Hyperspectral remote sensing identification method for soft rock stratum belt in railway engineering geological survey

Abstract

The present application discloses a kind of soft rock zone high spectral remote sensing identification method for railway engineering geological survey, comprising: obtaining railway survey area hyperspectral image and constructing data set;Subsample set is constructed by random sampling, and the space division of multiple groups of isolated hyperspheres and residual space is generated based on the distance of pixel and its nearest neighbor;According to the position of pixel falling into each space division, generate isolated kernel feature vector;With all or part of pixel as background pixel set, calculate the average value of its isolated kernel feature vector as isolated distribution kernel mean model;The inner product of the isolated kernel feature vector of the pixel to be measured and the mean model is calculated to obtain the initial abnormal similarity, and the contiguous area is extracted as soft rock zone after spatial neighborhood regularization.The present application reduces the computational complexity from quadratic to linear, has data density adaptive ability, and can efficiently and accurately identify soft rock zone under complex geological background along railway by fusing spatial continuity constraint.

Description

Technical Field

[0001] This invention relates to the field of railway engineering geological inversion technology, and in particular to a hyperspectral remote sensing method for identifying weak rock strata in railway engineering geological exploration. Background Technology

[0002] In railway engineering geological surveys, due to the long length of railway lines and the complex geological structures they cross, the identification of adverse geological bodies along the route (such as weak rock strata and concealed faults) is crucial for route selection and construction safety. Traditional manual ground surveys are inefficient and have blind spots. In recent years, geological remote sensing technology, especially hyperspectral remote sensing, has shown great potential in railway engineering geological surveys because it can simultaneously acquire continuous spectral characteristics and spatial texture information of surface targets. From a data processing perspective, weak rock strata occur infrequently against a large background of normal rock, and their spectral absorption characteristics, caused by factors such as mineral composition, differ from those of normal surrounding rocks. Therefore, remote sensing identification of weak rock strata essentially falls under the category of hyperspectral anomaly detection (HAD). Existing hyperspectral anomaly detection algorithms mainly include statistical model-based methods (such as RX detectors), representation-based methods (such as low-rank and sparse representations), and kernel-based methods. Kernel Mean Embedding (KME), as a cutting-edge technology for hyperspectral remote sensing geological anomaly detection, is no longer limited to the traditional method of comparing the similarity between "single pixels". It can map the messy geological background data group to the Reproducing Kernel Hilbert Space (RKHS) and extract it as an overall distribution measure feature, thereby measuring the degree of deviation between the pixel to be measured and the overall distribution of the geological background. This technology has shown great application potential in dealing with hyperspectral anomaly detection in complex backgrounds.

[0003] Defects and shortcomings of existing technology: Currently, directly applying existing kernel mean embedding (KME) methods and other conventional kernel anomaly detection algorithms to hyperspectral remote sensing identification in railway engineering geology has the following shortcomings: 1. The computational complexity is quadratic, making it difficult to meet the data processing needs of large-scale railway line survey areas: Existing kernel methods (such as Gaussian radial basis kernels) have infinite-dimensional feature maps, and the algorithms cannot explicitly extract feature vectors, but can only rely on constructing a kernel matrix for computation. This results in both time and space complexity reaching [insert value here]. ( (This refers to the total number of pixels in the image). Hyperspectral images of railway corridors have a large number of pixels (often tens of millions), and the computational efficiency of existing kernel matrix calculation methods is insufficient to meet engineering requirements.

[0004] 2. Kernel functions are independent of data distribution and have poor adaptability to local density variations in complex geological features: Existing kernels such as the Gaussian kernel are data-independent kernel functions, with their kernel width parameter being globally fixed. However, in the real geological environment along railway lines, due to the intermingling of vegetation cover, exposed rocks, Quaternary soils, etc., the pixel density in the image feature space varies greatly. Kernel functions with fixed parameters cannot reflect the objective law that "the similarity between two points in a sparse area should be higher than that between two points at the same distance in a dense area." This results in the weak spectral features of soft rock layers being easily submerged by complex geological background noise, leading to high false alarm and missed alarm rates.

[0005] 3. Lack of spatial structural constraints and underutilization of the "banded / contiguous" characteristics of geological strata: Existing nuclear anomaly detection typically treats pixels as independent samples. However, in real geological environments, weak rock strata are usually distributed continuously in "banded" or "planar" patterns. Ignoring the spatial neighborhood correlation of images easily leads to a large number of isolated noise points (shot noise) in the identification results, resulting in poor usability of the identification results.

[0006] In view of the shortcomings of the prior art, the technical problem to be solved by the present invention is to provide a hyperspectral remote sensing identification method for weak rock strata that has low computational complexity, high adaptability to complex geological background data density, and integrates spatial continuity characteristics. Summary of the Invention

[0007] Therefore, the purpose of this invention is to provide a hyperspectral remote sensing identification method for weak rock strata in railway engineering geological exploration. This method replaces the infinite-dimensional, highly complex feature mapping in traditional kernel methods with a finite-dimensional binary "isolated kernel" feature mapping based on local data density. It transforms the comparison between hyperspectral pixels and the global background distribution into an efficient vector inner product operation, thereby calculating the initial anomaly similarity of all pixels. Based on this anomaly similarity, the continuous distribution characteristics of weak rock strata are identified, enabling efficient and accurate extraction of continuously distributed weak rock strata from hyperspectral images of railway exploration areas, thus meeting the practical application needs of railway engineering geological exploration.

[0008] To achieve the above objectives, this invention provides a hyperspectral remote sensing method for identifying weak rock strata in railway engineering geological exploration, comprising the following steps: S1: Acquire hyperspectral remote sensing images of the railway survey area and preprocess them to form a hyperspectral dataset; S2: Randomly sample without replacement from the hyperspectral dataset to construct a subsample set, and construct an isolated hypersphere based on the nearest neighbor distance of each pixel in the subsample set. Each isolated hypersphere and the external residual space constitute a set of spatial partitions. Repeat the sampling and isolated hypersphere construction process multiple times to generate a set of isolated hyperspheres composed of multiple sets of spatial partitions. S3: For any pixel in the hyperspectral dataset, based on the position of the arbitrary pixel falling into the interior or exterior space of the isolated hypersphere in each group of spatial divisions, establish the isolated kernel feature mapping vector of the pixel; S4: Take all pixels in the hyperspectral dataset except the pixel to be measured as background pixels, calculate the average value of the isolated kernel feature mapping vector of all background pixels, and use it as the isolated distribution kernel mean feature of the global geological background. S5: Calculate the inner product of the isolated kernel feature mapping vector of the pixel to be tested and the isolated distribution kernel mean feature of the global geological background to obtain the initial anomaly similarity. The smaller the initial anomaly similarity, the higher the probability that the pixel to be tested belongs to a weak rock layer. S6: Based on the initial anomaly similarity of all pixels, identify and extract the weak rock strata zone in the railway survey area.

[0009] A further preferred method involves generating a set of isolated hyperspheres composed of multiple spatial partitions, specifically including the following steps: S201, From hyperspectral dataset Randomly selected Each pixel constructs a subsample set ; S202, For the subsample set The i-th pixel Calculate the pixels separately and The rest The Euclidean distance of each pixel is used to find the nearest neighbor pixel. S203, using pixels Center of the ball, with nearest neighbor distance Define an isolated hypersphere with radius . ; S204. Repeat the hypersphere construction process described in S201-S203 above. Next, generated by A set of isolated hyperspheres formed by grouping spatial partitions.

[0010] More preferably, in S3, based on the position of any pixel falling into the isolated hypersphere or external space in each group of spatial divisions, an isolated kernel feature mapping vector for that arbitrary pixel is established; including: Set the isolated kernel feature mapping vector The dimension is t×(w+1), where t is the number of spatial divisions and w is the number of pixels in the subset. Set indicator vector , representing the position where any pixel x falls. For the k-th spatial partition, if any pixel x falls into the j-th isolated hypersphere, then the indicator vector of that pixel is... The j-th bit is 1, and the rest are 0; If pixel x does not fall into any isolated hypersphere, then the indicator vector The (w+1)th bit is 1; The isolated kernel feature mapping vector Indicator vectors partitioned by all t groups of space It was pieced together.

[0011] Further preferably, in S4, it specifically includes: based on the obtained isolated kernel feature mapping vector The global geological background distribution of the area to be measured Mapped to a finite-dimensional kernel space, the feature The following formula is used to calculate it: ; in, The set of pixels that participate in the background distribution estimation.

[0012] More preferably, in S5, the initial anomaly similarity is calculated using the following formula: ; in, Represents the isolated kernel feature mapping vector of the pixel to be measured x. The inner product of the isolated distribution kernel mean characteristics of the global geological background; This indicates the initial anomaly similarity.

[0013] More preferably, in step S5, after calculating the initial anomaly similarity, spatial neighborhood regularization is introduced to calculate the similarity of each pixel. Final anomaly score: ; in, For pixels Local space window, and Representing pixels and pixels Spatial coordinates in the image, These are spatial scale control parameters. This is the weighting coefficient for the empty spectrum.

[0014] Further preferably, the spatial scale control parameters The range of values ​​is Where W is the local spatial window The number of pixels per side.

[0015] More preferably, the range of the spatial spectral weighting coefficient α is [0,1]. When the image signal-to-noise ratio is high and the rock layer boundary is clear, the value is 0.6≤α≤0.8. When the image noise is large or the surface is broken, the value is 0.3≤α≤0.5.

[0016] Furthermore, in S6, based on the initial anomaly similarity of all pixels, the weak rock strata zone of the railway survey area is identified, including: the final anomaly score of all pixels. The data is sorted in ascending order, segmented according to a preset threshold, and pixels with anomaly scores lower than the preset threshold are extracted. The extracted pixels are then merged, and the contiguous polygonal regions formed by the merged pixels are used as the final identified railway weak rock strata.

[0017] The hyperspectral remote sensing method for identifying weak rock strata in railway engineering geological exploration disclosed in this application has at least the following advantages compared to existing technologies: This invention constructs an explicit binary feature map based on hypersphere partitioning, transforming the infinite-dimensional feature transformation of traditional kernel methods into finite-dimensional sparse vector operations, which significantly reduces time and space complexity, while also reducing the demand for computing resources and memory, making it feasible for large-scale hyperspectral data processing in railway engineering.

[0018] The radius of the isolated hypersphere constructed in this invention is dynamically determined by the nearest neighbor distance in the local data. In dense background areas such as vegetation and Quaternary cover, the hypersphere radius is small and the subdivision is fine, while in sparse areas of anomalies, the radius is large and the subdivision is coarse. This data-dependent characteristic can automatically amplify the difference between weak anomaly signals and the background, improving the identification accuracy of weak rock strata in complex geological environments.

[0019] This invention does not require the prior acquisition of pure surrounding rock samples. It can characterize the global background distribution through random subsampling and kernel mean embedding. Even if a small number of abnormal pixels are mixed in the background samples, it will not cause serious deviation of the background mean model. It has high robustness to data noise and abnormal pollution and is more in line with real field exploration scenarios.

[0020] This invention introduces Gaussian neighborhood regularization based on spatial coordinate distance in the anomaly scoring stage, making full use of the geological characteristics of weak rock layers being distributed in bands or continuous areas, effectively suppressing false alarms caused by isolated noise points, and making the extraction results into continuous polygonal regions, which are convenient for engineering geologists to use directly. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the hyperspectral remote sensing identification method for weak rock strata in railway engineering geological exploration provided by the present invention. Detailed Implementation

[0022] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0023] like Figure 1 As shown, one embodiment of the present invention provides a hyperspectral remote sensing identification method for weak rock strata zones in railway engineering geological exploration, comprising the following steps: S1: Acquire hyperspectral remote sensing images of the railway survey area and form a hyperspectral dataset after preprocessing; the hyperspectral remote sensing images of the railway corridor obtained in this application, after preprocessing such as radiometric calibration, atmospheric correction, and band dimensionality reduction, remove bands severely affected by water vapor absorption, retain the effective spectral bands, and use principal component analysis (PCA) to reduce the dimensionality to the first 30 principal component bands. Select a section of the survey area (e.g., a data block containing 10,000 pixels). ); Constructing a hyperspectral dataset ,in, Total number of pixels (including background surrounding rock and potentially weak rock strata). This represents the number of spectral bands.

[0024] S2: Randomly sample without replacement from the hyperspectral dataset to construct a subset of samples. Construct isolated hyperspheres based on the nearest neighbor distance of each pixel in the subset of samples. Each isolated hypersphere and its external residual space constitute a set of spatial partitions. Repeat the sampling and isolated hypersphere construction process multiple times to generate a set of isolated hyperspheres consisting of multiple sets of spatial partitions. Set the number of sets of isolated hyperspheres. The subsampling rate parameter is set to 2.56% of the total number of pixels in the entire measurement area (i.e., the number of pixels sampled). Specifically, it includes: S201, From hyperspectral dataset Randomly selected Each pixel constructs a subsample set ; A random sampling method without replacement was used to sample from the hyperspectral dataset. Randomly selected Each pixel constructs a subsample set ,set up ; S202, For the subsample set The i-th pixel Calculate the pixels separately and The rest The Euclidean distance of each pixel is used to find the nearest neighbor pixel. S203, using pixels Center of the ball, with nearest neighbor distance Define an isolated hypersphere with radius . Among them, radius For pixels The minimum Euclidean distance to other pixels is calculated using the following formula: ; From subsample set In 1 pixel, generating a collection of A single spatial division of a hypersphere and its external space; S204. Repeat the hypersphere construction process from steps S201 to S203 above. Next, generated by A set of isolated hyperspheres formed by grouping spatial partitions.

[0025] In practice, 256 pixels are randomly selected from the dataset. For each of these 256 pixels, the Euclidean distance between the selected pixel and the remaining 255 pixels is calculated. The closest pixel is found, and 256 high-dimensional hyperspheres are constructed using this distance as the radius. This spatial partitioning is repeated 1000 times, generating 1000 sets of hypersphere spatial partitioning rules which are stored in memory.

[0026] S3: For any pixel in the hyperspectral dataset, based on the position of the arbitrary pixel falling within or outside the isolated hypersphere in each spatial partition, establish the isolated kernel feature mapping vector of the arbitrary pixel; specifically including: For hyperspectral datasets For any pixel x in the vector, define an explicit finite-dimensional feature map: as an isolated kernel feature map vector: ; The dimension of the isolated kernel feature mapping vector is t×(w+1), where t is the number of spatial divisions and w is the number of pixels in the subsample set. For the Group space partitioning, pixels The landing position is indicated by the indicator vector. Characterization; If pixel Only fell into the first For each hypersphere, the indicator vector is... The One bit is 1, and the rest are 0; if If a region falls into the overlapping region of multiple hyperspheres, it is assigned to the dimension of the hypersphere whose center is closest to it; if If it does not fall into any hypersphere, then the indicator vector will be... The Assign 1 to the bit (external residual space).

[0027] The isolated kernel feature mapping vector Indicator vectors partitioned by all t groups of space It is pieced together, therefore, a pixel The complete feature mapping uses the isolated kernel feature mapping vector as shown in the following formula. express: ; S4: Take all pixels in the hyperspectral dataset as background pixels, calculate the average value of the isolated kernel feature mapping vector of all background pixels, and use it as the isolated distribution kernel mean model of the global geological background. Based on the obtained isolated kernel feature mapping vector The global geological background distribution of the area to be measured Mapped to a finite-dimensional kernel space, the isolated distribution kernel mean characteristics of the global geological background The following formula is used to calculate it: ; in, This is the set of pixels involved in the background distribution estimation. This vector represents the comprehensive spectral distribution measure of the normal surrounding rock geological background in the exploration study area.

[0028] In this embodiment, 10,000 pixels are input one by one into the above partitioning rules. It is then determined whether each pixel falls within the hypersphere. If it does, the corresponding dimension of the pixel's feature vector is assigned 1; otherwise, it is assigned 0. Thus, a hypersphere is generated. A sparse binary vector of dimension as an isolated kernel feature mapping vector By calculating the arithmetic mean of the isolated kernel feature mapping vectors of all pixels across each dimension, the mean vector representing the normal surrounding rock background characteristics of this section is obtained. .

[0029] S5: Calculate the inner product of the isolated kernel feature mapping vector of the pixel to be tested and the mean feature of the isolated distribution kernel of the global geological background to obtain the initial anomaly similarity. The smaller the initial anomaly similarity, the higher the probability that the pixel to be tested belongs to a weak rock layer. The pixel to be tested Isolated kernel feature mapping vector The isolated distribution kernel mean characteristics of the measured global geological background Perform inner product operation to obtain the initial kernel anomaly similarity. : ; The smaller the value, the greater the difference in spectral distribution between the pixel and the normal geological background of the surrounding rock in the survey area, and the lower the probability that it belongs to a normal geological body.

[0030] S6: Based on the initial anomaly similarity of all pixels, identify and extract the weak rock strata zone in the railway survey area.

[0031] Furthermore, considering the spatial continuity of weak rock strata, for pixels... Spatial Gaussian neighborhood regularization is introduced to calculate the final anomaly score. : ; in, For pixels Local space window, and Representing pixels and Spatial coordinates in the image, These are spatial scale control parameters. These are the spatial spectral weighting coefficients. Specifically, they are the spatial scale control parameters. The standard deviation of the Gaussian spatial weighting function controls the rate at which the influence of other pixels in the spatial neighborhood on the central pixel decays with increasing spatial distance. The larger the value, the stronger the smoothing effect and the wider the range of influence. The value range is typically within... (Unit: pixels) between. The value and the set local space window Size (assuming the window side length is) Pixel, usually The correlation is strong (for odd numbers such as 3, 5, 7, etc.) and must be determined jointly based on the ground spatial resolution (GSD) of the hyperspectral sensor and the expected physical width of the target weak rock layer zone. According to the Gaussian distribution principle, to ensure that the weight of edge pixels within the window decays smoothly and does not prematurely approach zero, it is often correlated with the window side length in engineering. Establish a linear proportional relationship, with the following rules for value selection. To illustrate this with a practical engineering example: Assume the ground spatial resolution of the hyperspectral remote sensing data is... The geological report predicts that the width of the weak interlayer / rock strata zone within the survey area is approximately... To ensure that the Gaussian filter window precisely covers the core width of the rock strata, the local spatial window should ideally be [value missing]. Pixel (i.e., physical span) At this point, according to the value selection rules, The optimal range of values ​​is approximately Between (e.g., taking the default value) This value ensures that spatial regularization only smooths the rock formations within the strata, without forcibly smoothing normal surrounding rock into the strata.

[0032] Spatial Spectrum Weighting Coefficient This is used to balance the contribution ratio of "pixel's original spectral anomaly score" and "spatial neighborhood smoothness score" to the final identification result. Value range . The value is dynamically set based primarily on the signal-to-noise ratio (SNR) of the hyperspectral image and the degree of fragmentation (spatial heterogeneity) of the geological body surface. Higher values ​​are preferred for low-noise, well-defined scenes, when the image quality of the exploration area is high (high SNR), vegetation cover is sparse, and rock outcrop boundaries are clear. The value should be relatively large (e.g.) This reduces the weight of neighborhood smoothing, preserving the fine geometric contours of the weak rock layer edges to the greatest extent possible, and preventing "over-smoothing" from causing boundary blurring. In high-noise, fragmented surface scenarios (with smaller values), when the survey area is significantly affected by sensor stripe noise, or when the surface is severely disturbed by sparse vegetation, shadows, gravel, and other fragmented features, this method highlights the macroscopic "band-like" continuity of the weak rock layers. The value should be relatively small (e.g.) At this point, the algorithm focuses more on the consistency of the spatial neighborhood, effectively filtering out a large number of isolated, scattered "pseudo-anomalies." In the absence of prior geological information, it is empirically recommended to take... That is, the weights of the original spectral information and the spatial neighborhood smoothing information are divided in a 6:4 ratio, with spectral information being the dominant factor.

[0033] In this embodiment, the pixel mapping vector is calculated and The initial score is obtained by the dot product. Set up the space window for Pixel neighborhood, spatial spectral weight Spatial scale The final score, which incorporates spatial contiguousness attributes, is obtained by performing filtering and smoothing calculations according to the formula. .

[0034] Furthermore, in S6, the final anomaly score for all pixels is calculated. The images are sorted in ascending order. Pixels with scores below the threshold are extracted using either a preset threshold segmentation method or the Otsu threshold segmentation algorithm. The resulting contiguous polygonal regions are output as the final identified railway weak rock strata zone. In this embodiment, the final scored image is binarized using Otsu threshold segmentation, and morphological opening and closing operations are used to further smooth the boundaries. The output polygonal anomalous patches are plotted in the GIS system and marked as "suspected weak rock strata zone".

[0035] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A hyperspectral remote sensing method for identifying weak rock strata in railway engineering geological exploration, characterized in that, Includes the following steps: S1: Acquire hyperspectral remote sensing images of the railway survey area and preprocess them to form a hyperspectral dataset; S2: Randomly sample without replacement from the hyperspectral dataset to construct a subsample set, and construct an isolated hypersphere based on the nearest neighbor distance of each pixel in the subsample set. Each isolated hypersphere and the external residual space constitute a set of spatial partitions. Repeat the sampling and isolated hypersphere construction process multiple times to generate a set of isolated hyperspheres composed of multiple sets of spatial partitions. S3: For any pixel in the hyperspectral dataset, based on the position of the arbitrary pixel falling into the interior or exterior space of the isolated hypersphere in each group of spatial divisions, establish the isolated kernel feature mapping vector of the arbitrary pixel; S4: Take all pixels in the hyperspectral dataset except the pixel to be measured as background pixels, calculate the average value of the isolated kernel feature mapping vector of all background pixels, and use it as the isolated distribution kernel mean feature of the global geological background. S5: Calculate the inner product of the isolated kernel feature mapping vector of the pixel to be tested and the isolated distribution kernel mean feature of the global geological background to obtain the initial anomaly similarity. The smaller the initial anomaly similarity, the higher the probability that the pixel to be tested belongs to a weak rock layer. S6: Based on the initial anomaly similarity of all pixels, identify and extract the weak rock strata zone in the railway survey area.

2. The hyperspectral remote sensing identification method for weak rock strata zones in railway engineering geological exploration according to claim 1, characterized in that, In S2, the generation of a set of isolated hyperspheres consisting of multiple spatial partitions includes the following steps: S201, From hyperspectral dataset Randomly selected Each pixel constructs a subsample set ; S202, For the subsample set The i-th pixel Calculate the pixels separately and The rest The Euclidean distance of each pixel is used to find the nearest neighbor pixel. S203, using pixels Center of the ball, with nearest neighbor distance Define an isolated hypersphere with radius . ; S204. Repeat the hypersphere construction process described in S201-S203 above. Next, generated by A set of isolated hyperspheres formed by grouping spatial partitions.

3. The hyperspectral remote sensing identification method for weak rock strata zones in railway engineering geological exploration according to claim 1, characterized in that, In S3, based on the position of any pixel falling into the isolated hypersphere or external space in each group of spatial partitions, an isolated kernel feature mapping vector for that pixel is established; including: Set the isolated kernel feature mapping vector The dimension is t×(w+1), where t is the number of spatial divisions and w is the number of pixels in the subset. Set indicator vector , representing the position where any pixel x falls. For the k-th spatial partition, if any pixel x falls into the j-th isolated hypersphere, then the indicator vector of that pixel is... The j-th bit is 1, and the rest are 0; If pixel x does not fall into any isolated hypersphere, then the indicator vector The (w+1)th bit is 1; The isolated kernel feature mapping vector Indicator vectors partitioned by all t groups of space It was pieced together.

4. The hyperspectral remote sensing identification method for weak rock strata zones in railway engineering geological exploration according to claim 3, characterized in that, In S4, specifically it includes: based on the obtained isolated kernel feature mapping vector The global geological background distribution of the area to be measured Mapped to a finite-dimensional kernel space, the isolated distribution kernel mean characteristics of the global geological background The following formula is used to calculate it: ； in, The set of pixels that participate in the background distribution estimation.

5. The hyperspectral remote sensing identification method for weak rock strata zones in railway engineering geological exploration according to claim 1, characterized in that, In S5, the initial anomaly similarity is calculated using the following formula: ； in, Represents the isolated kernel feature mapping vector of the pixel to be measured x. The inner product of the isolated distribution kernel mean characteristics of the global geological background; This indicates the initial anomaly similarity.

6. The hyperspectral remote sensing identification method for weak rock strata zones in railway engineering geological exploration according to claim 1, characterized in that, In step S5, after calculating the initial anomaly similarity, spatial neighborhood regularization is introduced to calculate the similarity of each pixel. Final anomaly score: ； in, For pixels Local space window, and Representing pixels and pixels Spatial coordinates in the image, These are spatial scale control parameters. This is the weighting coefficient for the empty spectrum.

7. The hyperspectral remote sensing identification method for weak rock strata zones in railway engineering geological exploration according to claim 6, characterized in that, The spatial scale control parameters The range of values ​​is Where W is the local spatial window The number of pixels per side.

8. The hyperspectral remote sensing identification method for weak rock strata zones in railway engineering geological exploration according to claim 6, characterized in that, The value range of the spatial spectral weighting coefficient α is [0,1]. When the image signal-to-noise ratio is high and the rock layer boundary is clear, the value is 0.6≤α≤0.

8. When the image noise is large or the surface is broken, the value is 0.3≤α≤0.

5.

9. The hyperspectral remote sensing identification method for weak rock strata zones in railway engineering geological exploration according to claim 1, characterized in that, In S6, based on the initial anomaly similarity of all pixels, the weak rock strata zone of the railway survey area is identified and extracted, including: the final anomaly score of all pixels. The data is sorted in ascending order, segmented according to a preset threshold, and pixels with anomaly scores lower than the preset threshold are extracted. The extracted pixels are then merged, and the contiguous polygonal regions formed by the merged pixels are used as the final identified railway weak rock strata.

Images