Method for identifying hidden dangers of color steel tiles along railway based on guidance between SAR (Synthetic Aperture Radar) polarization channels

Abstract

The invention discloses a method for identifying hidden dangers of color steel tiles along a railway based on guidance between SAR polarization channels, and relates to the technical field of computer vision and remote sensing image processing. The method comprises the steps that firstly, a Sentinel-1 image covering a research area along a railway is obtained, color steel tile and non-color steel tile samples are marked in the research area, and a training set and a test set are divided; then, VV and VH polarization channel images are mapped to a logarithm domain from a linear domain, multi-scale structure components and detail components of VV and VH channels are extracted in combination with multi-scale decomposition and adaptive fusion, and an inter-polarization channel guide structure index and a polarization channel joint guide structure-detail ratio index are constructed; and finally, constructing a feature vector, training a random forest classifier by using the training set to obtain a color steel tile hidden danger recognition model, evaluating model performance through the test set, and realizing automatic detection of a color steel tile hidden danger target on the whole Sentinel-1 image. The method can obviously improve the hidden danger identification precision of the color steel tiles along the railway.

Description

Technical Field

[0001] This invention relates to the fields of computer vision and remote sensing image processing technology, and in particular to a method for identifying potential hazards in corrugated steel roofing along railway lines based on SAR polarization channel guidance. Background Technology

[0002] Corrugated steel roofing sheets are widely distributed in industrial parks, logistics parks, warehousing bases, and urban and rural construction land. Their spatial distribution and expansion pattern are closely related to local industrial layout, safety production management, and illegal construction. Accurate identification and dynamic monitoring of corrugated steel roofing sheets, especially in sensitive areas such as along railway lines, are of great significance for planning management and risk prevention.

[0003] Traditional methods for identifying color steel roof tiles primarily rely on on-site surveys and manual or semi-automatic interpretation based on high-resolution optical images. These methods are labor-intensive, subjective, and have long update cycles, making it difficult to meet the operational demands for "high-frequency updates and automated processing." Furthermore, optical remote sensing is easily affected by cloud cover and lighting conditions, leading to insufficient data timeliness and availability in cloudy, rainy areas or emergency monitoring scenarios.

[0004] Synthetic Aperture Radar (SAR) possesses all-weather, all-day imaging capabilities, unaffected by cloud cover or lighting conditions. Dual-polarization SAR satellites, such as Sentinel-1, provide a stable data foundation for identifying potential hazards in corrugated steel roofing when optical data is lacking or obscured by clouds or fog. However, SAR images suffer from significant speckle noise and a mixed effect from the superposition of multiple scattering mechanisms. Most existing SAR-based methods for corrugated steel roofing detection involve independent filtering or feature construction of polarization channels such as VV and VH, failing to fully utilize the polarization scattering characteristics of corrugated steel roofing. Summary of the Invention

[0005] The technical problem to be solved by this invention is how to provide a method for identifying hidden dangers of corrugated steel roofing along railway lines based on SAR polarization channel guidance, which has high identification accuracy and good identification robustness.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a method for identifying hidden dangers of corrugated steel roofing sheets along railway lines based on SAR polarization channel guidance, characterized by the following steps:

[0007] S1. Acquire Sentinel-1 images covering the study area along the railway line, label the color steel tile and non-color steel tile sample points in the study area, and divide the sample dataset into training set and test set. The Sentinel-1 images include the backscattering coefficients of the VV polarization channel and the VH polarization channel.

[0008] S2, performs logarithmic transformation on the VV and VH polarization channel images in the linear domain, and performs structure-detail decomposition based on polarization channel guidance in the logarithmic domain;

[0009] S3 performs structure-detail decomposition guided by polarization channels at small, medium and large scales respectively, and adopts an adaptive fusion strategy to adaptively adjust the weights of structural components and detail components at each scale according to the local texture intensity, thereby obtaining fused multi-scale structural components and detail components.

[0010] S4, using the fused multi-scale structural components to construct the polarization channel guide structure index CG_SBI;

[0011] S5, using the fused multi-scale structural components and detail components to construct a polarization channel joint guidance structure – detail ratio index JG_SDRI, where the polarization channel joint guidance includes both polarization channel self-guidance and polarization channel inter-guidance;

[0012] S6. The polarization channel guidance structure index CG_SBI, the polarization channel joint guidance structure-detail ratio index JG_SDRI, the backscattering coefficient of the VV polarization channel and the backscattering coefficient of the VH polarization channel are combined into a feature vector. A random forest classifier is trained based on the training set samples to obtain the color steel tile hidden danger identification model.

[0013] S7 uses a color steel tile hazard identification model to predict test set samples, calculates confusion matrix and various accuracy indicators, and identifies color steel tile hazards on the entire Sentinel-1 image.

[0014] The beneficial effects of adopting the above technical solution are as follows: the method enables the structural estimation of the VH polarization channel to benefit from the guidance of the VV polarization channel, making full use of the scattering characteristics of the color steel tile, which exhibits strong backscattering in the VV polarization channel but has a weaker response in the VH polarization channel, and significantly reducing the structural estimation instability caused by the weak signal and heavy speckle noise in the VH polarization channel.

[0015] By using the variance of VV polarization channel images within a local window as a measure of texture intensity, the fusion weights of small-scale, medium-scale, and large-scale structural components and detail components are adaptively adjusted. This enhances fine-scale structures in areas with complex textures and strengthens large-scale structures in flat background areas. While meeting the requirements of speckle suppression and background smoothing, this approach also takes into account and strengthens the preservation of structural details of the corrugated steel tile target, achieving a better overall balance.

[0016] The CG_SBI and JG_SDRI indices were constructed. Based on the guidance between polarization channels, the former highlights the polarization scattering advantage of VV over VH in a smooth and stable structural characterization, improving the distinguishability of corrugated steel tile hazards from background features. The latter, by comprehensively characterizing the differences in structural dominance and detail undulation between the VV and VH polarization channels, ensures that corrugated steel tile hazards maintain high discriminability and robustness of identification even in complex backgrounds. Attached Figure Description

[0017] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0018] Figure 1 This is an overall flowchart of the method described in the embodiments of the present invention;

[0019] Figure 2 This is a location image map of the study area along the railway line in the method described in this embodiment of the invention;

[0020] Figure 3 This is a pseudo-color map of the Sentinel-1 image of the research area along the railway line in the method described in this embodiment of the invention;

[0021] Figure 4 This is a confusion matrix diagram of the color steel tile hazard identification model in the method described in this embodiment of the invention on the test set samples;

[0022] Figure 5 The method described in this embodiment of the invention performs color steel tile detection on the entire Sentinel-1 image and outputs a color steel tile distribution result map. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0024] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0025] Overall, such as Figure 1 As shown in the figure, this invention discloses a method for identifying potential hazards in color steel roofing sheets along railway lines based on SAR polarization channel guidance, comprising the following steps:

[0026] S1. Obtain Sentinel-1 images covering the study area along the railway line from the GEE platform, mark the color steel tile and non-color steel tile sample points in the study area, and divide the sample dataset into training set and test set. The Sentinel-1 images include the backscattering coefficients of the VV polarization channel and the backscattering coefficients of the VH polarization channel.

[0027] S2 performs a logarithmic transformation on the VV and VH polarization channel images in the linear domain, and performs structure-detail decomposition based on polarization channel guidance in the logarithmic domain. This fully utilizes the stronger and more stable backscattering physical properties of the color steel tile in the VV polarization channel compared to the VH polarization channel to obtain a more robust structural characterization and noise suppression.

[0028] S3 performs structure-detail decomposition guided by polarization channels at small, medium and large scales respectively, and adopts an adaptive fusion strategy to adaptively adjust the weights of structural components and detail components at each scale according to the local texture intensity, thereby obtaining fused multi-scale structural components and detail components.

[0029] S4. The cross-guided structure-based index CG_SBI (Cross-Guided Structure-Based Index) is constructed using the fused multi-scale structural components.

[0030] S5. The fused multi-scale structural components and detail components are used to construct the joint-guided structure-detail ratio index JG_SDRI (Joint-Guided Structure-Detail Ratio Index), in which the joint guidance of polarization channels includes both self-guidance of polarization channels and guidance between polarization channels.

[0031] S6. The backscattering coefficients of CG_SBI, JG_SDRI, VV polarization channel and VH polarization channel are used to form a feature vector. A random forest classifier is trained based on the training set samples to obtain the color steel tile hidden danger identification model.

[0032] S7 uses a color steel tile hazard identification model to predict test set samples, calculates confusion matrix and various accuracy indicators, and identifies color steel tile hazards on the entire Sentinel-1 image.

[0033] The above steps will be explained in detail below with reference to specific content:

[0034] Specifically, step S1 includes:

[0035] S1-1) Data Acquisition: Acquire Sentinel-1 GRD imagery covering the study area along the railway line during the summer of 2024 (June 1 to August 31) using the GEE platform, such as... Figure 2 As shown, the study area is a 1 km buffer zone on both sides of some railway lines within Xinhua District, Qiaoxi District, and Luquan District of Shijiazhuang City, Hebei Province; the image selection criteria include: IW imaging mode, VV / VH dual polarization combination, and 10 m spatial resolution; representative synthetic images for this period are generated by pixel-by-pixel median synthesis. Figure 3 The image shown is a pseudo-color map of the study area. The pseudo-color composite scheme is R corresponding to VV, G to VH, and B to VV-VH, with the unit being dB. It is generated after percentile linear stretching. The purple areas in the image correspond to land features with large differences in VV and VH polarization (such as corrugated steel roofing).

[0036] S1-2) Data Labeling and Classification: Different land features in the study area were labeled from optical remote sensing images through manual visual interpretation. The land feature types in the study area were divided into seven main categories: corrugated steel roofing, vegetation, bare land, highways, railways, water bodies, and buildings. 600 representative sample points of corrugated steel roofing were labeled, and 150 sample points of each of the other categories were labeled, for a total of 1500 sample points. The original seven categories were transformed into a binary classification problem, with corrugated steel roofing marked as 0, and all other categories (vegetation, bare land, highways, railways, water bodies, and buildings) uniformly marked as 1 (non-corrugated steel roofing). Finally, a stratified sampling strategy was used to divide the data into training and test sets in a 7:3 ratio.

[0037] Specifically, step S2 includes:

[0038] S2-1) Perform logarithmic transformation on the VV and VH polarization channel images in the linear domain;

[0039] S2-2) In the logarithmic domain at a scale of k, guided filtering is used to estimate three structural components, namely the VV polarization channel self-guided structural components. VH polarization channel self-guided structure component and the guiding structure components between VH polarization channels ;

[0040] (S2-3) Construct a joint weight for polarization channels determined by polarization correlation weights and normalized polarization difference index, which is used to guide the self-guided structural components of VH polarization channels. Guided structural components between VH polarization channels Adaptive adjustment is performed to obtain the joint structural components of the VH polarization channels. ;

[0041] S2-4) The VV polarization channel self-guided structure component VH polarization channel inter-guided structure components and VH polarization channel joint structure components By performing a natural exponential transform, the self-guided structure components of the VV polarization channel in the linear domain are obtained. VH polarization channel inter-guided structure components and VH polarization channel joint structure components And the joint structural components of the VH polarization channels. Apply upper limit constraints;

[0042] S2-5) Calculate the detail components of the VV polarization channel in the linear domain based on the structural components. Detail components of VH polarization channels .

[0043] Further, step S2-1) includes:

[0044] VV polarization channel image in dB domain and VH polarization channel images Reconstructed into VV polarization channel images in the linear domain and VH polarization channel images And the VV polarization channel image under the linear domain. and VH polarization channel images Perform a natural logarithmic transformation to obtain the VV polarization channel image in the logarithmic domain. and VH polarization channel images :

[0045] ;

[0046] ;

[0047] ;

[0048] ;

[0049] in It is the natural logarithm function.

[0050] Further, step S2-2) includes:

[0051] In the logarithmic domain at scale k, the VV polarization channel self-guided structure component VV polarization channel image Structural components obtained from the guide image and the source image:

[0052] ;

[0053] VH polarization channel self-guided structure component VH polarization channel image Structural components obtained from the guide image and the source image:

[0054] ;

[0055] VH polarization channel inter-guided structural components VV polarization channel image As a guide image, using the VH polarization channel image As the structural components obtained from the source image, this method improves the stability and accuracy of VH polarization channel structural component estimation while ensuring that the polarization features of the VH polarization channel are not erased.

[0056] ;

[0057] in It is a guided filter function. To guide the image, For source image, The filter radius is... For regularization parameters, The scale is The filter radius at that time, and The relationship is:

[0058] ;

[0059] Further, steps S2-3) include:

[0060] Calculate the VV polarization channel image in the logarithmic domain within a local window. and VH polarization channel images The mean, variance, and covariance are used to calculate the polarization correlation coefficient, resulting in a polarization correlation weight used to characterize the coherence of the VV and VH polarization channels in a local region.

[0061] ;

[0062] ;

[0063] ;

[0064] ;

[0065] ;

[0066] ;

[0067] ;

[0068] in , These represent VV polarization channel images in the logarithmic domain. Mean and VH polarization channel images within a local window The mean within a local window, , These represent the VV polarization channel images in the logarithmic domain, respectively. Variance and VH polarization channel images within a local window Variance within a local window, VV polarization channel image in logarithmic domain and VH polarization channel images Covariance within a local window It is a local mean. Polarization correlation coefficient, It is a function that takes the maximum value. It is the polarization correlation weight;

[0069] A normalized polarization difference index was constructed to characterize the polarization response of the color steel tile target, which exhibits "significantly higher backscattering in the VV polarization channel than in the VH polarization channel":

[0070] ;

[0071] ;

[0072] in It is the polarization difference index. and These represent the VV polarization channel image and the VH polarization channel image in the linear domain, respectively; It is the normalized polarization difference index. It is a truncation function, which will... Limited to Within the range;

[0073] Construct joint weights for polarization channels to guide the self-guided structural components of the VH polarization channels. Guided structural components between VH polarization channels Adaptive adjustment is performed to enhance the guiding effect of VV on VH in regions where the VV polarization channel of the color steel tile has a significant scattering advantage and the correlation between the VV and VH polarization channels is high:

[0074] ;

[0075] in It is the joint weight of polarization channels, and its value range is... It is used to control the fusion ratio of guided and self-guided polarization channels. It is the polarization correlation weight. It is the normalized polarization difference index. It is a truncation function, which will... Limited to Within the range;

[0076] In order to maintain the independent polarization response of the VH polarization channel while utilizing the VV polarization channel for guidance, the self-guided structural components of the VH polarization channel are... Guided structural components between VH polarization channels Weighted fusion is performed to obtain the joint structural components of the VH polarization channels. :

[0077] ;

[0078] in It is the joint weight of polarization channels;

[0079] Further, steps S2-4) include:

[0080] Scale is VV polarization channel self-guided structure component VH polarization channel inter-guided structure components and VH polarization channel joint structure components By performing a natural exponential transform, the self-guided structure components of the VV polarization channel in the linear domain are obtained. VH polarization channel inter-guided structure components and VH polarization channel joint structure components :

[0081] ;

[0082] ;

[0083] ;

[0084] in Natural exponential function;

[0085] To address the physical characteristics of color-coated steel sheets exhibiting strong backscattering in the VV polarization channel but weak response in the VH polarization channel, the joint structural components of the VH polarization channels in the linear domain are analyzed within the high polarization difference region where the polarization difference index exceeds a preset threshold. An upper limit constraint is imposed, limiting it to no more than a preset scaling factor and the VV polarization channel self-guided structure component in the linear domain. The product range; within the non-high polarization difference region, the original VH polarization channel joint structure components are preserved. The estimated value:

[0086] ;

[0087] when hour:

[0088] ;

[0089] when hour:

[0090]

[0091] in High polarization difference mask, used to label polarization difference index Exceeding the preset polarization difference threshold The set of pixels, These are pixel space coordinates. The lower scale of the linear domain after upper bound constraint is The VH polarization channel joint structure components, It is a constraint proportionality coefficient. It is a function that takes the minimum value;

[0092] Further, steps S2-5) include:

[0093] Image of VV polarization channel in logarithmic domain The lower scale of the logarithmic field is Self-guided structure components The difference is calculated and remapped to the linear domain to serve as the detail components of the VV polarization channel. ; VH polarization channel image in the linear domain Its joint structural components with the VH polarization channel in the linear domain after upper limit constraint The ratio obtained by division is used as the detail component of the VH polarization channel. :

[0094] ;

[0095] ;

[0096] in Natural exponential function;

[0097] Specifically, step S3 includes:

[0098] For small-scale (k=3), medium-scale (k=5), and large-scale (k=7), after structure-detail decomposition guided by the polarization channels, the linear domain lower-scale is obtained as follows: VV polarization channel self-guided structure component VH polarization channel inter-structural components VH polarization channel joint structure components after upper limit constraint VV polarization channel detail components and VH polarization channel detail components ;

[0099] Calculate VV polarization channel images in the linear domain The variance within a local window is normalized and used as a measure of local texture intensity.

[0100] ;

[0101] ;

[0102] in It is a VV polarization channel image. Variance within a local window, It is a local mean. Normalize local texture intensity, It is a function that takes the maximum value;

[0103] The small-scale component weights are adaptively set based on the local texture intensity. Mesoscale component weights and large-scale component weights :

[0104] ;

[0105] ;

[0106] ;

[0107] in It is the normalized local texture intensity;

[0108] Weighted fusion of structural and detail components at each scale yields the fused VV polarization channel self-guided structural components. VH polarization channel inter-guided structure components VH polarization channel joint structure components VV polarization channel detail components and VH polarization channel detail components :

[0109] ;

[0110] ;

[0111] ;

[0112] ;

[0113] ;

[0114] in , , These represent the lower scale of the linear domain as follows: The self-guided structure components of the VV polarization channel, the inter-channel guided structure components of the VH polarization channel, and the joint structure components of the VH polarization channel. and These represent the lower scale of the linear domain as follows: The VV polarization channel detail components and the VH polarization channel detail components. It is the first Weights for each scale It is the first Each scale value.

[0115] Specifically, step S4 includes:

[0116] Compared to VH polarization channels, the VV polarization channels of color steel roofing sheets typically exhibit stronger and more stable backscattering. To reduce the interference of speckle and detail noise on the polarization difference measurement, the polarization scattering advantage of VV over VH is highlighted at the structural level, and the inter-polarization channel guiding structure index CG_SBI is constructed. This index incorporates the inter-polarization channel guiding structure component of VH polarization channels.

[0117] ;

[0118] in , These represent the self-guided structure component of the VV polarization channel and the guided structure component between the VH polarization channels after multi-scale fusion, respectively.

[0119] Specifically, step S5 includes:

[0120] Color-coated steel sheets typically exhibit a scattering characteristic of "structure dominance and weak detail" in the VV polarization channel, while the structure-detail ratio is relatively low in the VH channel. To comprehensively characterize the differences in structure dominance and detail variability between VV and VH, a joint structure-detail ratio index JG_SDRI guided by polarization channels is constructed. This index introduces the joint structure component of the VH polarization channel obtained by the fusion of self-guided polarization channels and inter-polarization channel guidance.

[0121] ;

[0122] ;

[0123] ;

[0124] in It is the VV polarization channel structure and detail ratio. It refers to the VH polarization channel structure and detail ratio. and These represent the self-guided structural components of the VV polarization channel and the joint structural components of the VH polarization channel after multi-scale fusion, respectively. and These represent the VV polarization channel detail components and the VH polarization channel detail components after multi-scale fusion, respectively.

[0125] Specifically, step S7 includes:

[0126] The various accuracy metrics include overall accuracy (OA), Kappa coefficient, and producer accuracy (PA), user accuracy (UA), and F1 score for the color steel tile category.

[0127] To verify the effectiveness of the method proposed in this application, an experiment was first conducted on the Sentinel-1 image, and the experimental results are shown in Table 1.

[0128] Table 1 Experimental results based on Sentinel-1 images

[0129]

[0130] In a pure SAR scenario using only Sentinel-1 imagery with dual VV and VH channels, experimental results show that the detection performance of corrugated steel roof tiles is significantly improved by introducing the inter-polarity channel guided structure index CG_SBI and the joint polarity channel guided structure-detail ratio index JG_SDRI. Compared with the random forest baseline using only the original intensities of VV and VH, the addition of CG_SBI increases the PA, UA, and F1 scores of the corrugated steel roof tile category by approximately 4 percentage points, OA by approximately 3.3 percentage points, and Kappa by approximately 7 percentage points. This demonstrates that, based on inter-polarity channel guidance, CG_SBI can highlight the polarization scattering advantage of VV over VH in a smooth and stable structural representation, thereby improving the distinguishability of corrugated steel roof tile targets from background features. Figure 4 The confusion matrix of the model after adding CG_SBI is shown. The TP of the corrugated steel tile category is 145 and the FN is 35, while the TN of the non-corrugated steel tile category is 237 and the FP is 33. The high proportion of diagonal elements indicates that the model after adding CG_SBI has a good ability to distinguish between the two types of land cover. Figure 5This paper presents the spatial distribution results of corrugated steel roof tiles in the entire Sentinel-1 image, using a random forest classifier trained with the CG_SBI index. Based on polarization channel guidance and self-guided fusion, JG_SDRI comprehensively characterizes the differences in structural dominance and detail undulation between the VV and VH polarization channels, enabling corrugated steel roof tile hazard targets to maintain high discriminability and robustness in complex backgrounds, while also delivering stable accuracy gains, although the improvement is slightly lower than that of CG_SBI.

[0131] Based on the experiments relying solely on Sentinel-1 imagery, it can be verified that the proposed polarization channel guided structure-decomposition of details and its constructed CG_SBI and JG_SDRI indices can significantly improve the detection accuracy of color steel tiles even without optical information. Building upon this, to further examine the applicability and robustness of this method in multi-source data scenarios, this application designs supplementary experiments using fused Sentinel-1 and Sentinel-2 images as the data source to evaluate the effectiveness of the proposed method in high-dimensional optical-radar feature spaces.

[0132] Sentinel-1 and Sentinel-2 fused images of railway lines were acquired during the summer of 2024 (June 1 to August 31). The study area was a 1 km buffer zone on both sides of some railway lines within Xinhua District, Qiaoxi District, and Luquan District of Shijiazhuang City, Hebei Province. For Sentinel-1 SAR data, images with IW imaging mode, VV / VH dual polarization, and 10m resolution within this period were selected and generated as seasonal SAR images through median composite. For Sentinel-2 optical data, images with cloud cover below 30% during the same period were selected, and cloud masking denoising was performed using the SCL band, retaining bands B2, B3, B4, B5, B6, B7, B8, B8A, B11, and B12, which were also composited using median composite. Finally, the VV and VH bands of the Sentinel-1 images were spatially registered and stacked with the 10 bands of the Sentinel-2 images at the pixel scale to generate a fused image with a uniform 10m resolution.

[0133] Table 2. Experimental results based on Sentinel-1 and Sentinel-2 fused images

[0134]

[0135] In the Sentinel-1 and Sentinel-2 fused imagery scenario, under the high accuracy conditions of already including VV, VH, and 10 optical bands, with a baseline F1 score of 93.44% and a Kappa of nearly 89%, the further introduction of CG_SBI and JG_SDRI still showed a clear gain effect. The results showed that adding CG_SBI slightly improved the detection accuracy, mainly reflected in the improvement of producer accuracy (PA) for color steel tile categories; while adding JG_SDRI improved the F1 score to 94.05%, OA to 95.11%, and Kappa to 89.91%, achieving the best overall performance. This indicates that this index can provide additional discriminative information from the perspective of polarization structure-detail ratio, given that the optical spectral and texture features are already relatively complete, thus effectively complementing the multi-source high-dimensional features and further improving the overall accuracy and robustness of the color steel tile detection model.

[0136] Specific embodiments of the present invention have been described above. Those skilled in the art should understand that this disclosure is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of this disclosure. Various changes and modifications can be made to this disclosure without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present disclosure as claimed.

Claims

1. A method for identifying potential hazards in railway corrugated steel roofing sheets based on SAR polarization channel guidance, characterized in that... Includes the following steps: S1. Acquire Sentinel-1 images covering the study area along the railway line, label the color steel tile and non-color steel tile sample points in the study area, and divide the sample dataset into training set and test set. The Sentinel-1 images include the backscattering coefficients of the VV polarization channel and the VH polarization channel. S2, performs logarithmic transformation on the VV and VH polarization channel images in the linear domain, and performs structure-detail decomposition based on polarization channel guidance in the logarithmic domain; S3 performs structure-detail decomposition guided by polarization channels at small, medium and large scales respectively, and adopts an adaptive fusion strategy to adaptively adjust the weights of structural components and detail components at each scale according to the local texture intensity, thereby obtaining fused multi-scale structural components and detail components. S4, using the fused multi-scale structural components to construct the polarization channel guide structure index CG_SBI; S5, using the fused multi-scale structural components and detail components to construct a polarization channel joint guidance structure – detail ratio index JG_SDRI, where the polarization channel joint guidance includes both polarization channel self-guidance and polarization channel inter-guidance; S6. The polarization channel guidance structure index CG_SBI, the polarization channel joint guidance structure-detail ratio index JG_SDRI, the backscattering coefficient of the VV polarization channel and the backscattering coefficient of the VH polarization channel are combined into a feature vector. A random forest classifier is trained based on the training set samples to obtain the color steel tile hidden danger identification model. S7 uses a color steel tile hazard identification model to predict test set samples, calculates confusion matrix and various accuracy indicators, and identifies color steel tile hazards on the entire Sentinel-1 image.

2. The method for identifying potential hazards in railway corrugated steel roofing tiles based on SAR polarization channel guidance as described in claim 1, characterized in that, Step S2 specifically includes the following steps: Step S21, the VV polarization channel image in the dB domain and VH polarization channel images Reconstructed into VV polarization channel images in the linear domain and VH polarization channel images And the VV polarization channel image under the linear domain. and VH polarization channel images Perform a natural logarithmic transformation to obtain the VV polarization channel image in the logarithmic domain. and VH polarization channel images ; Step S22, in the logarithmic field, at a scale of At that time, guided filtering was used to estimate three structural components, namely the VV polarization channel self-guided structural components. VH polarization channel self-guided structure component and the guiding structure components between VH polarization channels ; Step S23: Construct a joint weight for the polarization channel determined by the polarization correlation weight and the normalized polarization difference index, which is used to guide the self-guided structural components of the VH polarization channel. Guided structural components between VH polarization channels Adaptive adjustment is performed to obtain the joint structural components of the VH polarization channels. ; Step S24, the VV polarization channel self-guided structure component VH polarization channel inter-guided structure components and VH polarization channel joint structure components By performing a natural exponential transform, the self-guided structure components of the VV polarization channel in the linear domain are obtained. VH polarization channel inter-guided structure components and VH polarization channel joint structure components And the joint structural components of the VH polarization channels. Apply upper limit constraints; Step S25: Calculate the detail components of the VV polarization channel in the linear domain based on the structural components. Detail components of VH polarization channels .

3. The method for identifying potential hazards in railway corrugated steel roofing tiles based on SAR polarization channel guidance as described in claim 2, characterized in that: Step S22 specifically includes the following steps: In the logarithmic field, the scale is At that time, the VV polarization channel self-guided structure component VV polarization channel image Structural components are obtained from the guide image and the source image: ； VH polarization channel self-guided structure component VH polarization channel image Structural components are obtained from the guide image and the source image: ； VH polarization channel inter-guided structural components VV polarization channel image As a guide image, using the VH polarization channel image Structural components are obtained from the source image: ； in It is a guided filter function. To guide the image, For source image, The filter radius is... For regularization parameters, The scale is The filtering radius at that time, and The relationship is: 。 4. The method for identifying potential hazards along railway color steel tiles based on SAR polarization channel guidance as described in claim 2, characterized in that: S23 includes the following steps: Calculate the VV polarization channel image in the logarithmic domain within a local window. and VH polarization channel images The mean, variance, and covariance are used to calculate the polarization correlation coefficient, resulting in a polarization correlation weight used to characterize the coherence of the VV and VH polarization channels in a local region. ； in It is the polarization correlation weight. Polarization correlation coefficient; A normalized polarization difference index was constructed to characterize the polarization response of the color steel tile target, where the backscattering of the VV polarization channel is significantly higher than that of the VH polarization channel: ； ； in It is the polarization difference index. and These represent the VV polarization channel image and the VH polarization channel image in the linear domain, respectively; It is the normalized polarization difference index. It is a truncation function, which will... Limited to Within the range; Construct joint weights for polarization channels to guide the self-guided structural components of the VH polarization channels. Guided structural components between VH polarization channels Adaptive adjustment is performed to enhance the guiding effect of VV on VH in regions where the VV polarization channel of the color steel tile has a significant scattering advantage and the correlation between the VV and VH polarization channels is high: ； in It is the joint weight of polarization channels, and its value range is... It is used to control the fusion ratio of guided and self-guided polarization channels. It is the polarization correlation weight. It is the normalized polarization difference index. It is a truncation function, which will... Limited to Within the range; Self-guided structural components of VH polarization channels Guided structural components between VH polarization channels Weighted fusion is performed to obtain the joint structural components of the VH polarization channels. : ； in It is the joint weight of polarization channels.

5. The method for identifying potential hazards along railway color steel tiles based on SAR polarization channel guidance as described in claim 2, characterized in that: S24 includes the following steps: Scale is VV polarization channel self-guided structure component VH polarization channel inter-guided structure components and VH polarization channel joint structure components By performing a natural exponential transform, the self-guided structure components of the VV polarization channel in the linear domain are obtained. VH polarization channel inter-guided structure components and VH polarization channel joint structure components : ； ； ； in It is a natural exponential function; Within the high polarization difference region where the polarization difference index exceeds a preset threshold, the joint structural components of the VH polarization channel in the linear domain are analyzed. An upper limit constraint is imposed, limiting it to no more than a preset scaling factor and the VV polarization channel self-guided structure component in the linear domain. The product range; within the non-high polarization difference region, the original VH polarization channel joint structure components are preserved. The estimated value: ； when hour: ； when hour: ； in High polarization difference mask, used to label polarization difference index Exceeding the preset polarization difference threshold The set of pixels, These are pixel space coordinates. The lower scale of the linear domain after upper bound constraint is The VH polarization channel joint structure components, It is a constraint proportionality coefficient. It is a function that takes the minimum value.

6. The method for identifying potential hazards along railway color steel tiles based on SAR polarization channel guidance as described in claim 2, characterized in that: S25 includes the following steps: Image of VV polarization channel in logarithmic domain The lower scale of the logarithmic field is Self-guided structure components The difference is calculated and remapped to the linear domain to serve as the detail components of the VV polarization channel. ; VH polarization channel image in the linear domain Its joint structural components with the VH polarization channel in the linear domain after upper limit constraint The ratio obtained by division is used as the detail component of the VH polarization channel. : ； ； in It is a natural exponential function.

7. The method for identifying potential hazards in railway corrugated steel roofing tiles based on SAR polarization channel guidance as described in claim 1, characterized in that: S3 includes the following steps: For small-scale, medium-scale, and large-scale operations, after structure-detail decomposition guided by the polarization channels, the linear domain lower-scale values ​​are obtained as follows: VV polarization channel self-guided structure component VH polarization channel inter-structural components VH polarization channel joint structure components after upper limit constraint VV polarization channel detail components and VH polarization channel detail components ; Calculate VV polarization channel image in the linear domain The variance within a local window is normalized and used as a measure of local texture intensity. ； ； in It is a VV polarization channel image. Variance within a local window, It is a local mean. Normalize local texture intensity, It is a function that takes the maximum value; The small-scale component weights are adaptively set based on the local texture intensity. Mesoscale component weights and large-scale component weights : ； ； ； in It is the normalized local texture intensity; Weighted fusion of structural and detail components at each scale yields the fused VV polarization channel self-guided structural components. VH polarization channel inter-guided structure components VH polarization channel joint structure components VV polarization channel detail components and VH polarization channel detail components : ； ； ； ； ； in , , These represent the lower scale of the linear domain as follows: The self-guided structure components of the VV polarization channel, the inter-channel guided structure components of the VH polarization channel, and the joint structure components of the VH polarization channel. and These represent the lower scale of the linear domain as follows: The VV polarization channel detail components and the VH polarization channel detail components. It is the first Weights for each scale It is the first Each scale value.

8. The method for identifying potential hazards of corrugated steel roofing sheets along railway lines based on SAR polarization channel guidance as described in claim 1, characterized in that: S4 includes the following steps: Construct the inter-polarization channel guiding structure index CG_SBI, which incorporates the VH inter-polarization channel guiding structure component: ； in , These represent the self-guided structure component of the VV polarization channel and the guided structure component between the VH polarization channels after multi-scale fusion, respectively.

9. The method for identifying potential hazards in railway corrugated steel roofing tiles based on SAR polarization channel guidance as described in claim 1, characterized in that: S5 includes the following steps: Constructing a joint polarization channel guidance structure – the detail ratio index JG_SDRI, which introduces the VH polarization channel joint structure component obtained by the fusion of self-guidance and inter-channel guidance: ； ； ； in It is the VV polarization channel structure and detail ratio. It refers to the VH polarization channel structure and detail ratio. and These represent the self-guided structural components of the VV polarization channel and the joint structural components of the VH polarization channel after multi-scale fusion, respectively. and These represent the VV polarization channel detail components and the VH polarization channel detail components after multi-scale fusion, respectively.

Images