Saturation quantification method and device in forest stock remote sensing estimation and medium

Abstract

The invention discloses a saturation quantification method and device in forest stock remote sensing estimation and a medium, and the method comprises the steps: determining a fitting curve based on an FSV estimation value and an observation value, and determining a saturation starting point and a saturation point based on the fitting curve; and evaluating the saturation level in the FSV estimation according to the saturation starting point and the saturation point. According to the method, two indexes of a saturation starting point (FSVSP) and an estimated saturation point (FSVESAT) are used for evaluating the saturation level in FSV estimation of the system. And according to a scatter diagram and an error of an estimated value and a measured value of the FSV, defining the corresponding FSV as FSVSP when systematic underestimation begins to occur in the estimated value of the FSV, and defining the corresponding FSV as FSVESAT when an error caused by saturation is greater than an estimated random error. When the FSV reaches the FSVSP, a systematic underestimation phenomenon begins to occur in the estimated value; and when the FSV reaches the FSVESAT, the average value of FSV estimation is lower than the minimum value of the FSV in a random error range, so that quantitative expression of the saturation level from a single remote sensing variable to multiple remote sensing variables in a complex forest scene is realized.

Description

Technical Field

[0001] The present invention relates to the field of forestry remote sensing technology, and more specifically, to a saturation quantification method, device and medium in remote sensing estimation of forest volume. Background Art

[0002] Among the many available remote sensing data, optical remote sensing data has the longest research time, the most mature technology accumulation, the richest data source, and the most stable data performance. It has been widely used in forest resource monitoring. However, when the forest canopy is closed, the spectral reflectance of the canopy remains unchanged, and the increase in height causes the FSV to continue to increase. At this time, the spectral reflectance no longer changes with the change of FSV, that is, spectral saturation occurs. Optical remote sensing data will inevitably experience spectral saturation in forest stands with high stock volumes, which seriously affects the accuracy and reliability of FSV remote sensing mapping. Spectral saturation is an inherent attribute of optical remote sensing data and cannot be avoided. The level of saturation determines the upper limit of FSV estimated by optical remote sensing data. Therefore, a comprehensive evaluation and quantification of the saturation level of FSV estimated by optical remote sensing data is the key to achieving accurate and reliable FSV remote sensing mapping.

[0003] The characteristic variables extracted from optical remote sensing data mainly include two categories: band reflectance and vegetation index. Optical remote sensing can obtain information such as vegetation health, structure and quality by recording different wavelength spectral information reflected or radiated by the surface. The wavelength range captured by optical sensors usually includes visible light, near infrared and infrared bands, which cover the absorption peaks of plant chlorophyll and information related to vegetation structure and water content. The reflection of near infrared and red edge bands can provide information about the structure and coverage of vegetation leaves. The reflection of plants in these bands is affected by leaf structure and surface roughness, so these data can be used to estimate the biophysical parameters of vegetation. In tropical forests, due to the more vigorous growth of vegetation canopy, the spectral saturation effect is more serious than that in temperate coniferous forests, and it is more obvious in the red light band than in the green light band and near infrared band. Compared with visible light, the near infrared region shows higher variability in dense forests, and its saturation level is also higher than that of visible light bands.

[0004] Vegetation index is a type of numerical index extracted from remote sensing data to describe the status and characteristics of vegetation. These indices provide information about the physiological and structural characteristics of vegetation by analyzing the reflectance or radiation data of different bands. Vegetation index enhances the sensitivity of vegetation attribute observation by performing algebraic operations on spectral reflectance. Studies have shown that there is a high correlation between vegetation index and FSV. By analyzing the changes in vegetation index, the growth status of forest vegetation can be inferred and FSV can be estimated. Different vegetation indices have different sensitivities to forest attributes. For example, the Normalized Difference Vegetation Index (NDVI) has a high sensitivity to vegetation coverage and monitoring vegetation health; the Enhanced Vegetation Index (EVI) makes up for the inaccurate observation of NDVI in high vegetation coverage areas and improves the sensitivity to vegetation ecological processes. Although vegetation index can more intuitively characterize the growth characteristics and status of vegetation, the same as single-band spectral reflectance is that vegetation index still faces the problem of saturation in high vegetation areas. The saturation level represents the upper limit of FSV estimation using vegetation indices, so analyzing the saturation level of vegetation indices is crucial to assessing the reliability of using vegetation indices to estimate FSV. Due to spectral saturation, the continued growth of vegetation until maturity and subsequent senescence is difficult to observe through vegetation indices, so when using optical data to estimate FSV, it is difficult to make a stable estimate of forests that exceed the spectral saturation level.

[0005] In addition, due to the complexity of forest ecosystems, the binomial reflectance distribution function (BRDF) of forest vegetation spectra makes it more complicated to obtain forest structural information from optical remote sensing data. Studies have shown that the impact of BRDF on biomass estimation is one-third higher than that on NDVI values. BRDF is related to the way light is reflected, which depends on the angle of the incident light relative to the earth's surface and the angle of the sensor relative to the earth's surface. Specifically, cloud cover, terrain, solar zenith angle, and the geometric position of the sensor greatly affect the spectral reflectance of forest vegetation, so the spectral saturation of forest vegetation is also seriously affected by these environmental factors. Summary of the invention

[0006] The present invention is provided to solve the above problems existing in the prior art. Therefore, a saturation quantification method, device and medium in remote sensing estimation of forest volume are needed to achieve quantitative expression of saturation levels from single remote sensing variables to multiple remote sensing variables in complex forest scenes by constructing a saturation quantification model for estimating FSV based on optical remote sensing data.

[0007] According to a first aspect of the present invention, a method for quantifying saturation in remote sensing estimation of forest volume is provided, the method comprising:

[0008] Determine a fitting curve based on the estimated value and the observed value of FSV, the fitting curve comprising a first curve, a second curve, a third curve, a fourth curve, a fifth curve and a sixth curve, the first curve and the third curve represent the range of random errors, the second curve is a 1:1 straight line, representing the spatial distribution of FSV and the relationship between the predicted value and the observed value when there is no error in the predicted value, the fourth curve and the sixth curve represent the range of saturation errors, and the fifth curve is a scatter plot fitting curve of the observed value and the predicted value;

[0009] Determine a saturation starting point and a saturation point based on the fitting curve;

[0010] The saturation level in the FSV estimation is evaluated based on the saturation starting point and the saturation point.

[0011] Furthermore, based on the fitting curve, the saturation starting point is determined by the following method:

[0012] Determine an observation value S1 among all observation values, and take the observation value S1 as a saturation starting point;

[0013] For the observed value S1, when the FSV is less than the observed value S1, the estimation error consists of random errors, and when the FSV is greater than the observed value S1, the predicted value of the FSV is saturated.

[0014] Further, based on the fitting curve, the saturation point is determined by the following method:

[0015] Determine an observation value S among all observation values, and take the observation value S as the saturation starting point;

[0016] The corresponding points of the observation value S on the fourth curve, the fifth curve and the sixth curve are A, B and C, wherein the length of the line segment AB is equal to the length of the line segment BC.

[0017] Furthermore, the first curve, the second curve, the third curve, the fourth curve, the fifth curve and the sixth curve are respectively expressed by the following formulas (1) to (6):

[0018] Y c =x+e (1)

[0019] Y b =x (2)

[0020] Y c =xe (3)

[0021] Y d=kx+b+e (4)

[0022] Y e =kx+b (5)

[0023] Y f =kx+be (6)

[0024] In the formula, Y c , Y b , Y c , Y d , Y e and Y f are the predicted values ​​of the corresponding observed values ​​x in the first, second, third, fourth, fifth and sixth curves respectively, x is the observed value of FSV, k and b are constants, e is the random error of the predicted value, that is, the average distance from the predicted value of all samples to the fitting line e, expressed as Where n is the total number of samples, y i is the predicted value of sample i, y ei is the corresponding point on the fifth curve.

[0025] Furthermore, the saturation starting point is calculated by the following formula:

[0026] FSV SP =b / (1-k) (7)

[0027] Where, FSV SP Indicates the saturation starting point.

[0028] Furthermore, the saturation point is calculated by the following formula:

[0029] FSV ESAT =(b+e) / (1-k) (8)

[0030] Where, FSV ESAT Indicates the saturation starting point.

[0031] Further, evaluating the saturation level in the FSV estimation according to the saturation starting point and the saturation point specifically includes:

[0032] When FSV reaches FSV SP When , the estimated value of FSV begins to be systematically underestimated;

[0033] When FSV reaches FSV ESAT When , the mean value of the FSV estimate is lower than the minimum value of the FSV within the random error range.

[0034] According to a second aspect of the present invention, there is provided a saturation quantification device for remote sensing estimation of forest volume, the device comprising:

[0035] A curve fitting module is configured to determine a fitting curve based on an estimated value and an observed value of the FSV, wherein the fitting curve includes a first curve, a second curve, a third curve, a fourth curve, a fifth curve and a sixth curve, wherein the first curve and the third curve represent a range of random errors, the second curve is a 1:1 straight line representing a spatial distribution of the FSV and a relationship between the predicted value and the observed value when there is no error in the predicted value, the fourth curve and the sixth curve represent a range of saturation errors, and the fifth curve is a scatter plot fitting curve of the observed value and the predicted value;

[0036] An indicator determination module, configured to determine a saturation starting point and a saturation point based on the fitting curve;

[0037] The saturation quantization module is configured to evaluate a saturation level in FSV estimation according to the saturation starting point and the saturation point.

[0038] Furthermore, the saturation quantization module is further configured as follows:

[0039] When FSV reaches FSV SP When , the estimated value of FSV begins to be systematically underestimated;

[0040] When FSV reaches FSV ESAT When , the mean value of the FSV estimate is lower than the minimum value of the FSV within the random error range.

[0041] According to a third aspect of the present invention, there is provided a readable storage medium storing one or more programs, wherein the one or more programs can be executed by one or more processors to implement the method as described above.

[0042] The present invention has at least the following beneficial effects:

[0043] The present invention proposes a saturation threshold (FSV SP ) and estimate the saturation point (FSV ESAT ) are used to systematically evaluate the saturation level in FSV estimation. Based on the scatter plot and error of the estimated and measured values ​​of FSV, the FSV corresponding to the point where the estimated value of FSV begins to be systematically underestimated is defined as FSV SP , the FSV corresponding to the error caused by saturation being greater than the estimated random error is defined as FSV ESAT When FSV reaches FSV SP When FSV reaches FSV, its estimated value begins to be systematically underestimated; ESAT When the average value of FSV estimation is lower than the minimum value of FSV within the random error range, the quantitative expression of the saturation level from single remote sensing variable to multiple remote sensing variables in complex forest scenes can be achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 A flow chart of a method for quantifying saturation in remote sensing estimation of forest volume according to an embodiment of the present invention is shown.

[0045] Figure 2 A scatter plot of the predicted value and the observed value of FSV according to an embodiment of the present invention is shown, wherein (a1) and (a2) respectively represent the spatial distribution of FSV and the relationship between the predicted value and the observed value when there is no error in the predicted value, (b1) and (b2) respectively represent the spatial distribution of FSV and the relationship between the predicted value and the observed value when there is a random error in the predicted value, and (c1) and (c2) respectively represent the spatial distribution of FSV and the relationship between the predicted value and the observed value when there is a systematic error in the predicted value.

[0046] Figure 3 FIG. 1 is a schematic diagram of a fitting curve according to an embodiment of the present invention, wherein (a) is a curve obtained by fitting a scatter plot of predicted values ​​and observed values, and (b) is a curve obtained by fitting a scatter plot of predicted values ​​and observed values. Figure 3 The curve shown in (a) is a simplified diagram.

[0047] Figure 4 The LiDAR_FSV product of Wangyedian Forest Farm and the spatial distribution map of Pinus tabulaeformis and Larix gmelinii samples according to an embodiment of the present invention are shown.

[0048] Figure 5 The FSV product of Gaofeng Forest Farm and the spatial distribution map of eucalyptus samples according to an embodiment of the present invention are shown.

[0049] Figure 6 A precision verification diagram of LiDAR_FSV according to an embodiment of the present invention is shown.

[0050] Figure 7 The FSV of the eucalyptus FSV prediction value according to an embodiment of the present invention is shown. SP With FSV ESAT Schematic diagram of .

[0051] Figure 8 The FSV of larch FSV prediction value according to an embodiment of the present invention is shown. SP With FSV ESAT Schematic diagram of .

[0052] Fig. 9 The FSV of the predicted value of the FSV of Pinus tabulaeformis according to an embodiment of the present invention is shown. SP With FSV ESAT Schematic diagram of .

[0053] Fig.10A scatter plot of FSV observation values ​​and predicted values ​​according to an embodiment of the present invention is shown, wherein (a) is eucalyptus, (b) is larch, and (c) is Pinus tabulaeformis.

[0054] Fig.11 A structural diagram of a saturation quantification device in remote sensing estimation of forest volume according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0055] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. The embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings and specific embodiments, but are not intended to limit the present invention. For the various steps described herein, if there is no necessity for a causal relationship between each other, the order in which they are described as examples herein should not be regarded as a limitation, and those skilled in the art should know that they can be adjusted in order, as long as the logic between them is not destroyed, resulting in the inability to implement the entire process.

[0056] The embodiment of the present invention provides a saturation quantification method in remote sensing estimation of forest volume. Figure 1 As shown, it is a flow chart of the method, which includes the following steps S100 to S300, which are described in detail as follows:

[0057] Step S100, determining a fitting curve based on the estimated value and the observed value of FSV.

[0058] The fitting curves described in this embodiment include a first curve, a second curve, a third curve, a fourth curve, a fifth curve and a sixth curve. The first curve and the third curve represent the range of random errors. The second curve is a 1:1 straight line, which represents the spatial distribution of FSV and the relationship between the predicted values ​​and the observed values ​​when there is no error in the predicted values. The fourth curve and the sixth curve represent the range of saturation errors. The fifth curve is a scatter plot fitting curve of the observed values ​​and the predicted values.

[0059] Exemplarily, this embodiment proposes an estimation method for the estimated value of FSV. Specifically, a single variable saturation quantification model, namely a three-stage saturation quantification model, is constructed based on the gradual relationship curve between vegetation index and FSV. The saturation inflection point (FSV) is proposed in the model. INF ) and saturation point (FSV SAT ) two indicators to evaluate the saturation process of vegetation index to FSV. This paper uses a two-stage linear regression model to determine the turning point of the functional relationship between vegetation index and FSV, and defines it as FSV INF , when FSV is greater than FSV INFWhen the vegetation index is saturated, the vegetation index begins to saturate; the sensitivity of the vegetation index to FSV when it is not saturated is Sensitivity, and the FSV corresponding to the sensitivity of the vegetation index to FSV equals 0.1 times the Sensitivity is defined as FSV SAT , indicating that the vegetation index and FSV are completely saturated. Compared with the use of semivariogram function to quantify the saturation of vegetation index, the three-stage saturation quantification model can more accurately describe the gradual relationship between vegetation index and FSV.

[0060] When the estimated value of FSV has been determined, the observed value of FSV can be directly observed through existing observation devices. Therefore, the fitting curve can be fitted using a scatter plot of the predicted value and the observed value of FSV, where the fitting curve includes multiple curves. It should be noted that a straight line is a type of curve, and one or more of the first curve, second curve, third curve, fourth curve, fifth curve and sixth curve described in this article can be in the form of a straight line.

[0061] Step S200: determining a saturation starting point and a saturation point based on the fitting curve.

[0062] In this example, the saturation threshold (FSV) is proposed for the first time. SP ) and saturation point (FSV ESAT ) are used to systematically evaluate the saturation level in FSV estimation. Based on the scatter plot and error of the estimated and measured values ​​of FSV, the FSV corresponding to the point where the estimated value of FSV begins to be systematically underestimated is defined as FSV SP , the FSV corresponding to the error caused by saturation being greater than the estimated random error is defined as FSV ESAT .

[0063] Step S300 , evaluating the saturation level in FSV estimation according to the saturation starting point and the saturation point.

[0064] In some embodiments, the specific method of evaluating the saturation level in the FSV estimation according to the saturation starting point and the saturation point is: when the FSV reaches FSV SP When FSV reaches FSV, its estimated value begins to be systematically underestimated; ESAT When , the mean value of the FSV estimate is lower than the minimum value of the FSV within the random error range.

[0065] In some embodiments, the saturation quantification method in the remote sensing estimation of forest volume realizes the quantitative expression of the saturation level from a single remote sensing variable to multiple remote sensing variables in a complex forest scene by constructing a saturation quantification model for estimating FSV based on optical remote sensing data. The saturation quantification model is a multivariate saturation quantification model, and the core components of the model include the fitting curves described above. For ease of description, the first curve, the second curve, the third curve, the fourth curve, the fifth curve, and the sixth curve included in the fitting curve are represented by the characters a, b, c, d, e, and f, respectively.

[0066] Figure 2 The scatter plot of the predicted value and the observed value of FSV is shown in Figure 1. Generally speaking, when the accuracy of the predicted data is 100%, the scatter plot of the predicted value and the measured value is as follows: Figure 2 As shown in (a1) and (a2), all points are distributed on the 1:1 straight line. In actual FSV estimation, there are usually estimation errors caused by various factors. According to the factors causing the errors, the errors of FSV estimation can be divided into two categories. One is random error, which comes from the uncertainty of FSV estimation caused by multiple factors such as sensors, environmental factors, and ground measurements. This type of error is highly random and difficult to predict. When the estimation error only contains random errors and is not affected by saturation, the scatter plot of the estimated value and the observed value will be evenly distributed on both sides of the x=y straight line, as shown in Figure 2 The other type of error is due to data saturation, which comes from the spectral saturation phenomenon and leads to underestimation of samples with higher FSV. This type of error is similar to systematic error, which stably underestimates samples with high FSV, and when FSV exceeds a certain value, the predicted value will no longer change. When the estimated error is composed of random error and saturation error, the scatter plot of the estimated value and the observed value is as follows: Figure 2 As shown in (c1) and (c2), when FSV is between 0 and S, the estimation error only contains random errors, and the scatter plot of the estimated value and the observed value is approximately distributed on both sides of the straight line 1:1; when FSV is greater than S, the predicted FSV is always smaller than the observed FSV due to spectral saturation. Therefore, from the overall point of view, the estimation error can be regarded as composed of random errors and saturation errors.

[0067] By analyzing the sources and influencing factors of estimation errors, this embodiment constructs a multivariate saturation quantification method based on error analysis. Figure 3 As shown, by analyzing the relationship between the estimated value and the observed value of FSV, the scatter plot can be approximately fitted as Figure 3The fitting curve shown in (a). Wherein straight line b is a 1:1 straight line; straight lines a and c represent the range of random errors; straight line e is the scatter plot fitting curve of observed values ​​and predicted values; d and f represent the range of random errors; points A and C represent the range of predicted FSV (A to C) when the observed FSV is equal to S; point B is the corresponding point on the dotted line c when the observed FSV is S, indicating the theoretical minimum value (B) for predicting samples with FSV equal to S when only random errors are considered; points O1, O2, O3 and O4 are the four intersection points of straight lines a, c, d and f respectively, and the curve quadrilateral O1 O2 O3 O4 represents the distribution range of predicted values ​​of FSV when the observed values ​​of FSV are S1 to S2. If there is no data saturation problem, the scatter plot of predicted values ​​and observed values ​​should be evenly distributed within the range of dotted lines a and c; after the data saturation problem occurs, the scatter plot of predicted values ​​and observed values ​​in the actual situation should be evenly distributed between dotted lines d and f.

[0068] When the scatter plot of the predicted values ​​and the observed values ​​is distributed in the curved triangle O1 O2 O3, the predicted values ​​are all within the range of the dotted lines a and c, indicating that the overall prediction error is smaller than the theoretical random error, and the current prediction error can be explained by the random error. When the scatter plot of the predicted values ​​and the observed values ​​is distributed in the curved triangle O2 O3 O4, the overall prediction error is greater than the theoretical random error, and the overall prediction error cannot be explained by the random error alone. Therefore, it is believed that the part where the prediction error is greater than the random error is caused by saturation. That is, when the scatter points fall in the curved triangle O2 O3 O4, it means that the saturation error caused by data saturation begins to affect the prediction results.

[0069] Depend on Figure 3 It can be seen that when FSV is less than S1, straight line e and straight line b are approximately coincident, and the error at this time is only composed of random errors; when FSV is greater than S1, the predicted value begins to be systematically underestimated, that is, the yellow area in the figure. Therefore, S1 is defined as the saturation starting point (FSV SP ), that is, when FSV is greater than FSV SP , the predicted value of FSV begins to saturate. The line segment AC in the figure represents the range of variation of the predicted FSV when the observed FSV is equal to S, and the line segment BC represents that the predicted value is affected by saturation. As FSV increases from S1 to S2, the range of variation of the estimated value changes from O1 to O3 to O2 to O4, and the length of BC increases from 0 to the maximum (AC). The error caused by data saturation has an increasing impact on the prediction results. When AB is greater than BC, that is, the impact of random errors on the results is greater than the saturation error. When AB is less than BC, the impact of saturation errors on the results is greater than random errors. Therefore, when AB is equal to BC, the observed value S of the corresponding FSV is the saturation point (FSV) of the estimated FSV under the remote sensing data source, variable set and prediction model.ESAT ).

[0070] In order to solve FSV SP and FSV SAT , this embodiment will Figure 3 (a) can be further simplified to Figure 3 In (b), the predicted value and observed value of FSV are approximately fitted into a straight line. The meaning of all elements in the figure remains unchanged. Then the straight lines a to f can be expressed by a linear parameter model as follows:

[0071] Y c =x+e (1)

[0072] Y b =x (2)

[0073] Y c =xe (3)

[0074] Y d =kx+b+e (4)

[0075] Y e =kx+b (5)

[0076] Y f =kx+be (6)

[0077] In the formula, Y c , Y b , Y c , Y d , Y e and Y f are the predicted values ​​of the corresponding observed values ​​x in the first, second, third, fourth, fifth and sixth curves respectively, x is the observed value of FSV, k and b are constants, e is the random error of the predicted value, that is, the average distance from the predicted value of all samples to the fitting line e, expressed as Where n is the total number of samples, y i is the predicted value of sample i, y ei is the corresponding point on the fifth curve.

[0078] Saturation start point FSV SP It can be obtained by the intersection of straight line f and straight line c, that is:

[0079] FSV SP =b / (1-k) (7)

[0080] FSV ESAT It can be obtained by the intersection of straight line c and straight line e, that is:

[0081] FSV ESAT=(b+e) / (1-k) (8)

[0082] In order to further illustrate the feasibility and advancement of the present application, this embodiment will compare the multivariate saturation quantification model constructed by this embodiment with other models / algorithms for saturation quantification.

[0083] In this example, FSV products (LiDAR_FSV) obtained by using Lidar data were collected in Wangyedian Forest Farm and Gaofeng Forest Farm respectively to replace ground survey data. The spatial resolution of the product is 30m, and each pixel is regarded as a 30m×30m square plot, such as Figure 4 and 5 The second-class forest resource survey data was used to remove other tree species, leaving only samples of Chinese pine, larch and eucalyptus.

[0084] The final sample information used for modeling is shown in Table 1. A total of 26,462 sample plots were extracted in Wangyedian Forest Farm using the Lidar_FSV product, including 10,208 Chinese pine sample plots and 16,254 larch sample plots. The FSV range of the Chinese pine sample plot is 7.27m 3 / ha to 182.34m 3 / ha, with the average value and coefficient of variation being 182.34 m 3 / ha and 30.47%; the FSV of larch plots ranged from 5.13 to 419, with the mean and coefficient of variation being 173.87m 3 / ha and 37.32%. Compared with the Pinus tabulaeformis plot, the Larix gmelinii plot had a slightly smaller mean value, but a larger coefficient of variation. In the Gaofeng Forest Farm, 9803 Eucalyptus plots were obtained in the same way, and their FSV ranged from 3.38m 3 / ha to 366.62m 3 / ha, the average value and coefficient of variation were 117.81m 3 / ha and 47.27%. Since eucalyptus is a fast-growing tree species with a rotation period of 5-10 years, the FSV obtained is significantly lower than that of the Pinus tabulaeformis and Larix gmelinii samples in Wangyedian Forest Farm.

[0085] Table 1 Sample information statistics

[0086] species Sample size <![CDATA[Maximum value (m 3 / ha)]]> <![CDATA[Minimum value (m 3 / ha)]]> <![CDATA[Mean value (m 3 / ha)]]> Coefficient of variation (%) Pinus tabulaeformis 10208 407.58 7.27 182.34 30.47 Larch 16254 419 5.13 173.87 37.32 eucalyptus 9803 366.62 3.38 117.81 47.27

[0087] In order to verify the accuracy of Lidar_FSV, this example surveyed 73 Chinese pine and larch plots and 60 eucalyptus plots in the LiDAR_FSV coverage area for comparison with LiDAR_FSV. And calculated the correlation coefficient (Pearson correlation coefficient, r) and mean relative error (Mean Relative Error, MRE) two indicators to evaluate the accuracy of Lidar_FSV. The calculation formula is as follows:

[0088]

[0089]

[0090] Where n is the sample size, y i is the measured value of the volume of the ith plot, is the average of the measured values ​​of the volume of all plots. is the estimated value of the volume of the i-th plot. The higher the accuracy, the smaller the MRE and the larger r. Figure 6 The scatter plot between LiDAR_FSV and measured FSV was drawn in Figure 1. Since the FSV distribution of Pinus tabulaeformis and Larix gmelinii samples is similar and in the same study area, Pinus tabulaeformis and Larix gmelinii are combined into one sample set when verifying LiDAR_FSV. The verification results of LiDAR_FSV show that there is a high correlation between LiDAR_FSV and measured FSV, and the relative errors are all less than 20%. Among them, the correlation coefficient r between the LiDAR_FSV of Pinus tabulaeformis and Larix gmelinii obtained in Wangyedian Forest Farm and the measured FSV reaches 0.91, and the relative deviation is 11.82%; the correlation coefficient r between the LiDAR_FSV of Eucalyptus obtained in Gaofeng Forest Farm and the measured FSV reaches 0.93, and the relative deviation is 17.85%. In summary, there is a high correlation between LiDAR_FSV and the actual measured FSV, and the error basically meets the requirements of the second-class survey of forest resources in my country (accuracy reaches 85%), so it is feasible to use LiDAR_FSV to expand the number of ground survey plots.

[0091] According to the multivariate saturation quantification method proposed in this embodiment, the saturation level of the FSV prediction value obtained by multi-scale optical remote sensing data, two variable selection methods of Pearson correlation coefficient and distance correlation coefficient, and four regression models of multiple linear regression (MLR), random forest (RF), support vector machine (SVM) and K-nearest neighbor (KNN) is quantified. The saturation starting point and FSV of each prediction result are ESAT Statistics Figure 7 (eucalyptus), Figure 8 (Larch) and Fig. 9 (Pinus tabulaeformis). Figures 7 to 9It can be seen that among the three tree species, there is no significant difference in the saturation starting points obtained by the variable sets of different spatial resolutions. The saturation starting point of eucalyptus is the lowest, at 117m. 3 / ha to 120m 3 / ha; the saturation starting point of larch is second, at 170m 3 / ha to 173m 3 / ha; the saturation starting point of Pinus tabulaeformis is the highest, at 179m 3 / ha to 184m 3 / ha. FSV ESAT As the spatial resolution decreases, it shows a slight upward trend and gradually stabilizes when the spatial resolution is lower than 20m and 25m. ESAT As the spatial resolution changes in the range of 172m 3 / ha to 261m 3 / ha; FSV of larch ESAT At 213m 3 / ha to 279m 3 / ha; FSV of Pinus tabulaeformis ESAT At 236m 3 / ha to 307m 3 / ha.

[0092] Among the three tree species, the FSV of Pinus tabulaeformis ESAT The highest, followed by larch and eucalyptus ESAT The lowest. Eucalyptus is a typical fast-growing tree species, and its harvest period is 5 to 10 years. Studies have shown that the crown closure speed of young eucalyptus forests is relatively fast during growth, and the crown diameter increases with the increase of FSV before the crown closure. Therefore, the spectral reflectance can capture the change of FSV. When the forest is closed, the spectral characteristics cannot reflect the increase of forest stand density. In contrast, the growth cycle of Pinus tabulaeformis and Larix gmelinii is longer, the crown closure is slower, and the spectral saturation occurs later. In addition, in the eucalyptus forest, the crown shape of eucalyptus is approximately an ellipsoid, and the spectral four-component changes in remote sensing images are not obvious, while the crown shape of Pinus tabulaeformis and Larix gmelinii can be approximately regarded as a cone, and the crown four-component fluctuations are larger, so its crown structure is easier to be captured by optical images, resulting in the spectral saturation level of Pinus tabulaeformis and Larix gmelinii being greater than that of eucalyptus.

[0093] FSV obtained by four regression models in the same tree species ESAT There are some differences, among which the FSV of MLR ESAT Minimum, FSV of the MLR model of eucalyptus ESAT At 197m 3 / ha to 205m 3 / ha, FSV of larchESAT At 233m 3 / ha to 238m 3 / ha; FSV of Pinus tabulaeformis ESAT At 267m 3 / ha to 280m 3 / ha. And as the spatial resolution of the optical image changes, the FSV obtained by the MLR model ESAT The smallest range of change is 8m 3 / ha(Eucalyptus), 5m 3 / ha (larch) and 13m 3 / ha (Pinus tabulaeformis). In contrast, the FSVs obtained by the three nonparametric models ESAT The differences were large with the change of spatial resolution, among which the SVM model in the eucalyptus sample had the largest difference at 60m. 3 / ha; the SVM model in the larch sample had the largest difference, which was 63m 3 / ha; the RF difference was the largest in Pinus tabulaeformis samples, which was 52m 3 / ha. In addition, the results also show that for high-resolution optical variables, the FSV of MLR ESAT Higher than the other three models. The reason is FSV ESAT The calculation of is related to the estimation error of the model and is proportional. For example, when the spatial resolution is 1m, the relative root mean square error of FSV estimated using the MLR model is close to 40%, and R 2 Less than 0.3, due to the influence of random errors, the FSV ESAT Therefore, when using the predicted value to evaluate the multivariate saturation level, it is necessary to analyze the prediction error and goodness of fit together.

[0094] In order to further verify the performance and stability of the multivariate saturation quantification model, this embodiment further used a small amount of ground survey data and Landsat 8 OLI remote sensing data with a spatial resolution of 30m to conduct FSV remote sensing estimation experiments on Pinus tabulaeformis, Larix olgensis and Eucalyptus in Wangyedian Forest Farm and Gaofeng Forest Farm, and used the multivariate saturation quantification model to quantify the saturation of the FSV estimation value of each tree species. First, the above 15 vegetation indices and single-band reflectivity were extracted from the Landsat 8 OLI data, and then the best set of modeling variables was selected based on the Pearson correlation coefficient combined with the forward feature selection method, and four regression models, MLR, SVM, RF and KNN, were constructed for estimating FSV. The FSV estimation results of each model are shown in Table 2.

[0095] Table 2 FSV estimation results based on ground survey data

[0096]

[0097] The four models showed different FSV estimation capabilities in the samples of the three tree species, and the estimation results of the three machine learning models were better than those of the MLR model. Among the three tree species, Eucalyptus had the lowest estimation accuracy, with an R 2 The range of variation is between 0.25 and 0.31, and the rRMSE is between 35.38% and 39.19%. 2 The range of variation was between 0.30 and 0.38, and the rRMSE was between 32.99% and 38.93%. The highest value was found in Pinus tabulaeformis, with an R 2 The range of variation is between 0.35 and 0.42, and the rRMSE is between 33.57% and 38.56%. For the eucalyptus sample, the RF model obtained the best result, with R 2 is 0.31, RMSE is 42.45m 3 / ha, and the RRMSE was 35.35%. For larch samples, the SVM model obtained the best result, and its R 2 is 0.38, RMSE is 70.26m 3 / ha, and the RRMSE was 32.99%. For Pinus tabulaeformis samples, the RF model obtained the best result. 2 is 0.42, RMSE is 85.26m 3 / ha, rRMSE is 33.57%.

[0098] In order to further explore whether the multivariate saturation quantification model can effectively detect the saturation of optical data when using a small amount of ground data, this embodiment uses the FSV prediction results obtained by the optimal model among the three tree species to construct a multivariate saturation quantification model, and calculates the corresponding saturation values, such as Fig.10 When only optical data and a small number of ground survey samples were used for modeling, obvious data saturation occurred. The saturation value of the eucalyptus sample from the RF model was 101m 3 / ha, and the saturation values ​​of larch and pine samples from the SVM model were 279m 3 / ha and 314m 3 / ha. The results show that the multivariate saturation quantification model constructed using a small number of ground samples is basically consistent with the data saturation values ​​obtained using a large number of LiDAR_FSV samples. Therefore, it is feasible to use a small number of evenly distributed ground survey samples to evaluate the saturation of optical data and modeling methods.

[0099] The multivariate saturation quantification model constructed based on the estimation error can effectively evaluate the saturation level of each data source, variable combination method and estimation model according to the estimation results. In the FSV remote sensing estimation process, the spatial resolution of the remote sensing data source and the estimation model are the main factors affecting the FSV estimation accuracy and saturation level, and different variable selection methods have little effect on the estimation accuracy and saturation level. This embodiment evaluates the impact of the spatial resolution of optical remote sensing data, linear and nonlinear variable selection methods, and four regression models of random forest, K-nearest neighbor, support vector machine and multivariate linear regression on the saturation level of FSV estimation based on the multivariate saturation quantification model. The results show that the sensitivity between optical variables and FSV, the prediction accuracy of FSV and the saturation level gradually increase with the decrease of spatial resolution, and tend to be stable when the spatial resolution is lower than 20m and 25m. When estimating FSV using optical remote sensing data with different spatial resolutions, there is no obvious difference in the saturation starting point, and the range is 3m. 3 / ha(Eucalyptus), 3m 3 / ha (larch) and 5m 3 / ha(Pinus tabulaeformis);FSV ESAT Affected by the spatial resolution, the range is 89m. 3 / ha(eucalyptus), 66m 3 / ha (larch) and 71m 3 / ha (Pinus tabulaeformis). Among the four regression models, the saturation levels of the three non-parametric models were significantly affected by the spatial resolution of the optical data, while the saturation level of the multivariate linear regression model was almost unaffected by the spatial resolution of the optical data.

[0100] The embodiment of the present invention also provides a saturation quantification device in remote sensing estimation of forest volume, such as Fig.11 As shown, the device 1100 includes:

[0101] The curve fitting module 1101 is configured to determine a fitting curve based on the estimated value and the observed value of the FSV, wherein the fitting curve includes a first curve, a second curve, a third curve, a fourth curve, a fifth curve and a sixth curve, wherein the first curve and the third curve represent the range of random errors, the second curve is a 1:1 straight line representing the spatial distribution of the FSV and the relationship between the predicted value and the observed value when there is no error in the predicted value, the fourth curve and the sixth curve represent the range of saturation errors, and the fifth curve is a scatter plot fitting curve of the observed value and the predicted value;

[0102] An indicator determination module 1102 is configured to determine a saturation starting point and a saturation point based on the fitting curve;

[0103] The saturation quantization module 1103 is configured to evaluate the saturation level in the FSV estimation according to the saturation starting point and the saturation point.

[0104] In some embodiments, the indicator determination module is further configured to determine the saturation starting point based on the fitting curve by the following method:

[0105] Determine an observation value S1 among all observation values, and take the observation value S1 as a saturation starting point;

[0106] For the observed value S1, when the FSV is less than the observed value S1, the estimation error consists of random errors, and when the FSV is greater than the observed value S1, the predicted value of the FSV is saturated.

[0107] In some embodiments, the indicator determination module is further configured to determine the sum of the saturation points based on the fitting curve by the following method:

[0108] Determine an observation value S among all observation values, and take the observation value S as the saturation starting point;

[0109] The corresponding points of the observation value S on the fourth curve, the fifth curve and the sixth curve are A, B and C, wherein the length of the line segment AB is equal to the length of the line segment BC.

[0110] In some embodiments, the first curve, the second curve, the third curve, the fourth curve, the fifth curve and the sixth curve are respectively expressed by the following formulas (1) to (6):

[0111] Y c =x+e (1)

[0112] Y b =x (2)

[0113] Y c =xe (3)

[0114] Y d =kx+b+e (4)

[0115] Y e =kx+b (5)

[0116] Y f =kx+be (6)

[0117] In the formula, Y c , Y b , Y c , Y d , Y e and Y fare the predicted values ​​of the corresponding observed values ​​x in the first, second, third, fourth, fifth and sixth curves respectively, x is the observed value of FSV, k and b are constants, e is the random error of the predicted value, that is, the average distance from the predicted value of all samples to the fitting line e, expressed as Where n is the total number of samples, y i is the predicted value of sample i, y ei is the corresponding point on the fifth curve.

[0118] In some embodiments, the indicator determination module is further configured to calculate the saturation starting point by the following formula:

[0119] FSV SP =b / (1-k) (7)

[0120] Where, FSV SP Indicates the saturation starting point.

[0121] In some embodiments, the indicator determination module is further configured to calculate the saturation point by the following formula:

[0122] FSV ESAT =(b+e) / (1-k) (8)

[0123] Where, FSV ESAT Indicates the saturation starting point.

[0124] In some embodiments, the saturation quantization module is further configured to:

[0125] When FSV reaches FSV SP When , the estimated value of FSV begins to be systematically underestimated;

[0126] When FSV reaches FSV ESAT When , the mean value of the FSV estimate is lower than the minimum value of the FSV within the random error range.

[0127] It should be noted that the various device structures described in this embodiment belong to the same technical concept as the previously described method, and achieve the same technical effect through the same principle, which will not be repeated here.

[0128] An embodiment of the present invention further provides a readable storage medium, which stores one or more programs. The one or more programs can be executed by one or more processors to implement the method described in any of the above embodiments.

[0129] In addition, although exemplary embodiments have been described herein, the scope includes any and all embodiments based on the present invention with equivalent elements, modifications, omissions, combinations (e.g., various embodiments intersecting schemes), adaptations or changes. The elements in the claims will be interpreted broadly based on the language adopted in the claims, and are not limited to the examples described in this specification or during the implementation of this application, and the examples will be interpreted as non-exclusive. Therefore, this specification and examples are intended to be considered as examples only, and the true scope and spirit are indicated by the following claims and the full scope of their equivalents.

[0130] The above description is intended to be illustrative rather than restrictive. For example, the above examples (or one or more of them) can be used in combination with each other. For example, those of ordinary skill in the art can use other embodiments when reading the above description. In addition, in the above-mentioned specific embodiments, various features can be grouped together to simplify the present invention. This should not be interpreted as a feature of an invention that is not claimed for protection being necessary for any claim. On the contrary, the subject matter of the present invention may be less than all the features of the embodiments of a specific invention. Thus, the following claims are incorporated into the specific embodiments as examples or embodiments, wherein each claim is independently used as a separate embodiment, and it is considered that these embodiments can be combined with each other in various combinations or arrangements. The scope of the present invention should be determined with reference to the full scope of the equivalent forms of the attached claims and these claims.

Claims

1. A saturation quantification method for remote sensing estimation of forest volume, characterized in that: The method comprises: Determine a fitting curve based on the estimated value and the observed value of FSV, the fitting curve comprising a first curve, a second curve, a third curve, a fourth curve, a fifth curve and a sixth curve, the first curve and the third curve represent the range of random errors, the second curve is a 1:1 straight line, representing the spatial distribution of FSV and the relationship between the predicted value and the observed value when there is no error in the predicted value, the fourth curve and the sixth curve represent the range of saturation errors, and the fifth curve is a scatter plot fitting curve of the observed value and the predicted value; Determine a saturation starting point and a saturation point based on the fitting curve; The saturation level in the FSV estimation is evaluated based on the saturation starting point and the saturation point.

2. The method according to claim 1, characterized in that Based on the fitting curve, the saturation starting point is determined by the following method: Determine an observation value S1 among all observation values, and take the observation value S1 as a saturation starting point; For the observed value S1, when the FSV is less than the observed value S1, the estimation error consists of random errors, and when the FSV is greater than the observed value S1, the predicted value of the FSV is saturated.

3. The method according to claim 1, characterized in that Based on the fitting curve, the sum saturation point is determined by the following method: Determine an observation value S among all observation values, and take the observation value S as the saturation starting point; The corresponding points of the observation value S on the fourth curve, the fifth curve and the sixth curve are A, B and C, wherein the length of the line segment AB is equal to the length of the line segment BC.

4. The method according to claim 1, characterized in that: The first curve, the second curve, the third curve, the fourth curve, the fifth curve and the sixth curve are respectively expressed by the following formulas (1) to (6): AND c =x+e (1) AND b =x (2) AND c =xe (3) Y d kx+b+e (4) Yes e =kx+b (5) Y f =kx+be (6) In the formula, Y c , Y b , Y c , Y d , Y e and Y f are the predicted values ​​of the corresponding observed values ​​x in the first, second, third, fourth, fifth and sixth curves respectively, x is the observed value of FSV, k and b are constants, e is the random error of the predicted value, that is, the average distance from the predicted value of all samples to the fitting line e, expressed as Where n is the total number of samples, y i is the predicted value of sample i, y ei is the corresponding point on the fifth curve.

5. The method according to claim 4, characterized in that The saturation starting point is calculated by the following formula: FSV SP =b / (1-k) (7) Where, FSV SP Indicates the saturation starting point.

6. The method according to claim 4, characterized in that The saturation point is calculated using the following formula: FSV ESAT =(b+e) / (1-k) (8) Where, FSV ESAT Indicates the saturation starting point.

7. The method according to claim 1, characterized in that The saturation level in the FSV estimation is evaluated based on the saturation starting point and the saturation point, specifically including: When FSV reaches FSV SP When , the estimated value of FSV begins to be systematically underestimated; When FSV reaches FSV ESAT When , the mean value of the FSV estimate is lower than the minimum value of the FSV within the random error range.

8. A saturation quantification device for remote sensing estimation of forest volume, characterized in that: The device comprises: A curve fitting module is configured to determine a fitting curve based on an estimated value and an observed value of the FSV, wherein the fitting curve includes a first curve, a second curve, a third curve, a fourth curve, a fifth curve and a sixth curve, wherein the first curve and the third curve represent a range of random errors, the second curve is a 1:1 straight line representing a spatial distribution of the FSV and a relationship between the predicted value and the observed value when there is no error in the predicted value, the fourth curve and the sixth curve represent a range of saturation errors, and the fifth curve is a scatter plot fitting curve of the observed value and the predicted value; An indicator determination module, configured to determine a saturation starting point and a saturation point based on the fitting curve; The saturation quantization module is configured to evaluate a saturation level in FSV estimation according to the saturation starting point and the saturation point.

9. The device according to claim 8, characterized in that The saturation quantization module is further configured as follows: When FSV reaches FSV SP When , the estimated value of FSV begins to be systematically underestimated; When FSV reaches FSV ESAT When , the mean value of the FSV estimate is lower than the minimum value of the FSV within the random error range. 10 . A non-transitory computer-readable storage medium storing instructions, which, when executed by a processor, perform the method according to claim 1 .

Images