A forestry resource intelligent supervision method and system based on satellite remote sensing technology

Abstract

The application discloses a kind of forestry resource intelligent supervision method and system based on satellite remote sensing technology, belong to forestry resource supervision technical field, including the following steps: based on multi-temporal satellite remote sensing image, to forest area pixel is layered according to canopy density gradient, the obtained each canopy density layer is constructed shielding effect weight, obtain canopy density layering weight chart;With canopy density layering weight chart as constraint, the pixel radiation value of each canopy density layer is corrected, and the radiation correction image is obtained;Extract the spectral feature of the radiation correction image, weight the spectral feature with the canopy density layering weight chart, and obtain the canopy density perception feature set.The application constructs canopy density layering weight chart, uses canopy density gradient information as unified constraint throughout the whole process of radiation correction, spectral feature weighting and abnormal detection threshold setting, solves the problem of radiation correction deviation and abnormal detection misalignment caused by ignoring the difference of canopy density in the existing method.

Description

Technical Field

[0001] This invention relates to the field of forestry resource monitoring technology, specifically to an intelligent monitoring method and system for forestry resources based on satellite remote sensing technology. Background Technology

[0002] With the increasing demand for ecological protection and forestry resource management, satellite remote sensing technology has been widely used for resource monitoring and anomaly detection in large-scale forest areas.

[0003] However, in actual monitoring scenarios, the canopy closure levels vary across different areas within forest regions, causing shading interference with the radiometric signals of remote sensing images. Taking the dynamic monitoring of stock volume in the southern subtropical mixed forest region as an example, high-canopy-closure patches coexist within the forest area. When uniform radiometric correction parameters and feature extraction methods are applied to all pixels in the region, the radiometric values ​​of pixels in high-canopy-closure areas are systematically underestimated due to the strong shading effect of the canopy layer on the surface reflection signals, leading to deviations in the stock volume regression results for that region.

[0004] Based on this, if a uniform threshold for judging abnormal changes in stock volume is adopted for the entire region, false alarms or omissions may occur in high-canopy-density areas because the basic fluctuation characteristics are different from those in low-canopy-density areas, making it difficult for regulators to accurately locate forest areas where abnormal changes have actually occurred. Summary of the Invention

[0005] In view of the above-mentioned problems, the present invention provides a method and system for intelligent monitoring of forestry resources based on satellite remote sensing technology.

[0006] To address the aforementioned technical problems, this invention provides the following technical solution: an intelligent forestry resource monitoring method based on satellite remote sensing technology, comprising the following steps: based on multi-temporal satellite remote sensing images, forest area pixels are layered according to canopy closure gradients, and shading effect weights are constructed for each obtained canopy closure layer to obtain a canopy closure layer weight map; using the canopy closure layer weight map as a constraint, the pixel radiance values ​​of each canopy closure layer are corrected to obtain a radiance-corrected image; the spectral features of the radiance-corrected image are extracted, and the spectral features are weighted using the canopy closure layer weight map to obtain a canopy closure perception feature set; volume regression inference is performed on the canopy closure perception feature set, the change in volume results for adjacent canopy closure layers is calculated, abnormal forest areas are located based on each change, and a forestry resource monitoring report is output.

[0007] As a preferred embodiment of the intelligent forestry resource monitoring method based on satellite remote sensing technology described in this invention, the step of obtaining the canopy closure layer weight map includes: calculating the estimated canopy closure value for each pixel in the forest area based on multi-temporal satellite remote sensing images; dividing the pixels according to the estimated canopy closure value and a preset threshold interval to obtain canopy closure layers; assigning shading effect weights to each canopy closure layer; and writing the shading effect weights corresponding to the canopy closure layer to which each pixel belongs into a spatial raster to generate a canopy closure layer weight map.

[0008] As a preferred embodiment of the intelligent forestry resource monitoring method based on satellite remote sensing technology described in this invention, the step of obtaining the radiometrically corrected image includes: using the calculated surface reflectance as the pixel radiometric value for each pixel in each canopy closure layer, and performing weighted correction on the pixel radiometric value using the corresponding weights in the canopy closure layer weight map; verifying the correction results of pixels in the same canopy closure layer in adjacent temporal satellite remote sensing images, and if the difference in pixel radiometric value after correction exceeds a preset threshold, then re-weighting correction is performed on the pixel in an iterative manner; merging the pixel radiometric values ​​after correction of each canopy closure layer according to spatial location to generate a radiometrically corrected image.

[0009] As a preferred embodiment of the intelligent forestry resource monitoring method based on satellite remote sensing technology described in this invention, the step of obtaining the canopy closure perception feature set includes: taking the spectral feature vector of each pixel extracted from the radiometrically corrected image, and using the corresponding weight in the canopy closure layer weight map as the weighting coefficient to obtain the canopy closure perception feature vector; performing time-series splicing of the canopy closure perception feature vectors of each time phase to construct a time-series canopy closure perception feature matrix, which is output as the canopy closure perception feature set.

[0010] As a preferred embodiment of the intelligent forestry resource monitoring method based on satellite remote sensing technology described in this invention, the step of performing volume regression inference on the canopy closure sensing feature set includes: taking the temporal canopy closure sensing feature matrix of each pixel in the canopy closure sensing feature set as input, serializing and flattening the temporal canopy closure sensing feature matrix into a canopy closure sensing feature vector, and then estimating the volume of each pixel to obtain the volume estimation value. The calculation formula is expressed as:

[0011] ;

[0012] In the formula, This is the estimated accumulation value for the i-th pixel. For the total number, Let be the regression function of the t-th decision tree. This is a feature vector for canopy closure perception.

[0013] The estimated volume values ​​are aggregated at the stand / pattern level to obtain the patch volume.

[0014] As a preferred embodiment of the intelligent forestry resource monitoring method based on satellite remote sensing technology described in this invention, the step of performing stand patch-level aggregation of the estimated volume includes: spatially merging the estimated volume of each pixel according to its forest stand patch based on the spatial boundary of the forest stand patch; summing the estimated volume of all pixels within the forest stand patch to obtain the patch volume of each forest stand patch.

[0015] As a preferred embodiment of the intelligent forestry resource monitoring method based on satellite remote sensing technology described in this invention, the step of calculating the change in the volume of adjacent time phases according to the canopy closure layer and locating abnormal forest areas based on each change includes: calculating the relative change rate of the volume of patches within the same canopy closure layer in adjacent time phases; setting an anomaly judgment threshold, and marking an abnormal forest area when the change rate exceeds the anomaly judgment threshold corresponding to its canopy closure layer.

[0016] As a preferred embodiment of the intelligent forestry resource monitoring method based on satellite remote sensing technology described in this invention, the step of setting anomaly judgment thresholds includes: based on the canopy closure stratification weight map, calculating the time-series standard deviation of the relative change rate of the volume of all forest patches in each historical phase within each canopy closure layer, and using the time-series standard deviation as the baseline fluctuation level; and setting anomaly judgment thresholds for each canopy closure layer according to the baseline fluctuation level of each canopy closure layer, with the anomaly judgment threshold corresponding to a high canopy closure layer being lower than the anomaly judgment threshold corresponding to a low canopy closure layer.

[0017] As a preferred embodiment of the intelligent forestry resource monitoring method based on satellite remote sensing technology described in this invention, the steps for determining high-canopy-density layers and low-canopy-density layers include: sorting the shading effect weights of all canopy-density layers, and using the average of the shading effect weights of all canopy-density layers as the boundary benchmark; canopy-density layers with shading effect weights higher than the boundary benchmark are determined to be high-canopy-density layers; and canopy-density layers with shading effect weights lower than the boundary benchmark are determined to be low-canopy-density layers.

[0018] This invention provides an intelligent monitoring system for forestry resources based on satellite remote sensing technology.

[0019] To solve the above technical problems, the present invention provides the following technical solution: an intelligent forestry resource monitoring system based on satellite remote sensing technology, comprising: a weight map framework module, which, based on multi-temporal satellite remote sensing images, divides forest area pixels into layers according to canopy closure gradient, and constructs shading effect weights for each obtained canopy closure layer to obtain a canopy closure layer weight map.

[0020] The correction module uses the canopy closure layer weight map as a constraint to correct the pixel radiation values ​​of each canopy closure layer, thus obtaining a radiation-corrected image.

[0021] The feature extraction module is used to extract the spectral features of the radiometrically corrected image, and to perform weighted processing on the spectral features using a canopy closure layer weight map to obtain a canopy closure perception feature set.

[0022] The output module is used to perform volume regression inference on the canopy closure perception feature set, calculate the change in volume of adjacent canopy closure layers, locate abnormal forest areas based on each change, and output a forestry resource supervision report.

[0023] The beneficial effects of this invention are as follows: By constructing a canopy closure layer weight map, canopy closure gradient information is used as a unified constraint throughout the entire process of radiometric correction, spectral feature weighting, and anomaly detection threshold setting. This solves the problems of radiometric correction bias and anomaly detection inaccuracy caused by neglecting canopy closure differences in existing methods. Specifically, radiometric correction with corresponding shading effect weights is applied to different canopy closure layers, so that the pixel radiometric values ​​in high canopy closure areas are reasonably compensated, improving the accuracy of volume regression estimation. At the same time, anomaly judgment thresholds are set according to the historical fluctuation levels of each canopy closure layer, making the location results of abnormal forest areas more consistent with the actual volume change patterns of each layer, reducing false alarm and false negative rates, and improving the overall reliability of intelligent forestry resource supervision. Attached Figure Description

[0024] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 The above is a flowchart of an intelligent forestry resource monitoring method based on satellite remote sensing technology, provided as an embodiment of the present invention. Detailed Implementation

[0026] To make the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0027] Example 1, referring to Figure 1This is one embodiment of the present invention, which provides an intelligent monitoring method for forestry resources based on satellite remote sensing technology, comprising the following steps:

[0028] S1. Based on multi-temporal satellite remote sensing images, forest area pixels are layered according to canopy closure gradient, and shading effect weights are constructed for each canopy closure layer to obtain a canopy closure layer weight map.

[0029] S2. Using the canopy closure layer weight map as a constraint, the pixel radiation values ​​of each canopy closure layer are corrected to obtain a radiation-corrected image.

[0030] S3. Extract the spectral features of the radiometrically corrected image, and weight the spectral features using a canopy closure layer weight map to obtain a canopy closure perception feature set.

[0031] S4. Perform volume regression inference on the canopy closure perception feature set, calculate the change in volume for adjacent volume results according to the canopy closure layer, locate abnormal forest areas based on each change, and output a forestry resource supervision report.

[0032] It should be noted that if a uniform threshold for judging abnormal changes in forest volume is adopted for the entire region, false alarms or omissions may occur in high-canopy-density areas because the basic fluctuation characteristics are different from those in low-canopy-density areas, making it difficult for regulators to accurately locate forest areas where abnormal changes have actually occurred.

[0033] This invention constructs a layered weighted map of canopy closure, using canopy closure gradient information as a unified constraint throughout the entire process of radiometric correction, spectral feature weighting, and anomaly detection threshold setting. This solves the problems of radiometric correction bias and anomaly detection inaccuracies caused by neglecting differences in canopy closure in existing methods. Specifically, radiometric correction with corresponding shading effect weights is applied to different canopy closure layers, ensuring reasonable compensation for pixel radiometric values ​​in high-canopy-closure areas and improving the accuracy of volume regression estimation. Simultaneously, anomaly judgment thresholds are set based on the historical fluctuation levels of each canopy closure layer, making the location results of abnormal forest areas more consistent with the actual volume change patterns of each layer, reducing false alarm and false negative rates, and improving the overall reliability of intelligent forestry resource monitoring.

[0034] Example 2 is an embodiment of the present invention, which provides an intelligent monitoring method for forestry resources based on satellite remote sensing technology, based on the previous embodiment, including the following steps:

[0035] S1. Based on multi-temporal satellite remote sensing images, forest area pixels are layered according to canopy closure gradient, and shading effect weights are constructed for each canopy closure layer to obtain a canopy closure layer weight map.

[0036] The steps for obtaining the canopy closure stratification weight map include S1.1~S1.4:

[0037] S1.1. Based on multi-temporal satellite remote sensing images, calculate the estimated canopy closure value for each pixel in the forest area.

[0038] Specifically, for multi-temporal satellite remote sensing images that have undergone atmospheric correction and geometric registration preprocessing, the estimated value of canopy closure is calculated for each pixel within the forest area using the empirical regression relationship between the normalized vegetation index and canopy closure.

[0039] First, extract the red band reflectance and near-infrared band reflectance of each pixel in the satellite remote sensing images of each time phase, and then calculate the normalized vegetation index for each time phase:

[0040] ;

[0041] In the formula, Let be the normalized vegetation index for the t-th time phase. and These are the near-infrared band reflectance and red band reflectance at the t-th time phase, respectively.

[0042] Furthermore, the average normalized vegetation index (NDI) for multiple time phases is taken to obtain the time-averaged NDI, thereby reducing the estimation error caused by factors such as cloud cover and seasonal variations in a single time phase.

[0043] ;

[0044] In the formula, The mean normalized vegetation index is T, and the total number of time phases is T=3 in this embodiment.

[0045] Convert the normalized vegetation index mean into an estimate of canopy closure:

[0046] ;

[0047] In the formula, Let be the estimated canopy closure value of the i-th pixel, and a and b be the regression coefficients.

[0048] For example, the normalized vegetation index of this pixel in spring, summer and autumn is 0.52, 0.78 and 0.61 respectively. The mean normalized vegetation index is 0.637. Substituting into the above formula, the estimated canopy closure is 1.12×0.637-0.08=0.634, that is, the canopy closure of this pixel is about 63.4%.

[0049] S1.2. Based on the estimated canopy closure value, the pixels are divided according to the preset threshold range to obtain the canopy closure layer.

[0050] Based on the actual conditions of the forest area and referring to the conventions for canopy closure classification, the range of estimated canopy closure values ​​in the range [0,1] is divided into four preset threshold ranges, corresponding to four canopy closure layers. Taking the example pixel in correction S1.1 as an example, its estimated canopy closure value is 0.634, which falls into the range [0.5,0.7). Therefore, this pixel is classified into the medium-high canopy closure layer.

[0051] S1.3, assign shading effect weights to each canopy closure layer respectively.

[0052] The shading effect weight is used to quantify the degree of shading of the canopy on the radiation signal of the understory by different canopy closure layers. The higher the canopy closure, the greater the degree of canopy shading, and the higher the weight assigned.

[0053] In this embodiment, based on the intra-layer mean of the estimated canopy closure values ​​for each canopy closure layer, the shading effect weight of each canopy closure layer is calculated using the following formula:

[0054] ;

[0055] In the formula, The weight of the shading effect of the k-th canopy closure layer. The mean of the estimated canopy closure values ​​of all pixels in the k-th canopy closure layer is given by K, where K is the total number of canopy closure layers. In this embodiment, K=4.

[0056] S1.4 Write the shading effect weights corresponding to the canopy closure layer of each pixel into the spatial raster to generate a canopy closure layer weight map.

[0057] After obtaining the shading effect weights of each canopy closure layer, based on the canopy closure layer classification of all pixels in the forest area in step S1.2, the shading effect weight value corresponding to the canopy closure layer of each pixel is written into a raster dataset with the same spatial resolution, coordinate system and spatial range as the original remote sensing image. Each spatial raster cell stores the shading effect weight value of its respective canopy closure layer, thereby generating a canopy closure layer weight map.

[0058] For example, if a pixel belongs to a medium-high canopy closure layer, the corresponding shading effect weight is 0.44. Therefore, the pixel value of the corresponding raster in the canopy closure layer weight map is written as 0.44.

[0059] The final generated canopy closure layer weight map is a single-band raster image covering the entire forest area. Its pixel values ​​include only four values ​​(0.08, 0.25, 0.44, 0.61), which correspond to the shading effect weights of the four canopy closure layers, respectively.

[0060] S2. Using the canopy closure layer weight map as a constraint, the pixel radiation values ​​of each canopy closure layer are corrected to obtain a radiation-corrected image.

[0061] The steps for obtaining the radiometrically corrected image include S2.1 to S2.3:

[0062] S2.1. Using the calculated surface reflectance as the pixel radiation value, the pixel radiation value is weighted and corrected using the corresponding weights in the canopy closure layer weight map.

[0063] In satellite remote sensing observations, the radiation signal received by the sensor is a mixture of canopy-reflected radiation and forest surface-reflected radiation. In areas with higher canopy closure, the canopy more strongly obscures the forest surface radiation signal. Without correction, directly using the raw radiation values ​​would lead to incomparable volume estimates between areas with different canopy closures. Therefore, this step uses surface reflectance as the representation of pixel radiation values ​​and applies weighted correction based on the corresponding shading effect weights in the canopy closure stratification weight map to eliminate the systematic influence of canopy shading differences on the radiation signal.

[0064] Specifically, for a pixel with a spatial location of (x,y) within the forest area, its atmospherically corrected surface reflectance is first extracted as the original pixel radiance value, where b represents the band number and t represents the time phase number.

[0065] This embodiment uses surface reflectance data from three bands and three time phases: red band, near-infrared band, and short-wave infrared band.

[0066] Subsequently, the shading effect weight of the corresponding raster is read from the canopy closure layer weight map, and the radiance value of the pixel is weighted and corrected according to the following formula to obtain the corrected pixel radiance value:

[0067] ;

[0068] In the formula, To correct the pixel radiation value, Where is the original surface reflectance, W(x,y) is the shading effect weight, and α is the shading correction coefficient, which is used to control the adjustment range of the shading effect weight on the radiation value correction. In this embodiment, α=0.35 is obtained based on the ground measured sample calibration.

[0069] Taking a certain pixel as an example, it belongs to the medium-high canopy closure layer, and the shading effect weight is 0.44. Its original surface reflectance in the near-infrared band at time 1 is 0.312. Substituting it into the above formula, we get the corrected pixel radiation value, which is 0.369 after calculation.

[0070] S2.2. The correction results of pixels in the same canopy closure layer in adjacent time-phase satellite remote sensing images are checked. If the difference in the radiance value of the pixel after correction exceeds the preset threshold, the pixel is re-weighted and corrected in an iterative manner.

[0071] Specifically, for adjacent time phases t and t+1, for each pixel within the same canopy closure layer k, the relative difference of its corrected pixel radiance value across each band is calculated. The average of the relative differences across each band is then used to obtain the comprehensive temporal difference index for that pixel.

[0072] A preset threshold for consistency testing is set, which varies depending on the canopy closure level. For the four canopy closure levels, the thresholds are set to 0.25, 0.20, 0.15, and 0.12, respectively. The threshold for high canopy closure levels is smaller because the canopy structure is stable in high-canopy-closure areas, and the normal fluctuation range of radiation values ​​in adjacent time phases is relatively small, making them more sensitive to abnormal changes.

[0073] When the overall temporal difference index exceeds a preset threshold, the pixel correction result is deemed to have failed the consistency check, triggering iterative correction. The iterative correction process includes:

[0074] Based on the current corrected radiation value, the shading correction coefficient α is adjusted according to the following formula to obtain the updated correction coefficient for the nth iteration:

[0075] ;

[0076] In the formula, Here, λ is the correction coefficient for the nth iteration, and λ is the iteration step size control parameter. In this embodiment, λ is set to 0.1. This is the sign function. When the comprehensive phase difference index exceeds the threshold, this formula drives the correction coefficient to adjust in a decreasing direction, thereby reducing the correction amplitude and causing the difference in radiation values ​​between adjacent phases to converge within the threshold range.

[0077] Substitute the updated correction coefficients into the weighted correction formula and recalculate the corrected radiometric value of the pixel. Repeat this process iteratively until the iteration condition is met.

[0078] Taking a certain L3 layer pixel as an example, the calculated comprehensive temporal difference index after the initial correction is 0.22, which exceeds the corresponding threshold of 0.15 for L3 layer, triggering iterative correction. After the first iteration, the output correction coefficient is 0.334, and the comprehensive temporal difference index drops to 0.17, which still exceeds the threshold. The second iteration continues until the comprehensive temporal difference index is less than or equal to the preset threshold.

[0079] S2.3. Merge the pixel radiation values ​​after correction of each canopy closure layer according to spatial location to generate a radiation-corrected image.

[0080] S3. Extract the spectral features of the radiometrically corrected image, and weight the spectral features using a canopy closure layer weight map to obtain a canopy closure perception feature set.

[0081] The steps for obtaining the canopy closure perception feature set include S3.1~S3.2:

[0082] S3.1. For the spectral feature vector of each pixel extracted from the radiometrically corrected image, the corresponding weight in the canopy closure layer weight map is used as the weighting coefficient to obtain the canopy closure perception feature vector.

[0083] Specifically, for a pixel with a spatial location of (x,y) in a radiometrically corrected image, the following spectral feature vectors are extracted at the t-th time phase, including the extraction of red band corrected reflectance, near-infrared band corrected reflectance, and short-wave infrared band corrected reflectance, as well as the extraction of normalized vegetation index, enhanced vegetation index, and normalized differential moisture index.

[0084] The above 6 spectral feature components are arranged in order to form the spectral feature vector of the pixel at the t-th time phase and spatial location. The shading effect weight corresponding to the pixel is read from the canopy closure layer weight map. Each component in the spectral feature vector is weighted to obtain the canopy closure perception feature vector.

[0085] S3.2. Perform time-series splicing of the canopy closure perception feature vectors of each time phase to construct a time-series canopy closure perception feature matrix, which is output as the canopy closure perception feature set.

[0086] Specifically, for a pixel with a spatial location of (x,y) in the forest area, its canopy closure perception feature vectors in T time phases are spliced ​​together in the row direction according to the time phase order to construct a time-series canopy closure perception feature matrix.

[0087] The matrix has a dimension of T×D, where T is the total number of time phases and D is the dimension of the single-phase canopy closure sensing feature vector. In this embodiment, T=3 and D=6, so the dimension of the temporal canopy closure sensing feature matrix corresponding to each pixel is 3×6. The t-th row of the matrix corresponds to the canopy closure sensing feature vector of the t-th time phase, and the d-th column corresponds to the value sequence of the d-th spectral feature component in each time phase.

[0088] S4. Perform volume regression inference on the canopy closure perception feature set, calculate the change in volume for adjacent volume results according to the canopy closure layer, locate abnormal forest areas based on each change, and output a forestry resource supervision report.

[0089] The steps for performing accumulation regression inference on the canopy closure perception feature set include S4.1~S4.2:

[0090] S4.1. Using the temporal canopy closure sensing feature matrix of each pixel in the canopy closure sensing feature set as input, after serializing and flattening the temporal canopy closure sensing feature matrix into a canopy closure sensing feature vector, the accumulation amount of each pixel is estimated to obtain the accumulation amount estimate. The calculation formula is expressed as follows:

[0091] ;

[0092] In the formula, This is the estimated accumulation value for the i-th pixel. For the total number, Let be the regression function of the t-th decision tree. This is the feature vector for canopy closure perception.

[0093] For a pixel at spatial location (x, y) in the canopy closure sensing feature set, its temporal canopy closure sensing feature matrix has dimensions T×D. In this embodiment, T=3, D=6, and the matrix dimension is 3×6. This matrix is ​​then serialized and flattened in row-major order to obtain the canopy closure sensing feature vector. The dimension is T×D=18, and the flattening operation is represented as:

[0094] ;

[0095] In the formula, This is the temporal canopy closure perception feature matrix. Let be the canopy closure sensing feature value corresponding to the d-th spectral feature component in the t-th time phase. The flattening operation transforms the two-dimensional temporal matrix into a one-dimensional vector, which can then be used as the input feature of the decision tree regression model.

[0096] After obtaining the canopy closure perception feature vector, a random forest regression model is used to estimate the volume of each pixel. The random forest regression model consists of T decision trees. In this embodiment, the total number of decision trees is T=100. During the training phase, each decision tree is trained in a supervised manner using the volume survey data of the forest area ground measured samples as labels and the canopy closure perception feature vector of the corresponding pixel as input. The training samples consist of 350 forest stand standard plot measured records, with the measured volume ranging from 15 to 420.

[0097] For example, the 18-dimensional canopy closure sensing feature vector is input into a trained random forest model. Each of the 100 decision trees outputs a volume regression value. The average of all regression values ​​is taken to obtain an estimated volume value of 187.3 for that pixel. The above operation is repeated for all pixels in the canopy closure sensing feature set to obtain a raster result of pixel-by-pixel volume estimation values ​​covering the entire forest area.

[0098] S4.2. Aggregate the estimated volume values ​​at the stand / pattern level to obtain the patch volume.

[0099] The steps for aggregated forest stand patches of the estimated volume include A1 to A2:

[0100] A1. Based on the spatial boundaries of forest stand patches, the estimated volume of each pixel is spatially merged according to the forest stand patch to which it belongs.

[0101] It is important to know that forest stand patches are the basic spatial units for forestry resource surveys and management, and their spatial boundaries are provided by forestry survey vector data.

[0102] In this embodiment, a total of 382 forest stands were delineated in the study area, with the area of ​​each stand ranging from 0.8 to 35.6 hm². 2 Based on the spatial boundary vector data of each forest stand patch, the raster of estimated volume per pixel is spatially merged, and all pixels whose spatial location falls within the boundary range of the same forest stand patch are identified as the set of pixels to which the patch belongs.

[0103] For a forest stand patch j, its set of associated pixels is denoted as . ,in, Let j be the spatial boundary range of the j-th forest stand patch. Let be the spatial coordinates of the i-th pixel. Spatial merging uses the criterion that the pixel center point falls within the patch boundary. For pixels whose center point falls exactly on the patch boundary line, they are assigned to the adjacent patch with the smaller number.

[0104] A2. Sum the estimated volume values ​​of all pixels within the forest stand patch to obtain the patch volume of each forest stand patch.

[0105] For the j-th forest stand patch, the patch volume is obtained by multiplying the estimated volume of all pixels in its set by the area of ​​a single pixel and then summing the results.

[0106] For example, the area of ​​the patch is 4.2 hm². 2 The number of assigned pixels is 420, and the estimated average volume of each pixel is 193.6m. 3 / hm 2 Therefore, the patch accumulation is 420 × 193.6 × 0.01 = 812.1 m³. 3 The above aggregation operation was repeated for all 382 forest stands in the study area to obtain the distribution of forest stock volume across the entire area at each time phase.

[0107] The steps for calculating the change in volume between adjacent time phases based on canopy closure layer, and locating anomalous forest areas based on these changes, include S4.3~S4.4:

[0108] S4.3 Calculate the relative rate of change of patch accumulation within the same canopy closure layer at adjacent time phases.

[0109] For the same forest stand patch j, its patch volume at time t and time t+1 are denoted as follows: and The formula for calculating the relative rate of change of patch accumulation is:

[0110] ;

[0111] In the formula, R_j(t,t+1) is the relative rate of change, expressed as a percentage. When it is greater than 0, it indicates that the volume of the patch has increased; conversely, it indicates that the volume of the patch has decreased. Abnormal changes in the direction of decrease (such as illegal logging, fire, pests and diseases) are the key focus of forestry resource supervision.

[0112] The calculation of relative change rate is strictly carried out independently between patches within the same canopy closure layer, that is, the relative change rate of patch accumulation in each canopy closure layer is calculated separately, without cross-layer mixing.

[0113] For example, the plaque accumulation of a certain patch in spring is 812.1 m³. 3 The summer patch accumulation was 694.5 m³. 3 The relative rate of change is -14.5%.

[0114] S4.4 Set an anomaly detection threshold. When the rate of change exceeds the anomaly detection threshold corresponding to its canopy closure layer, it is marked as an abnormal forest area.

[0115] It is also important to know that the steps for determining high-canopy-density and low-canopy-density layers include C1~C2:

[0116] C1. Sort the shading effect weights of all canopy closure layers and use the mean of the shading effect weights of all canopy closure layers as the dividing benchmark.

[0117] The mean value of the shading effect weights of the four canopy closure layers in this embodiment is calculated, and the result is 0.345. 0.345 is then used as the dividing line.

[0118] C2. Canopy layers with a shading effect weight higher than the boundary benchmark are determined to be high canopy layers; canopy layers with a shading effect weight lower than the boundary benchmark are determined to be low canopy layers.

[0119] The steps for setting the anomaly detection threshold include B1~B2:

[0120] B1. Based on the canopy closure stratification weight map, calculate the time series standard deviation of the relative change rate of the stock volume of all forest stand patches in each historical phase within each canopy closure layer, and use the time series standard deviation as the benchmark fluctuation level.

[0121] This embodiment utilizes historical remote sensing imagery data from the past 5 years (a total of 15 time phases) of the study area to calculate the time series of relative change rates of patch volume for each forest stand patch over historical time phases. The time series standard deviation of the relative change rates for all adjacent historical time phases of all forest stand patches within the same canopy closure layer is calculated and used as the baseline fluctuation level for that canopy closure layer.

[0122] B2. Based on the baseline fluctuation level of each canopy closure layer, anomaly judgment thresholds are set for each canopy closure layer. The anomaly judgment thresholds for high canopy closure layers are lower than those for low canopy closure layers.

[0123] Based on the baseline fluctuation level of each canopy closure layer, an anomaly judgment threshold is set for each canopy closure layer. The anomaly judgment threshold is the product of the threshold multiplier and the baseline fluctuation level. In this embodiment, the threshold multiplier is set to 2.0, that is, twice the baseline fluctuation level is used as the anomaly judgment boundary.

[0124] After the anomaly detection threshold is determined, the relative change rate of all forest stand patches in each canopy closure layer within the current monitoring time phase is compared with the anomaly detection threshold corresponding to its respective canopy closure layer. The detection rules include:

[0125] An anomaly is defined as a relative rate of change that is greater than the anomaly threshold, and a normal rate of change is defined as a rate of change that is less than the threshold.

[0126] Taking the layered forest patch in step S4.3 as an example, the relative change rate of this patch is -14.5%, and the corresponding anomaly judgment threshold is 8.4%. Since |-14.5%|=14.5%>8.4%, this patch is marked as an abnormal forest area. Furthermore, since the change rate of this patch is negative, it is judged as an anomaly of declining stock volume, and priority should be given to investigating whether there is illegal logging, natural disasters, or pest infestation.

[0127] Finally, the results of abnormal forest area marking, the estimated volume of each patch, the statistical distribution of relative change rate, and the preliminary analysis of the causes of abnormality are summarized. Combined with information such as the spatial location of abnormal patches, the canopy closure layer to which they belong, the change rate value, and the magnitude of exceeding the threshold, a forestry resource supervision report is generated to provide quantitative basis for the forestry authorities to conduct on-site verification and law enforcement decisions.

[0128] Example 3 is an embodiment of the present invention. This embodiment provides an intelligent forestry resource monitoring system based on satellite remote sensing technology, including a weight map framework module. Based on multi-temporal satellite remote sensing images, the forest area pixels are layered according to the canopy closure gradient, and a shading effect weight is constructed for each obtained canopy closure layer to obtain a canopy closure layer weight map.

[0129] The correction module uses the canopy closure layer weight map as a constraint to correct the pixel radiation values ​​of each canopy closure layer, thus obtaining a radiation-corrected image.

[0130] The feature extraction module is used to extract the spectral features of the radiometrically corrected image, and to perform weighted processing on the spectral features using a canopy closure layer weight map to obtain a canopy closure perception feature set.

[0131] The output module is used to perform volume regression inference on the canopy closure perception feature set, calculate the change in volume of adjacent canopy closure layers, locate abnormal forest areas based on each change, and output a forestry resource supervision report.

[0132] This embodiment also provides an electronic device applicable to a method for intelligent monitoring of forestry resources based on satellite remote sensing technology, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the method for intelligent monitoring of forestry resources based on satellite remote sensing technology proposed in the above embodiment.

[0133] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, it implements a method for intelligent monitoring of forestry resources based on satellite remote sensing technology as proposed in the above embodiments.

[0134] The storage medium proposed in this embodiment belongs to the same inventive concept as the method for intelligent monitoring of forestry resources based on satellite remote sensing technology proposed in the above embodiments. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0135] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0136] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for intelligent monitoring of forestry resources based on satellite remote sensing technology, characterized in that, Includes the following steps: Based on multi-temporal satellite remote sensing images, forest area pixels are stratified according to canopy closure gradient, and shading effect weights are constructed for each canopy closure layer to obtain a canopy closure stratification weight map. Using the canopy closure layer weight map as a constraint, the pixel radiance values ​​of each canopy closure layer are corrected to obtain a radiance-corrected image. The spectral features of the radiometrically corrected image are extracted, and the spectral features are weighted using a canopy closure layer weight map to obtain a canopy closure perception feature set. The volume regression inference is performed on the canopy closure perception feature set. The change in volume of adjacent layers is calculated according to the canopy closure layer. Abnormal forest areas are located based on each change, and a forestry resource supervision report is output.

2. The intelligent forestry resource monitoring method based on satellite remote sensing technology as described in claim 1, characterized in that, The steps for obtaining the canopy closure stratification weight map include: Based on multi-temporal satellite remote sensing images, the estimated canopy density is calculated for each pixel in the forest area; Based on the estimated canopy closure value, the pixels are divided according to a preset threshold range to obtain the canopy closure layer; Each canopy closure layer is assigned a weight for its shading effect. The shading effect weights corresponding to the canopy closure layer of each pixel are written into the spatial raster to generate a canopy closure layer weight map.

3. The intelligent forestry resource monitoring method based on satellite remote sensing technology as described in claim 2, characterized in that, The steps for obtaining the radiometrically corrected image include: Using the pixels within each canopy closure layer, the calculated surface reflectance is used as the pixel radiation value, and the pixel radiation value is weighted and corrected using the corresponding weights in the canopy closure layer weight map. The correction results of pixels within the same canopy closure layer in adjacent temporal satellite remote sensing images are examined. If the difference in the radiance value of pixels after correction exceeds a preset threshold, the pixel is re-weighted and corrected in an iterative manner. The radiometric values ​​of pixels corrected for each canopy closure layer are merged according to their spatial location to generate a radiometrically corrected image.

4. The intelligent forestry resource monitoring method based on satellite remote sensing technology as described in claim 3, characterized in that, The steps for obtaining the canopy closure perception feature set include: For each pixel's spectral feature vector extracted from the radiometrically corrected image, the corresponding weight in the canopy closure layer weight map is used as the weighting coefficient to obtain the canopy closure perception feature vector. The canopy closure perception feature vectors of each time phase are spliced ​​together in time series to construct a time series canopy closure perception feature matrix, which is output as the canopy closure perception feature set.

5. The intelligent forestry resource monitoring method based on satellite remote sensing technology as described in claim 4, characterized in that, The steps for performing accumulation regression inference on the feature set of canopy closure perception include: Using the temporal canopy closure sensing feature matrix of each pixel in the canopy closure sensing feature set as input, the temporal canopy closure sensing feature matrix is ​​serialized and flattened into a canopy closure sensing feature vector. The accumulation amount is then estimated for each pixel to obtain the estimated accumulation amount. The calculation formula is as follows: ； In the formula, This is the estimated accumulation value for the i-th pixel. For the total number, Let be the regression function of the t-th decision tree. This is a feature vector for canopy closure perception. The estimated volume values ​​are aggregated at the stand / pattern level to obtain the patch volume.

6. The intelligent forestry resource monitoring method based on satellite remote sensing technology as described in claim 5, characterized in that, The steps for aggregated stand patch-level data on volume estimates include: Based on the spatial boundaries of forest stand patches, the estimated volume of each pixel is spatially merged according to the forest stand patch to which it belongs; The patch volume of each forest stand is obtained by summing the estimated volume values ​​of all pixels within the stand.

7. The intelligent forestry resource monitoring method based on satellite remote sensing technology as described in claim 6, characterized in that, The steps for calculating the change in volume of adjacent time phases based on canopy closure layer, and locating abnormal forest areas based on these changes, include: For the patch accumulation in the same canopy layer at adjacent time phases, calculate the relative rate of change of patch accumulation; An anomaly detection threshold is set, and when the rate of change exceeds the anomaly detection threshold corresponding to its canopy closure layer, it is marked as an abnormal forest area.

8. The intelligent forestry resource monitoring method based on satellite remote sensing technology as described in claim 7, characterized in that, The steps for setting anomaly detection thresholds include: Based on the canopy closure stratification weight map, for the historical phases of all forest stand patches in each canopy closure layer, the time series standard deviation of the relative change rate of patch volume in the same canopy closure layer is calculated, and the time series standard deviation is used as the benchmark fluctuation level. Based on the baseline fluctuation level of each canopy closure layer, anomaly detection thresholds are set for each canopy closure layer, with the anomaly detection thresholds for high canopy closure layers being lower than those for low canopy closure layers.

9. The intelligent forestry resource monitoring method based on satellite remote sensing technology as described in claim 8, characterized in that, The steps for determining high-canopy-density and low-canopy-density layers include: The shading effect weights of all canopy closure layers are sorted, and the mean of the shading effect weights of all canopy closure layers is used as the dividing benchmark. A canopy layer with a shading effect weight higher than the boundary benchmark is determined to be a high canopy layer; a canopy layer with a shading effect weight lower than the boundary benchmark is determined to be a low canopy layer.

10. A forestry resource intelligent monitoring system based on satellite remote sensing technology, employing the forestry resource intelligent monitoring method based on satellite remote sensing technology as described in any one of claims 1 to 9, characterized in that, include: The weighted map framework module, based on multi-temporal satellite remote sensing images, divides forest area pixels into layers according to canopy closure gradient, and constructs shading effect weights for each canopy closure layer to obtain a canopy closure layer weighted map. The correction module uses the canopy closure layer weight map as a constraint to correct the pixel radiation values ​​of each canopy closure layer, thus obtaining a radiation-corrected image. The feature extraction module is used to extract the spectral features of the radiometrically corrected image, and to perform weighted processing on the spectral features using a canopy closure layer weight map to obtain a canopy closure perception feature set. The output module is used to perform volume regression inference on the canopy closure perception feature set, calculate the change in volume of adjacent canopy closure layers, locate abnormal forest areas based on each change, and output a forestry resource supervision report.

Images