Intelligent identification method and system for creep landslide hazards based on winding interference phase

Abstract

The present invention discloses a method and system for intelligent identification of creep-type landslide hazards based on winding interference phase. The method first obtains multi-view time-continuous winding interference pattern data and upsamples it to a fixed input size of 128x128 through bilinear interpolation. Subsequently, the winding interference pattern is mapped using sine and cosine functions to generate sine phase and cosine phase to further enrich the phase information. Through the set mapping function, these phase data are converted into RGB color space to achieve multi-dimensional expression of data. In the feature extraction and fusion stage, a multi-branch network architecture with shared weights is adopted to extract independent features for each RGB channel, and the same-layer features are integrated through a feature fusion mechanism to extract high-dimensional semantic information. Then, the feature information is restored step by step through the decoder and the classification mapping is completed, and finally the prediction result of the single-view interference pattern is obtained.

Description

Technical Field

[0001] The invention belongs to the technical field of synthetic aperture radar, and in particular relates to a method and system for intelligently identifying hidden dangers of creep-type landslides based on winding interferometry phase. Background Art

[0002] Landslide is a common geological disaster, which is a process in which rocks and soil move downward along a sliding surface. The occurrence of landslides is affected by factors such as topography, climatic conditions and human activities. Landslides have a significant impact on human life and property, resulting in a large number of casualties and economic losses every year. Therefore, accurate identification of landslide hazards is of great significance for disaster prevention and mitigation. Traditional landslide identification relies on field surveys and on-site data collection, but is limited by limited spatial coverage, insufficient real-time capabilities and high costs. Advances in remote sensing technology have provided new possibilities for non-contact and wide-area coverage landslide detection. The high sensitivity of interferometric synthetic aperture radar (InSAR) to surface displacement changes makes it an important tool for landslide identification. However, the lack of effective automated interpretation methods and excessive reliance on expert knowledge have hindered the application of InSAR technology in landslide identification in large-area rapid response scenarios.

[0003] As a data-driven technology, deep learning has profoundly changed the task of geological disaster investigation with its efficient feature extraction capabilities. In the study of combining deep learning with InSAR technology for landslide hazard identification, two methods are usually used. One method involves identifying the difference between the atmospheric delay phase and the deformation phase in the unwrapped interferogram. By using image denoising technology to isolate the atmospheric delay phase, the displacement is restored and the deformation field boundary is depicted. However, the residual phase caused by the atmospheric phase screen shows spatial similarity with the deformation phase in the unwrapped interferogram, and this similarity often leads to misclassification, which limits the universality of these methods.

[0004] Another approach considers InSAR deformation products, which utilize time series deformation products as network input and aim to segment the landslide deformation area and accurately delineate its boundaries through semantic segmentation and object detection techniques. However, this approach is susceptible to factors such as decorrelation, significant terrain fluctuations, and steep deformation gradients during implementation, resulting in phase unwrapping errors and suboptimal data utilization. In addition, the inherent computational intensity of time series InSAR processing, coupled with a high reliance on expertise, significantly increases the time and labor expenditure in the preparation phase. Summary of the invention

[0005] In order to solve the technical problems existing in the background technology, the present invention aims to provide an intelligent identification method and system for creep landslide hazards based on winding interference phase, aiming to accurately extract the creep landslide area directly from the winding interference map, thereby solving the defects of high cost and high computational intensity in the prior art.

[0006] In order to solve the technical problem, the technical solution of the present invention is:

[0007] A method for intelligently identifying creep-type landslide hazards based on winding interference phase, the method comprising:

[0008] S1: Data acquisition and preprocessing: Obtain multi-view time-continuous winding interferograms of the target area, use bilinear interpolation to upsample the winding interferogram to the network input size, map the winding interferogram to sine phase and cosine phase through sine and cosine functions, map the upsampled winding interferogram, sine phase and cosine phase to RGB color space, and obtain a three-channel RGB image;

[0009] S2: Feature extraction and fusion: The three-channel RGB image is abstracted into high-dimensional semantic information through three parallel encoding processes with shared weights, the same-layer encoding information of different branches is fused through the feature fusion mechanism, the semantic information is restored to representation information step by step through the decoder, and finally the classification number mapping is completed, and the single-scene interference map prediction result output by the model is output;

[0010] S3: Model generalization mechanism based on intersection and union decision: The single-scene interference graph prediction result output by the model is used to produce the final prediction result through intersection and union decision.

[0011] Furthermore, in the step S1: the original size of the data is upsampled to 128 by bilinear interpolation to adapt to the fixed data input size of the network; the winding interference pattern is mapped into sine phase and cosine phase by function mapping;

[0012] The phase, sine phase and cosine phase are mapped into RGB three-channel matrix by color rendering:

[0013] RGB(x,y)=C(I(x,y)) (1)

[0014] Where C represents the mapping function, I(x, y) represents the pixel value of the input single-channel data, and RGB(x, y) represents the pixel value of the mapped three-channel data.

[0015] Furthermore, in step S2, the three-channel RGB image is abstracted into high-dimensional semantic information through three weight-sharing parallel encoding processes, including:

[0016] Through step-by-step encoding, the phase, cosine phase and sine phase are extracted according to three independent branches; it consists of five encoding blocks, each of which is responsible for feature extraction and data compression tasks. The convolution block contains two convolution layers, 3×3 convolution layer-batch normalization layer-linear rectification unit, and data compression is achieved through maximum pooling;

[0017] Among them, the calculation formula of the convolution layer is:

[0018]

[0019] In the formula, O ( :,: ,z) represents the output feature map of the zth channel, M ( :,: ,k) represents the input feature map of the kth channel, * represents the convolution operation, F (：,：,k,z) Indicates that (：,：,k) and O (：,：,z) The corresponding convolution kernel, B (：,：,z) represents the bias matrix of the zth channel; C represents the total number of channels of the input feature map;

[0020] The batch normalization layer calculation formula is:

[0021]

[0022] In the formula, O Z and M z denote the zth batch of output and input, μ z is the mean of the current batch, and the formula is:

[0023]

[0024] In the formula, x i ∈[x 1 ,x 2 ,…,x m ] represents a set of m small batches of input data, is the variance of the current batch, and the formula is:

[0025]

[0026] γ z and β z is a learnable parameter, ε z Prevent the denominator from being 0 during the calculation process;

[0027] The formula for the rectified linear unit is:

[0028] f(x)=max(0,x) (6)

[0029] In the formula, x is the input feature map, and the maximum pooling formula is expressed as:

[0030]

[0031] In the formula, x m,n is the value of the input feature map at (m,n), R i,j is the pooling domain in x m,n The collection of locations on .

[0032] Furthermore, in step S2, the same-layer coding information of different branches is fused through a feature fusion mechanism, including:

[0033] By superimposing multi-branch features in the feature map dimension, the same-layer information of different branches in the encoding process is fused. This mechanism has three parallel branches. In each branch, the input feature map first passes through a 3×3 convolution layer and then performs an algebraic AND operation with itself. The formula is expressed as:

[0034]

[0035] Where M k represents the input feature map of the kth branch, * represents the convolution operation, F k Represents the convolution kernel of the kth branch.

[0036] Furthermore, in step S2, the semantic information is gradually restored to representation information through a decoder, and finally the classification number mapping is completed, including:

[0037] The decoding process includes five decoding blocks in total: each decoding block is responsible for category mapping and contains only pixel-level convolution. The decoding block I consists of a convolution block-skip connection layer-rectified linear unit-batch normalization-3×3 convolution and a 2x upsampling layer;

[0038] Among them, the formula of the jump connection layer is expressed as:

[0039] F concat =concat(F e ,F d ) (9)

[0040] In the formula, F e represents the feature map output by the feature fusion mechanism, F d represents the feature map corresponding to the current decoding layer; concat represents the skip connection calculation, F concat Output result of the skip connection layer.

[0041] Further, the step S3 specifically includes:

[0042] The final prediction product is obtained by taking the intersection and union of the prediction results of the single-view interference graph; the formula can be expressed as:

[0043]

[0044] In the formula, A i , A i-1 , A i+1 It represents the prediction result of the adjacent three-scene interference graph, ∩ represents the intersection operation, and ∪ represents the union operation.

[0045] A creep-type landslide hazard intelligent identification system based on winding interference phase, the system is applied to the above-mentioned method, and the system comprises:

[0046] Data acquisition and preprocessing module: used to obtain the multi-view time-continuous winding interferogram of the target area, upsample the winding interferogram to the network input size using bilinear interpolation, map the winding interferogram to sine phase and cosine phase through sine and cosine functions, and map the upsampled winding interferogram, sine phase and cosine phase to RGB color space to obtain a three-channel RGB image;

[0047] Feature extraction and fusion module: used to abstract the three-channel RGB image into high-dimensional semantic information through three shared weight parallel encoding processes, fuse the same-layer encoding information of different branches through the feature fusion mechanism, restore the semantic information to representation information step by step through the decoder, and finally complete the classification number mapping, and output the single-scene interference map prediction result output by the model;

[0048] Intersection and merging decision processing module: the single-scene interference graph prediction results output by the model are processed through intersection and merging decision to produce the final prediction results.

[0049] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, any one of the above-mentioned intelligent identification methods for creep-type landslide hazards based on winding interference phase is implemented.

[0050] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements any one of the above-mentioned intelligent identification methods for creep-type landslide hazards based on winding interference phase.

[0051] Compared with the prior art, the advantages of the present invention are:

[0052] The deformation area and distribution of the creep landslide area obtained using InSAR data can provide an important reference for the prevention and control of landslide hazards, and greatly improve the accuracy and efficiency of identifying large-scale creep landslide hazards.

[0053] Improved analysis accuracy: Through the acquisition of multi-scene time-continuous entanglement interferograms and high-dimensional feature extraction, this method can effectively capture and express complex phase information, thereby significantly improving the detection and classification accuracy of the target.

[0054] Enhanced information expression: Through sine and cosine mapping, the phase information is converted into a three-channel RGB image, which enriches the color expression of the data, makes the subsequent feature extraction and analysis more visual, and improves the intuitiveness and interpretability of the results.

[0055] Flexible network architecture: The multi-branch feature extraction framework makes the independent extraction of each channel information more efficient, and the feature fusion mechanism can effectively integrate the features of each channel, thereby improving the generalization ability and adaptability of the model.

[0056] Fast convergence: Through color mapping and multi-branch feature fusion, the convergence speed of the neural network is accelerated, which helps to shorten the model training time and improve the overall efficiency.

[0057] Wide application potential: This method is applicable to many fields, such as geological exploration, engineering monitoring and material testing, etc. It has good practical value and promotion prospects, and provides effective technical support for related research. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is the main flow chart of the present invention;

[0059] Figure 2 It is a network structure diagram;

[0060] Figure 3 It is the model generalization graph. DETAILED DESCRIPTION

[0061] The specific implementation mode of the present invention is described below in conjunction with embodiments:

[0062] It should be noted that the structures, proportions, sizes, etc. shown in this specification are only used to match the contents disclosed in the specification so that people familiar with this technology can understand and read them, and are not used to limit the conditions under which the present invention can be implemented. Any structural modification, change in proportional relationship or adjustment of size should still fall within the scope of the technical content disclosed in the present invention without affecting the effects and purposes that can be achieved by the present invention.

[0063] At the same time, the terms such as "upper", "lower", "left", "right", "middle" and "one" cited in this specification are only for the convenience of description and are not used to limit the scope of implementation of the present invention. Changes or adjustments to their relative relationships should be regarded as the scope of implementation of the present invention without substantially changing the technical content.

[0064] Embodiment 1:

[0065] like Figure 1 As shown, a method for intelligently identifying creep-type landslide hazards based on winding interference phase is provided. The complete steps of the method of the present invention include:

[0066] (1) Winding interference pattern processing based on image rendering

[0067] Step S101: Obtaining a multi-view time-continuous winding interferogram of the study area

[0068] This step requires obtaining the winding interference pattern of the research area, and the data can be obtained by searching and ordering on the Internet or obtaining through a storage device.

[0069] Step S102: up-sample the winding interference pattern to the network input size.

[0070] This step uses bilinear interpolation to upsample the original data size to 128 to meet the fixed data input size of the network.

[0071] Step S103: Map the winding interference pattern into a sine phase and a cosine phase through sine and cosine functions.

[0072] In this step, the winding interference pattern is mapped into sine phase and cosine phase by function mapping, which is used to enrich the expression of phase information and accelerate network convergence.

[0073] Step S104: Map the phase (winding interference pattern), sine phase, and cosine phase to the RGB color space through a predetermined color interval.

[0074] This step uses color rendering to map the phase, sine phase, and cosine phase into an RGB three-channel matrix to further refine the data in the channel dimension and accelerate network convergence.

[0075] RGB(x,y)=C(I(x,y)) (2)

[0076] Where C represents the mapping function, I(x, y) represents the pixel value of the input single-channel data, and RGB(x, y) represents the pixel value of the mapped three-channel data.

[0077] (2) Multi-branch network framework based on parallel branch encoding and progressive feature fusion

[0078] Step S201 : abstracting the phase, cosine phase and sine phase into high-dimensional semantic information through three parallel encoding processes with shared weights.

[0079] In this step, the phase, cosine phase and sine phase are encoded step by step and feature extracted in three independent branches. This process consists of five encoding blocks, each of which is responsible for feature extraction and data compression tasks. The convolution block contains two convolution layers (3×3 convolution layer-batch normalization layer-linear rectifier unit). Data compression is achieved through maximum pooling.

[0080] Among them, the calculation formula of the convolution layer is:

[0081]

[0082] In the formula, O (：,：,z) represents the output feature map of the zth channel, M (：,：,k) represents the input feature map of the kth channel, * represents the convolution operation, F (：,：,k,z) Indicates that (：,：,k) and O (：,：,z) The corresponding convolution kernel, B (：,：,z) Represents the bias matrix of the z-th channel.

[0083] The batch normalization layer calculation formula is:

[0084]

[0085] In the formula, O z and M z denote the zth batch of output and input, μ z is the mean of the current batch, and the formula is:

[0086]

[0087] is the variance of the current batch, and the formula is:

[0088]

[0089] γ z and β z is a learnable parameter, ε z Prevent the denominator from being 0 during calculation.

[0090] The formula for the rectified linear unit is:

[0091] f(x)=max(0,x) (7)

[0092] The formula for maximum pooling is expressed as:

[0093]

[0094] In the formula, x m,n is the value of the input feature map at (m,n), R i,j is the pooling domain in xm,n The collection of locations on .

[0095] Step S202: Fusing the same-layer coding information of different branches through a feature fusion mechanism.

[0096] like Figure 2 As shown in the figure, this step combines the same-layer information of different branches in the encoding process by superimposing multi-branch features in the feature map dimension. This mechanism has three parallel branches. In each branch, the input feature map first passes through a 3×3 convolution layer and then performs an algebraic sum operation with itself. The formula is expressed as:

[0097]

[0098] Where M k represents the input feature map of the kth branch, * represents the convolution operation, F k Represents the convolution kernel of the kth branch.

[0099] Step S203: restore the semantic information to representation information step by step through the decoder, and finally complete the classification number mapping.

[0100] The decoding process of this step includes five specific decoding blocks. Each decoding block, except for decoding block I which is responsible for category mapping and only contains pixel-level convolution, consists of a convolution block-skip connection layer-rectified linear unit-batch normalization-3×3 convolution and a 2x upsampling layer.

[0101] Among them, the formula of the jump connection layer is expressed as:

[0102] F concat =concat(F e ,F d ) (10)

[0103] In the formula, F e represents the feature map output by the feature fusion mechanism, F d Represents the feature map corresponding to the current decoding layer.

[0104] (3) Model generalization mechanism based on intersection decision

[0105] Step S301: The single-scene interference graph prediction results output by the model are used to produce a final prediction product through intersection and union decision.

[0106] In this step, the prediction results of the single-view interference graph are used to obtain the final prediction product by taking the intersection and union. The formula can be expressed as:

[0107]

[0108] In the formula, A i , A i-1, A i+1 It represents the prediction result of the adjacent three-scene interference graph, ∩ represents the intersection operation, and ∪ represents the union operation.

[0109] Embodiment 2:

[0110] This example gives the optimization results, using the public data (099A_06014_131313) from the COMET-LiCSSentinel-1InSAR portal, the study area is the middle reaches of the Jinsha River in China, the data time is from July 30, 2019 to October 22, 2019, the time baseline is 12 days, and a total of 7 interferograms are included. Figure 3 .

[0111] Depend on Figure 3 It can be seen that the method of the present invention has the effect of accurately identifying the creep-type landslide area in the middle reaches of the Jinsha River in China. Specifically, MB-Net is used to perform sliding window prediction on these interference patterns with a step size of 128. This process generates 7 preliminary prediction outputs. Various noise factors such as atmospheric delay errors in single-scene interference patterns are considered, and creep-type landslides are considered to be a continuous process. Therefore, adjacent results are combined to generate secondary prediction products, a total of 5. In addition, in order to ensure that the boundaries of these areas are clear, the secondary prediction outputs are unioned to obtain the final product.

[0112] Embodiment 3:

[0113] This embodiment provides a terminal device, which includes a processor and a memory, wherein the memory is used to store a computer program, the computer program includes program instructions, and the processor is used to execute the program instructions stored in the computer storage medium. The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is suitable for implementing one or more instructions, and is specifically suitable for loading and executing one or more instructions to implement the corresponding method flow or corresponding function; the processor described in the embodiment of the present invention can be used for the operation of a method for intelligent identification of creep-type landslide hazards based on winding interference phase, including the following steps:

[0114] S1: Data acquisition and preprocessing: Obtain multi-view time-continuous winding interferograms of the target area, use bilinear interpolation to upsample the winding interferogram to the network input size, map the winding interferogram to sine phase and cosine phase through sine and cosine functions, map the upsampled winding interferogram, sine phase and cosine phase to RGB color space, and obtain a three-channel RGB image;

[0115] S2: Feature extraction and fusion: The three-channel RGB image is abstracted into high-dimensional semantic information through three parallel encoding processes with shared weights, the same-layer encoding information of different branches is fused through the feature fusion mechanism, the semantic information is restored to representation information step by step through the decoder, and finally the classification number mapping is completed, and the single-scene interference map prediction result output by the model is output;

[0116] S3: Model generalization mechanism based on intersection and union decision: The single-scene interference graph prediction result output by the model is used to produce the final prediction result through intersection and union decision.

[0117] Embodiment 4:

[0118] This embodiment provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a terminal device for storing programs and data. It is understandable that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and the extended storage medium supported by the terminal device. The computer-readable storage medium provides a storage space, which stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space, and these instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory.

[0119] The processor may load and execute one or more instructions stored in a computer-readable storage medium to implement the corresponding steps of the above-mentioned embodiment related to a method for intelligently identifying creep-type landslide hazards based on winding interference phase; the processor may load and execute the following steps of one or more instructions in the computer-readable storage medium:

[0120] S1: Data acquisition and preprocessing: Obtain multi-view time-continuous winding interferograms of the target area, use bilinear interpolation to upsample the winding interferogram to the network input size, map the winding interferogram to sine phase and cosine phase through sine and cosine functions, map the upsampled winding interferogram, sine phase and cosine phase to RGB color space, and obtain a three-channel RGB image;

[0121] S2: Feature extraction and fusion: The three-channel RGB image is abstracted into high-dimensional semantic information through three parallel encoding processes with shared weights, the same-layer encoding information of different branches is fused through the feature fusion mechanism, the semantic information is restored to representation information step by step through the decoder, and finally the classification number mapping is completed, and the single-scene interference map prediction result output by the model is output;

[0122] S3: Model generalization mechanism based on intersection and union decision: The single-scene interference graph prediction result output by the model is used to produce the final prediction result through intersection and union decision.

[0123] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0124] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0125] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0126] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0127] The preferred embodiments of the present invention are described in detail above, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge scope of ordinary technicians in this field without departing from the purpose of the present invention.

[0128] Many other changes and modifications may be made without departing from the concept and scope of the present invention.It should be understood that the present invention is not limited to the specific embodiments, and the scope of the present invention is defined by the appended claims.

Claims

1. A method for intelligent identification of creep-type landslide hazards based on winding interference phase, characterized in that: The method comprises: S1: Data acquisition and preprocessing: Obtain the multi-view time-continuous winding interferogram of the target area, use the bilinear interpolation method to upsample the winding interferogram to the network input size, map the winding interferogram to sine phase and cosine phase through sine and cosine functions, and map the upsampled winding interferogram, sine phase and cosine phase to RGB color space respectively to obtain a three-channel RGB image; S2: Feature extraction and fusion: The three-channel RGB image is abstracted into high-dimensional semantic information through three parallel encoding processes with shared weights. Through step-by-step encoding, the phase, cosine phase and sine phase are feature extracted according to three independent branches; it consists of five encoding blocks, each of which is responsible for feature extraction and data compression tasks. The convolution block contains two convolution layers, 3×3 convolution layer-batch normalization layer-linear rectifier unit. Data compression is achieved through maximum pooling. The same-layer encoding information of different branches is fused through the feature fusion mechanism. The same-layer information of different branches in the encoding process is fused by superimposing multi-branch features in the feature map dimension. The mechanism has three parallel branches; in each branch, the input feature map first passes through a 3×3 convolution layer, and then performs an algebraic and operation with itself. The formula is expressed as: Where M k represents the input feature map of the kth branch, * represents the convolution operation, F k Represents the convolution kernel of the kth branch; The semantic information is restored to representation information step by step through the decoder, and finally the classification number mapping is completed. The decoding process includes five decoding blocks: each decoding block is responsible for category mapping and only includes pixel-level convolution decoding block I, which consists of convolution block-skip connection layer-rectified linear unit-batch normalization-3×3 convolution and 2 times upsampling layer; the single scene interference map prediction result output by the output model; S3: Model generalization mechanism based on intersection and union decision: The final prediction product is obtained by taking the intersection and union of the prediction results of the single-view interference graph; the formula can be expressed as: In the formula, A i , A i-1 , A i+1 It represents the prediction result of the adjacent three-scene interference graph, ∩ represents the intersection operation, and ∪ represents the union operation.

2. The method for intelligent identification of creep landslide hazards based on winding interference phase according to claim 1 is characterized in that: In the step S1: the original size of the data is upsampled to 128 by bilinear interpolation to adapt to the fixed data input size of the network; the winding interference pattern is mapped into sine phase and cosine phase by function mapping; The phase, sine phase and cosine phase are mapped to RGB three-channel matrices respectively by color rendering: RGB(x,y)=C(I(x,y)) (1) Where C represents the mapping function, I(x, y) represents the pixel value of the input single-channel data, and RGB(x, y) represents the pixel value of the mapped three-channel data.

3. The method for intelligent identification of creep landslide hazards based on winding interference phase according to claim 1 is characterized in that: In step S2: The calculation formula of the convolution layer is: In the formula, O (：,：,z) represents the output feature map of the zth channel, M (：,：,k) represents the input feature map of the kth channel, * represents the convolution operation, F (：,：,k,z) Indicates that (：,：,k) and O (：,：,z) The corresponding convolution kernel, B (：,：,z) represents the bias matrix of the zth channel, and C represents the total number of channels of the input feature map; The batch normalization layer calculation formula is: In the formula, O z and M z denote the zth batch of output and input, μ z is the mean of the current batch, and the formula is: In the formula, x i ∈[x1,x2,…,x m ] represents a set of m small batches of input data, is the variance of the current batch, and the formula is: γ z and β z is a learnable parameter, ε z Prevent the denominator from being 0 during the calculation process; The formula for the rectified linear unit is: f(x)=max(0,x) (6) In the formula, x is the input feature map, and the maximum pooling formula is expressed as: In the formula, x m,n is the value of the input feature map at (m,n), R i,j is the pooling domain in x m,n The collection of locations on .

4. The method for intelligent identification of creep landslide hazards based on winding interference phase according to claim 1 is characterized in that: In step S2: The formula of the skip connection layer is expressed as: F concat =concat(F e ,F d ) (9) In the formula, F e represents the feature map output by the feature fusion mechanism, F d represents the feature map corresponding to the current decoding layer; concat represents the skip connection calculation, F concat Output result of the skip connection layer.

5. An intelligent identification system for creep-type landslide hazards based on winding interference phase, characterized in that: The system is applied to the method described in any one of claims 1 to 4, and the system comprises: Data acquisition and preprocessing module: used to obtain the multi-view time-continuous winding interferogram of the target area, upsample the winding interferogram to the network input size using bilinear interpolation, map the winding interferogram to sine phase and cosine phase through sine and cosine functions, and map the upsampled winding interferogram, sine phase and cosine phase to RGB color space to obtain a three-channel RGB image; Feature extraction and fusion module: used to abstract the three-channel RGB image into high-dimensional semantic information through three shared weight parallel encoding processes, fuse the same-layer encoding information of different branches through the feature fusion mechanism, restore the semantic information to representation information step by step through the decoder, and finally complete the classification number mapping, and output the single-scene interference map prediction result output by the model; Intersection and merging decision processing module: the single-scene interference graph prediction results output by the model are processed through intersection and merging decision to produce the final prediction results.

6. A computer device, characterized in that: The invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for intelligently identifying creep-type landslide hazards based on winding interference phase as described in any one of claims 1 to 4 is implemented.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the method for intelligently identifying creep-type landslide hazards based on winding interference phase as described in any one of claims 1 to 4.

Images