CN-122024078-A - Multi-source remote sensing fusion forest and grass wet resource dynamic identification and classification method and system

CN122024078ACN 122024078 ACN122024078 ACN 122024078ACN-122024078-A

Abstract

The invention discloses a method and a system for dynamically identifying and classifying forest and grass wet resources by multi-source remote sensing fusion, and relates to the technical field of resource identification and classification. The method comprises the steps of S1, collecting multi-source remote sensing observation data, carrying out data preprocessing and observation quality assessment, S2, carrying out object segmentation and feature aggregation to form an object-level three-dimensional feature set, S3, obtaining a cross-time object-level three-dimensional feature set, extracting pseudo-change driving features, carrying out pseudo-change assessment, carrying out true change consistency judgment, carrying out change type distinction by combining the pseudo-change assessment and the true change consistency judgment result, S4, carrying out supervision training and grading classification based on the object-level three-dimensional feature set, outputting forest wet type probability, and carrying out rule recognition by combining the change features. The problems of false change and true ground class transfer confusion, high false alarm rate and insufficient result stability caused by the fact that unreal fluctuation is difficult to distinguish in multi-source remote sensing cross-time comparison are solved.

Inventors

WANG ZIPING
Cheng sansheng
ZHENG YIMIN
LIU RUJIAN
HU SEN
FAN XIFENG
Du Huike
CHANG RUOSHI
WANG JINGLI

Assignees

天津市地质工程勘测设计院有限公司

Dates

Publication Date: 20260512
Application Date: 20260414

Claims (10)

1. The method for dynamically identifying and classifying the forest and grass wet resources by multi-source remote sensing fusion is characterized by comprising the following steps of: s1, acquiring multi-source remote sensing observation data, and carrying out data preprocessing on the multi-source remote sensing observation data, wherein the multi-source remote sensing observation data after preprocessing is used for executing observation quality assessment, and standardized observation data under reliability constraint is constructed; S2, carrying out object segmentation and feature aggregation based on standardized observation data under credibility constraint to form an object-level three-dimensional feature set fusing spectra, structures and textures; S3, constructing a cross-time object corresponding relation, utilizing the driving characteristics of the cross-time object level three-dimensional characteristic set to extract the characteristics of the weather, the water level, the geometry and the quality, and utilizing the driving characteristics to perform pseudo-change evaluation; And S4, based on the object-level three-dimensional feature set, taking a stability distinguishing result of the change type as a data screening basis, performing supervision training and classification, outputting classification probabilities of the woodland, the grassland and the wetland, and performing rule recognition by combining the change features to form a land type updating and change type labeling result.
2. The method for dynamically identifying and classifying the forest and grass wet resources by multi-source remote sensing fusion according to claim 1, wherein the specific processes of collecting multi-source remote sensing observation data and preprocessing the multi-source remote sensing observation data are as follows: acquiring multi-source remote sensing observation data in real time, wherein the multi-source remote sensing observation data comprises optical multi-spectral reflectivity, SAR backward scattering coefficient, point cloud height, solar altitude angle, observation zenith angle, cloud probability and shadow probability; The method comprises the steps of reading time stamps and coordinate reference information of multi-source remote sensing observation data, unifying the multi-source remote sensing observation data to the same geographic coordinate system and the same space grid through coordinate projection conversion and space resampling, resampling and pixel alignment the multi-source remote sensing observation data according to target resolution by adopting an interpolation resampling and grid alignment method, performing cross-source time phase registration based on a feature matching registration method to obtain pixel level registration residual fields, performing radiation consistency check and outlier rejection on optical multi-spectral reflectivity, performing radiation calibration and speckle noise filtering on SAR backward scattering coefficients, separating ground points from non-ground points through point cloud classification, unifying height references by adopting an elevation reference conversion method, reading solar altitude parameters and observation zenith angle parameters, performing gridding expansion according to the space range and resolution of corresponding images, enabling the solar altitude parameters and the observation zenith angles to correspond to image pixel spaces one by one, generating observation geometric auxiliary grids, performing dimensionless normalization processing on the multi-source remote sensing observation data, establishing a resource identification classification database, and storing the original and preprocessed multi-source remote sensing observation data.
3. The method for dynamically identifying and classifying forest and grass wet resources by multi-source remote sensing fusion according to claim 2, wherein the specific process of constructing standardized observation data under reliability constraint is as follows: Acquiring an image element level registration residual field, extracting the median and the median absolute deviation of the image element level registration residual, calculating the absolute difference of each image element level registration residual and the median for each grid unit, multiplying the median absolute deviation by an MAD normal consistency correction coefficient, adding an extremely small positive number as a denominator, dividing the absolute difference by the denominator to obtain a robust standardized residual value, and calculating the negative index of the robust standardized residual value to obtain a geometric registration stability item; Subtracting the cloud probability by one, subtracting the shadow probability by one, multiplying the two difference results, and then opening the root number to obtain an imaging quality stable item; multiplying the geometric registration stable item and the imaging quality stable item to obtain a multisource availability credibility value; Calculating the multisource availability credibility value of each grid unit, establishing a spatial index and a time index association with multisource remote sensing observation data, taking the multisource availability credibility value as a weight factor, and carrying out pixel-level weighting on the optical multispectral reflectivity, the SAR backscattering coefficient and the point cloud height which are subjected to corresponding pretreatment to obtain standardized observation data under the credibility constraint.
4. The method for dynamically identifying and classifying forest and grass wet resources by multi-source remote sensing fusion according to claim 3, wherein the specific process for developing object segmentation and feature aggregation based on standardized observation data under credibility constraint to form an object-level three-dimensional feature set of fusion spectrum, structure and texture is as follows: Taking the weighted optical multispectral reflectivity, SAR backscattering coefficient and point cloud height as a segmentation feature input layer, performing super-pixel segmentation on the weighted data by adopting a multiscale segmentation algorithm to generate an object unit set, and performing weighted feature statistics on each object unit in the object unit set; Respectively carrying out weighted average on the optical multispectral reflectivity and the SAR backscattering coefficient of all pixels in the object unit based on the multisource availability credibility value to obtain a spectral vector and SAR backscattering statistic; Sequencing all weighted point cloud heights in the object unit and counting the fractional numbers to obtain an upper bound fractional value of the point cloud height and a central fractional value of the point cloud height; based on the optical multispectral reflectivity, selecting a near infrared band as single-channel gray level input, linearly mapping the corresponding reflectivity into a limited gray level, constructing a gray level co-occurrence matrix in an object unit area, and calculating texture statistics; calculating the area of an object unit region by a grid counting method based on the space contour of the object unit generated by segmentation, calculating the perimeter of the object unit by a boundary pixel identification and boundary length accumulation method, and calculating the shape compactness by a shape compactness calculation model based on the area and the perimeter; And combining the spectrum vector, the SAR backscattering statistic, the point cloud height upper bound index value, the point cloud height center index value, the texture statistic and the shape compactness to construct the object-level three-dimensional feature set.
5. The method for dynamically identifying and classifying the forest and grass wet resources by multi-source remote sensing fusion according to claim 4, wherein the specific process of constructing the cross-time object corresponding relation and extracting the driving characteristics of the weather, water level, geometry and quality by using the cross-time object level three-dimensional characteristic set is as follows: For each object unit in the object unit set, acquiring object-level three-dimensional feature sets of two time phases, and establishing a cross-time-phase object corresponding relationship by adopting an object space superposition matching method; extracting the reflectivity of a near infrared band and the reflectivity of a red light band in a spectrum vector, dividing the difference value of the reflectivity of the near infrared band and the reflectivity of the red light band by the sum of the reflectivity of the near infrared band and the reflectivity of the red light band to obtain a vegetation index of the object unit, performing smooth spline fitting on the vegetation index of the object unit on a continuous time sequence, extracting a time point corresponding to a peak value from a vegetation index curve after fitting, and calculating the difference value of the peak value time points between two time phases to obtain a weather phase difference of the object unit; Extracting the reflectivity of a green light wave band and the reflectivity of a short wave infrared wave band in the spectrum vector, dividing the difference value of the reflectivity of the green light wave band and the reflectivity of the short wave infrared wave band by the sum of the reflectivity of the green light wave band and the reflectivity of the short wave infrared wave band to obtain a water body index, calculating the difference value of the water body indexes of two time phases, and respectively calculating the difference value of SAR backward scattering statistics of an object unit under the two time phases; calculating the difference value of the solar altitude angle and the difference value of the observed zenith angle of the object unit under two time phases, substituting the two difference values into cosine functions respectively, and subtracting the product of the two cosine functions from one to obtain the observed geometric difference of the object unit; and calculating the average value of the multisource availability credibility values in the object unit, and calculating the difference value of the average value of the two time-phase multisource availability credibility values to obtain the quality difference.
6. The method for dynamically identifying and classifying the forest and grass wet resources by multi-source remote sensing fusion according to claim 5, wherein the specific process of performing pseudo-change evaluation by using driving characteristics is as follows: Combining the physical phase difference, the water level wetting difference, the observation geometric difference and the quality difference to form a pseudo-variation driving vector of the object unit; Selecting object units with the absolute difference value of the cloud height upper bound bit dividing value in the object unit set smaller than a change threshold value under two time phases to obtain a reference object set; Calculating covariance among all the components based on pseudo-variation driving vectors of all the object units in the reference object set to form a covariance matrix, performing matrix multiplication operation on the pseudo-variation driving vectors and an inverse matrix of the covariance matrix to obtain intermediate result vectors, performing matrix multiplication operation on the intermediate result vectors and the pseudo-variation driving vectors to obtain scalar values, and opening square roots of the scalar values to obtain pseudo-variation driving intensity values of the object units.
7. The method for dynamically identifying and classifying the forest and grass wet resources by multi-source remote sensing fusion according to claim 6, wherein the specific process of performing true change consistency judgment by using the cross-time object-level three-dimensional feature set is as follows: For each object unit, three kinds of variation amplitude are calculated respectively, namely, euclidean distance between spectrum vectors under two time phases is calculated to obtain spectrum variation amplitude, absolute difference of upper boundary dividing values of point cloud height under the two time phases is calculated to obtain structure height variation amplitude; calculating the difference between each type of variation amplitude and the corresponding median based on the spectrum variation amplitude, the structure height variation amplitude and the scattering variation amplitude of all the object units, respectively calculating the difference between each type of variation amplitude and the corresponding median, multiplying the corresponding median absolute deviation by an MAD normal consistency correction coefficient, adding an extremely small positive number as a denominator, dividing the difference by the denominator to obtain standardized variation values corresponding to three types of variation amplitude; dividing the three natural index operation results by the addition result to obtain a spectrum change duty ratio, a structure change duty ratio and a scattering change duty ratio, multiplying each change duty ratio by a corresponding natural logarithmic value, summing, and obtaining a change evidence consistency entropy by taking the opposite number; and subtracting the ratio of the change evidence consistency entropy to the natural logarithm three from one, and multiplying the ratio by the comprehensive change intensity value to obtain the true change evidence consistency value of the object unit.
8. The method for dynamically identifying and classifying the forest and grass wet resources by multi-source remote sensing fusion according to claim 7, wherein the specific process of combining the pseudo-change evaluation result and the true change consistency judgment result to distinguish the stability of the change types is as follows: Respectively pseudo-variable driving intensity value of each object unit Comparing the true change evidence consistency value with the false change threshold t1 When compared with the consistency threshold t2 < T1 and When not less than t2, determining that the object unit is a real change object Not less than t1 and When < t2, the determination target unit is the pseudo-change target Not less than t1 and When not less than t2, determining the object unit as an uncertain change object < T1 and At < t2, the determination target unit is a stabilization target.
9. The method for dynamically identifying and classifying the forest and grass wet resources by multi-source remote sensing fusion according to claim 8, wherein the specific process for forming land class updating and change type labeling results is as follows by taking a stability distinguishing result of a change type as a data screening basis, performing supervision training and classification, outputting class probabilities of forest lands, grasslands and wet lands, and performing rule recognition by combining the change characteristics: Performing normalization processing on the input features of the supervision training samples, performing model training by using a random forest supervision learning algorithm, and optimizing model parameters through cross verification to construct a hierarchical classification model; Inputting an object-level three-dimensional feature set into a hierarchical classification model aiming at each object unit, wherein a time-phase object-level three-dimensional feature set after the real change object is input and a time-phase object-level three-dimensional feature set before the stable object and the pseudo change object are input are output, and the forest land classification probability, the grassland classification probability and the wetland classification probability of each object unit are output; And aiming at the real change object, reading the spectrum change amplitude, the structure height change amplitude, the scattering change amplitude, the physical phase difference and the water level wetting difference, comparing with corresponding judgment thresholds respectively, and carrying out rule recognition to obtain different change types of the real change object.
10. The system for dynamically identifying and classifying the forest and grass wet resources by multi-source remote sensing fusion is characterized by comprising the following components: the data acquisition processing module is used for acquiring multi-source remote sensing observation data, and carrying out data preprocessing on the multi-source remote sensing observation data; The object feature construction module is used for carrying out object segmentation and feature aggregation based on standardized observation data under credibility constraint to form an object-level three-dimensional feature set fusing spectrum, structure and texture; The pseudo-change judging module is used for constructing a cross-time object corresponding relation, carrying out pseudo-change evaluation by utilizing driving characteristics of the cross-time object level three-dimensional characteristic set extract, the water level, the geometry and the quality, carrying out true change consistency judgment by utilizing the cross-time object level three-dimensional characteristic set, and carrying out stability differentiation of a change type by combining a pseudo-change evaluation result and a true change consistency judgment result; The classification change recognition module is used for performing supervision training and classification based on the object-level three-dimensional feature set by taking the stability distinguishing result of the change type as a data screening basis, outputting the classification probability of the woodland, the grassland and the wetland, and performing rule recognition by combining the change feature to form a land type updating and change type marking result.

Description

Multi-source remote sensing fusion forest and grass wet resource dynamic identification and classification method and system Technical Field The invention relates to the technical field of resource identification and classification, in particular to a method and a system for dynamically identifying and classifying wet forest and grass resources by multi-source remote sensing fusion. Background Along with the rapid development of multi-source remote sensing technologies such as high-resolution optical remote sensing, synthetic aperture radar, laser radar and the like, dynamic monitoring of ecological resources such as woodland, grassland, wetland and the like gradually evolves from a single sensor to a multi-source fusion direction. By fusing multispectral reflection information, backward scattering characteristics and three-dimensional structure information, the earth surface coverage condition and structure attribute can be obtained under a larger space scale and higher time frequency, so that resource type identification, change monitoring and ecological state assessment are realized, and data support is provided for natural resource management, ecological protection restoration, carbon sink accounting and the like. For example, the invention patent with publication number of CN120655997A discloses a mineral resource image classification method based on VGG, which comprises an image preprocessing stage, a model construction stage and a model analysis stage, wherein the remote sensing image is segmented into small blocks according to mine coordinates and classification results and is labeled, then the operations of image up-sampling, sharpening, filtering and the like are carried out, the classification model based on VGG is designed, the tidied image is divided into a training set and a test set, the training set is used, and the model effect is evaluated by using the test set. And in the mineral reserve prediction stage, the trained model is subjected to containerization deployment, predicted pictures are input, and an output result is obtained. The invention uses VGG algorithm, obtains better effect in classifying mineral resource images under real geological data, and has wide application value and application prospect in the geological mineral field. For example, the invention patent with publication number CN120014498A discloses a quick spot inspection analysis method for classification of land resource land patterns, which comprises the following steps of S1, obtaining a land resource land vertical aerial view shot by an aerial camera, S2, carrying out land profile division on the land resource land vertical aerial view in the step S1, S3, carrying out land resource land marking on the land resource land vertical aerial view obtained by the land profile division in the step S2, and S4, classifying the land resource land vertical aerial view marked in the step S3 for quick spot inspection. The invention can rapidly check and identify according to the spot check image, and improve the efficiency and the accuracy. However, in dynamic monitoring of forest grass, it is common practice to register and classify remote sensing images of different sensors or different time phases, and then directly compare the time phase differences to determine the change. However, when the difference of the physical stage, the short-term water level fluctuation, the solar altitude angle and the observation geometry are large, the same ground object unit can generate non-variable fluctuation of spectral reflection and texture or structure characterization, and the fluctuation is confused with a true class transfer or degradation signal in difference comparison, so that 'pseudo change' is difficult to distinguish. Causing the false alarm and missing alarm of the change detection to rise and the comparability of the time phase crossing result to be reduced. Therefore, in order to solve the above problems, a method and a system for dynamically identifying and classifying the forest and grass wet resources by multi-source remote sensing fusion are needed. Disclosure of Invention Technical problem to be solved Aiming at the defects of the prior art, the invention provides a method and a system for dynamically identifying and classifying forest and grass wet resources by multi-source remote sensing fusion, which solve the problems of difficult distinction between unreal spectrum and structure fluctuation caused by physical difference, water level fluctuation and observation geometric change, confusion between pseudo-change and real ground class transfer, high false alarm rate of change detection and poor stability and comparability of results in multi-source remote sensing cross-time comparison. Technical proposal The method for dynamically identifying and classifying the forest and grass wet resources by multi-source remote sensing fusion comprises the following steps of S1, collecting multi-source remote sensing observation dat