east are trees, shrubs and trees. The water absorption process of canopy leaves in Chongli is similar, and the interception of some shrubs is higher than that of trees, so the water content of canopy vegetation in Chongli District will be higher in the west and lower in the east.

Figure 10 is the grayscale map and distribution map of water content in Chongli District based on the inversion of the input variable of spectral reflectance according to the MLP model. The main tree species in the southeastern part of Chongli District are arbor species. The litter has a more obvious effect of intercepting surface runoff and inhibiting the evaporation of soil water. At this time, the moisture content is higher, so the moisture content of litter in this area is higher than that in the west.

The average moisture content of the canopy vegetation in the entire Chongli area is 35%, and the average moisture content of the surface litter is 52%. The inversion accuracy of the moisture content of the canopy combustibles is relatively high, and the fitting degree is 0.843. Although the surface litter has the problem of canopy occlusion, the inversion of the remote sensing data is also obtained by the radiative transfer model. model with a goodness of fit of 0.448.

Claims (5)

1. A forest combustible water content inversion algorithm based on remote sensing data is characterized by comprising the following steps:

step 1: extracting remote sensing data, preprocessing the remote sensing data, and cutting and screening sample points;

step 2: proceeding to the sample points marked in the step 1 for sampling on the spot;

and step 3: performing inversion of the water content of the combustible materials in the canopy by adopting an MLP deep learning model;

and4, step 4: and performing surface combustible water content inversion by adopting an MLP deep learning model.

2. The forest combustible water content inversion algorithm based on remote sensing data as claimed in claim 1, wherein the step of preprocessing the remote sensing data is as follows:

step 1: sequentially carrying out atmospheric correction on the 10m resolution wave band and the 20m resolution wave band to respectively obtain two groups of L2A-level data of 10m resolution and two groups of 20m resolution wave bands;

step 2: resampling the data obtained in the step 1 to be a 10m resolution wave band by using a nearest neighbor method;

and step 3: and (3) performing band synthesis on the data obtained in the step (2) to generate a true color image.

3. The forest combustible water content inversion algorithm based on remote sensing data as claimed in claim 1, wherein the field sampling in step 2 is to collect canopy vegetation and litter on the earth surface at a sample point, measure and record information such as longitude and latitude, tree species, temperature and humidity and atmospheric pressure of the sample point, and calculate water content of all samples according to the following formula:

absolute water content

Relative water content

Wherein, W_HIs the wet weight (g) of combustible materials, W_DIs combustible dry weight (g).

4. The forest combustible water content inversion algorithm based on remote sensing data as claimed in claim 1, wherein the canopy water content inversion is that red light, green light, near infrared and two short wave infrared in original data are selected as input ends by using an MLP deep learning model, inversion of canopy combustible water content is directly carried out, machine learning is carried out for multiple times with data obtained by actual sampling, and the model is optimized.

5. The forest combustible water content inversion algorithm based on remote sensing data of claim 1, wherein the inversion of the surface combustible water content is that the original data is processed by using a two-way reflection distribution function to obtain multi-angle remote sensing data, and the expression is as follows:

where λ is the wavelength, θ_iIs the angle between the incident direction of the sunlight and the zenith angle, theta_rIs the angle between the observation direction and the zenith angle,

and

the angles of the incident direction and the observation direction in the direction are respectively indicated;

and substituting the obtained data into a 4-scale model in the radiation transmission model, wherein the reflectivity relation of the 4-scale model is as follows:

R＝R_TK_T+R_GK_G+R_ZTK_ZT+R_ZGK_ZG

wherein: r_TRepresenting the reflectivity of the canopy illuminated surface;

K_Trepresenting the probability of the sensor observing the ground illuminated surface

R_GRepresenting the surface illumination surface reflectivity;

K_Grepresenting the probability of the sensor observing the ground illuminated surface

R_ZTRepresenting the reflectivity of the background surface of the canopy;

K_ZTrepresenting the probability of the sensor observing the background surface of the canopy

R_ZGRepresenting the reflectivity of the ground background surface;

K_ZGrepresenting the probability of the sensor observing the ground background surface;

then obtaining surface remote sensing data by using the following formula;

in the formula: m is a multiple scattering factor, and A, B and C are relational expressions of M and K components;

and finally, selecting red light, green light, near infrared and two short wave infrared in the obtained surface remote sensing data as input ends in an MLP deep learning model, carrying out inversion on the water content of the surface combustible, and carrying out machine learning with the data obtained by actual sampling for multiple times to optimize the model.

Images