CN-121661503-B - Landslide extraction method based on multi-source remote sensing data fusion

CN121661503BCN 121661503 BCN121661503 BCN 121661503BCN-121661503-B

Abstract

The invention relates to the technical field of geological disaster monitoring and remote sensing image processing, and particularly discloses a landslide extraction method based on multi-source remote sensing data fusion. The method comprises the steps of constructing a multidimensional characteristic system comprising vegetation attenuation factors, topography factors and SAR dynamic factors by unifying resolution and projection of optical remote sensing images, synthetic aperture radar data and digital elevation model data, dynamically distributing factor weights by adopting an entropy weight method and combining a physical correction and variation coefficient method, constructing landslide risk indexes, designing a self-adaptive threshold model based on data distribution characteristics, dynamically optimizing and dividing threshold values by means of data distribution characteristics, spatial context characteristics and threshold reliability, and finally realizing accurate extraction of landslide areas. The method solves the problems of low precision, poor adaptability and the like caused by single data source, fixed threshold value and unreasonable weight distribution in the traditional landslide extraction method, improves the accuracy and the robustness of landslide extraction under complex terrain conditions, and provides reliable technical support for post-earthquake disaster assessment and emergency response.

Inventors

ZHAO ZHIHONG
Chen Daiyang
TONG AIHUA
MA LINGYU
ZHANG RAN

Assignees

石家庄铁道大学

Dates

Publication Date: 20260512
Application Date: 20251209

Claims (3)

1. A landslide extraction method based on multi-source remote sensing data fusion is characterized by comprising the following steps: s1, acquiring optical remote sensing images, synthetic aperture radar data and digital elevation model data of a target area before and after landslide, and carrying out standardized processing of uniform resolution and uniform projection on all the data to ensure spatial alignment; S2, calculating a normalized vegetation index variation delta NDVI based on the preprocessed optical remote sensing image, wherein delta NDVI is the difference value between the first time phase normalized vegetation index and the second time phase normalized vegetation index, and the difference value is preserved and normalized to obtain a vegetation attenuation factor; calculating SAR dynamic factors based on the preprocessed synthetic aperture radar data, wherein the SAR dynamic factors comprise SAR variation indexes and SAR spatial consistence, the SAR variation indexes are weighted by polarization logarithmic ratios, polarization ratio variation and texture variation, and the SAR spatial consistence is calculated through standard deviation of SAR variation indexes in a 5 multiplied by 5 window, and the SAR dynamic factors are respectively normalized; S3, calculating initial weights of gradient, curvature and roughness by adopting an entropy weight method, and improving the gradient weight ratio by physical correction to obtain final topography factor weights; calculating the weight of the SAR dynamic factor based on the variation coefficient, wherein the weight of SAR spatial consistency is the ratio of the variation coefficient to the sum of the variation coefficient of SAR variation indexes, and the weight of SAR variation indexes is the difference value between 1 and the ratio; S4, calculating static terrain risk according to the terrain factors and the weights thereof, calculating SAR dynamic risk according to the SAR dynamic factors and the weights thereof, and constructing a landslide risk index LRI by combining vegetation attenuation factors; S5, through weighting fusion of an OTSU threshold, a percentile threshold, a dynamic proportion threshold and a space adjustment factor, adding range constraint to ensure rationality calculation of a self-adaptive threshold based on data distribution characteristics, calculating skewness, kurtosis, data density characteristics and space up-down Wen Quanchong of LRI indexes of all pixel points of a target area, respectively adopting an OTSU algorithm to obtain a basic threshold Obtaining a basic threshold value by adopting a percentile method Obtaining basic threshold by LRI maximum dynamic proportion method The method comprises the steps of calculating the extraction proportion of each basic threshold value and the data density near the threshold value to obtain a threshold value reliability score, dynamically distributing weights of three types of basic threshold values and space context adjustment factors based on the data distribution characteristics and the threshold value reliability score, obtaining an adaptive threshold value based on the data distribution characteristics by weighting and fusing the basic threshold values and the space context adjustment factors and combining range constraint ; S6, performing binarization processing on the LRI index based on the threshold value in S5, removing small-area noise and isolated points in the binarization result through morphological opening operation, and filling small holes and connection fracture areas in the result after the opening operation processing through morphological closing operation to obtain a final landslide extraction result.
2. The landslide extraction method based on multi-source remote sensing data fusion according to claim 1, wherein in S4, the calculation method of LRI index comprises the following formula: Wherein, the Representing a static terrain risk factor, Representing vegetation attenuation factors, namely normalized vegetation index variation, Representing SAR dynamic risk factors; calculating static terrain risk factors The formula is as follows: Wherein, the 、、 The final topographic factor weights of the gradient, the curvature and the roughness obtained in the step S3 are respectively expressed, Indicating the gradient of the slope after the normalization, The curvature after normalization is indicated and the curvature after normalization, Representing the normalized terrain roughness; calculating SAR dynamic risk factors The formula is as follows: Wherein, the Represents the SAR spatial consistency weights obtained in step S3, Indicating the SAR variation index weight obtained in step S3, Representing the normalized SAR spatial consistency, Indicating the normalized SAR variation index.
3. The landslide extraction method based on multi-source remote sensing data fusion according to claim 1, wherein in the step S5, the adaptive threshold value calculation method based on the data distribution characteristics is implemented by weighting and fusing an OTSU threshold value, a percentile threshold value, a dynamic proportion threshold value and a space adjustment factor, and adding a range constraint to ensure rationality, wherein the formula is as follows: Wherein, the Representing an adaptive threshold based on the data distribution characteristics, Representing the base threshold value obtained by the OTSU algorithm, A preset percentile base threshold representing LRI valid data, Represents the LRI maximum dynamic ratio threshold, The spatial adjustment factor is represented as such, Representing the normalized OTSU threshold weights, Representing the normalized percentile threshold weights, Representing the normalized dynamic scaling threshold weights, Representing the normalized spatial context adjustment factor threshold weight; Firstly, the distribution characteristics and the spatial heterogeneity of LRI effective data are quantified by constructing innovative indexes, a basis is provided for threshold optimization, wherein the data density index breaks through the traditional evaluation mode depending on the full-range data, the extreme value interference is reduced by the ratio of the quartile range to the data range, the data aggregation condition is reflected more accurately, and the formula is as follows: Wherein, the Representing data density, a larger value indicates that the LRI valid data distribution is more concentrated, Represents the 75 percentile of LRI valid data, Represents the 25 percentile of LRI valid data, Represents the maximum value of the LRI valid data, Representing the minimum value of LRI valid data, Representing a preset small positive number; Calculating spatial context weight, wherein the weight breaks through the limitation that the traditional threshold value only depends on numerical distribution, combines the discrete degree of LRI effective data with the overall horizontal quantization spatial heterogeneity, reflects the influence of landslide spatial distribution difference on the threshold value, and has the formula: Wherein, the The spatial context weights are represented and, Representing the LRI active data set after the invalid value is removed, Representing the standard deviation calculation function, Representing a mean value calculation function; is a preset maximum value, is used for limiting the weight range, Is a preset small positive number; calculating an LRI maximum value dynamic proportion threshold value, wherein the threshold value breaks through the traditional fixed proportion design, dynamically adjusts the proportion coefficient through the kurtosis of LRI effective data, and adapts to different data distribution characteristics, and the formula is as follows: Wherein, the Represents the LRI maximum dynamic ratio threshold, Represents the maximum value of the LRI valid data, Representing the dynamic scaling factor of the reference, Representing a kurtosis adjustment coefficient; Kurtosis values representing LRI-valid data, for quantifying data distribution sharpness, Expressing a kurtosis constraint coefficient, and avoiding the abnormal proportionality coefficient caused by overlarge kurtosis; And then calculating the reliability score of a single threshold, wherein the score integrates the 'extraction proportion rationality' and the 'data distribution uniformity near the threshold', avoids the limitation of single index evaluation, provides a basis for dynamic weight distribution, and has the formula: Wherein, the A reliability score representing a single threshold, a larger value indicating a stronger threshold applicability, Representing the weight coefficient of the extraction scale score, Represents the data density score weight coefficient around the threshold value, and , Represents the extraction proportion score, evaluates the rationality of the landslide extraction area, Representing a data density score near a threshold, evaluating data distribution uniformity; Then calculating the OTSU threshold weight, wherein the weight breaks through the traditional equal weight design, and combines the reliability score of the OTSU threshold with the skewness dynamic adjustment duty ratio of the LRI effective data to indicate that the data is more asymmetric when the skewness is larger, and the OTSU threshold weight needs to be reduced to avoid the limitation, and the formula is as follows: Wherein, the Representing the threshold weight of the OTSU, A reliability score representing the OTSU threshold, Representing the total reliability score as the sum of the reliability score and the spatial context weight of the OTSU threshold, percentile threshold, dynamic scale threshold, A bias value representing LRI valid data, As a lower limit of the weight of the vehicle, For the OTSU threshold weight upper bound, Controlling the influence amplitude of the skewness on the weight for the kurtosis constraint coefficient; Meanwhile, calculating a percentile threshold weight, wherein the weight is combined with the reliability score of the percentile threshold and the kurtosis dynamic adjustment duty ratio of LRI effective data to adapt to the difference of the degree in the data set, and the formula is as follows: Wherein, the Representing the percentile threshold weight, A reliability score representing the percentile threshold, As the percentile threshold weight upper limit, Adapting the tolerance of the percentile threshold to kurtosis for the kurtosis constraint coefficient, 、、 Meaning is consistent with an OTSU threshold weight formula; Calculating a dynamic proportion threshold weight, wherein the formula is as follows: Wherein, the Representing the dynamic scaling threshold weight, A reliability score representing a dynamic scale threshold; the spatial context adjustment factor weight is calculated, and the formula is as follows: Wherein, the Representing spatial context adjustment factor weights; And normalizing the weights, wherein the formula is as follows: Wherein, the Representing the normalized OTSU threshold weights, Representing the normalized percentile threshold weights, Representing the normalized dynamic scaling threshold weights, Representing the normalized spatial context adjustment factor threshold weight; And calculating a space adjustment factor, wherein the factor is used for correcting a dynamic proportion threshold value, combining the variation degree and the mean square of LRI effective data, and improving the robustness of the threshold value to data fluctuation, and the formula is as follows: Wherein, the The spatial adjustment factor is represented as such, Represents the LRI maximum dynamic ratio threshold, The reference coefficients are adjusted for the space and, Representing the variance-calculating function, The square of the mean value is represented, For the spatial adjustment of the coefficients, Is a preset small positive number.

Description

Landslide extraction method based on multi-source remote sensing data fusion Technical Field The invention relates to the technical field of remote sensing image processing, in particular to a landslide extraction method based on multi-source remote sensing data fusion. Background With the rapid development of remote sensing technology and geographic information science, landslide hazard monitoring has gradually evolved from traditional ground investigation to space-to-ground integrated observation. The acquisition capability of multi-source data such as an optical satellite, a Synthetic Aperture Radar (SAR), a Digital Elevation Model (DEM) and the like is continuously improved, and rich surface information is provided for landslide monitoring. The optical data can capture vegetation coverage change, SAR data can penetrate through cloud and fog to obtain surface structure dynamics, and the DEM supports extraction of terrain factors such as gradient and curvature. Meanwhile, the space data processing technology is mature, so that space-time alignment, feature fusion and quantitative analysis of multi-source data are possible, and a technical foundation is laid for early identification and dynamic evaluation of landslide hidden danger. The traditional landslide identification method still has a remarkable bottleneck, the existing method is mostly dependent on single type data, for example, vegetation change is monitored only by utilizing an optical image, or terrain features are analyzed only based on a DEM, or ground surface deformation is detected only by adopting SAR data, and the complex process that landslide is driven by multi-factor coupling of terrain, vegetation and ground surface physical characteristics is ignored, so that potential landslide is incompletely identified and has insufficient interpretation. When the landslide area is extracted, the traditional method generally adopts a fixed threshold value, an empirical threshold value or a single automatic threshold value, and lacks of targeted analysis on the landslide index data distribution of a specific area. The cutting strategy of one cut is difficult to cope with the index value distribution difference caused by different geographic environments and landslide scales, so that a great deal of misjudgment or missed judgment of the extraction result appears in space. Aiming at the problems, a landslide extraction method based on multi-source remote sensing data fusion is provided. The method comprises the steps of constructing a multidimensional feature set containing vegetation attenuation factors, topography factors and SAR dynamic factors by comprehensively utilizing optical remote sensing data, SAR data and DEM data, further quantifying weights of the factors by adopting an entropy weight method and a variation coefficient, constructing a Landslide Risk Index (LRI), carrying out binarization processing on the landslide sensitivity index (LRI) by combining spatial context analysis and a self-adaptive threshold based on data distribution characteristics, realizing accurate extraction of a landslide region with multi-factor cooperation, and effectively solving the problems of single data source, poor threshold adaptability and the like in the traditional method. Disclosure of Invention 1.A landslide extraction method based on multi-source remote sensing data fusion is characterized by comprising the following steps: s1, acquiring optical remote sensing images, synthetic aperture radar data and digital elevation model data of a target area before and after landslide, and carrying out standardized processing of uniform resolution and uniform projection on all the data to ensure spatial alignment; S2, calculating a normalized vegetation index variation delta NDVI based on the preprocessed optical remote sensing image, wherein delta NDVI is the difference value between the first time phase normalized vegetation index and the second time phase normalized vegetation index, and the difference value is preserved and normalized to obtain a vegetation attenuation factor; calculating SAR dynamic factors based on the preprocessed synthetic aperture radar data, wherein the SAR dynamic factors comprise SAR variation indexes and SAR spatial consistence, the SAR variation indexes are weighted by polarization logarithmic ratios, polarization ratio variation and texture variation, and the SAR spatial consistence is calculated through standard deviation of SAR variation indexes in a 5 multiplied by 5 window, and the SAR dynamic factors are respectively normalized; S3, calculating initial weights of gradient, curvature and roughness by adopting an entropy weight method, and improving the gradient weight ratio by physical correction to obtain final topography factor weights; calculating the weight of the SAR dynamic factor based on the variation coefficient, wherein the weight of SAR spatial consistency is the ratio of the variation coefficient to the sum of the va