Embedded multi-channel spectral-topographic feature fusion coseismic landslide detection method

Abstract

An embodiment of the present invention discloses an embedded multi-channel spectral-topography feature fusion method for coseismic landslide detection. The method comprises: obtaining high-resolution remote sensing image data and satellite stereo image pair data from an earthquake-affected region; processing the satellite stereo image pair data using a multi-class image pair combined DSM extraction method and a median synthesis filter method to generate terrain data; fusing the high-resolution remote sensing image data and terrain information data through a convolution operation; and analyzing the fused data using a deep learning-based wide-area coseismic landslide semantic segmentation detection model to predict coseismic landslides in the earthquake-affected region. This embodiment fuses multi-source remote sensing image data and terrain data to detect landslides in the earthquake-affected region after an earthquake.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of coseismic landslide detection, and in particular to an embedded multi-channel spectral-topographic feature fusion coseismic landslide detection method. Background Art

[0002] A coseismic landslide is a landslide caused by an earthquake. After an earthquake, there is a high probability of coseismic landslides occurring in the affected area. Therefore, it is essential to conduct coseismic landslide detection in the affected area to predict the locations of possible landslides.

[0003] At present, coseismic landslide detection is mainly based on single remote sensing image data as the data source for human-computer interactive interpretation and analysis, which only uses low- and medium-level features such as spectrum, texture, and structure, which is prone to misjudgment of landslide detection. Summary of the Invention

[0004] In view of the shortcomings of the existing technology, the embodiments of the present invention provide an embedded multi-channel spectral-topographic feature fusion co-seismic landslide detection method.

[0005] In the first aspect, the present invention provides an embedded multi-channel spectral-topographic feature fusion coseismic landslide detection method:

[0006] Obtain high-resolution remote sensing image data and satellite stereo image pair data in earthquake-affected areas;

[0007] Processing the satellite stereo image pair data using a multi-image pair combination DSM extraction method and a median synthesis filtering method to generate terrain data information;

[0008] Performing feature fusion on the high-resolution remote sensing image data and the terrain information data, wherein the high-resolution remote sensing image data includes three channels of RGB, and the terrain information data includes three channels of digital elevation model data, slope data, and aspect data;

[0009] The feature fused data is analyzed using a wide-area co-seismic landslide semantic segmentation detection model based on deep learning to predict co-seismic landslides in the earthquake-prone area.

[0010] In a second aspect, an embodiment of the present invention provides an electronic device, characterized by including:

[0011] one or more processors;

[0012] a memory for storing one or more programs,

[0013] When the one or more programs are executed by the one or more processors, the one or more processors implement the embedded multi-channel spectrum-topography feature fusion co-seismic landslide detection method as described in any one of claims 1-8.

[0014] In a third aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that when the program is executed by a processor, the embedded multi-channel spectral-topography feature fusion coseismic landslide detection method as described in any one of claims 1-8 is implemented.

[0015] After an earthquake occurs, an embodiment of the present invention uses a deep learning-based wide-area co-seismic landslide semantic segmentation detection model to detect landslides in the earthquake-affected area. First, high-resolution remote sensing image data and satellite stereo image pair data are acquired for the earthquake-affected area. The satellite stereo image pair data are processed using a multi-class image pair combination DSM extraction method and a median synthesis filter method to generate terrain data information. The high-resolution remote sensing image data and terrain information data are then subjected to feature fusion through a convolution operation. The fused feature data is then detected using a deep learning-based wide-area co-seismic landslide semantic segmentation detection model to predict the locations of possible landslides in the earthquake-affected area. This method has the characteristics of fast detection speed and high detection accuracy. Furthermore, the co-seismic landslide detection model incorporates geological constraints into the model by utilizing multi-source optical images and digital elevation model data, as well as derived geoscientific data such as slope and aspect. This avoids misjudgments of flat-ground landslides and misjudgments of the main sliding direction of non-landslides, thereby improving the accuracy of co-seismic landslide detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0017] Figure 1 This is a flow chart of an embedded multi-channel spectral-topographic feature fusion coseismic landslide detection method provided by an embodiment of the present invention.

[0018] Figure 2 This is a flow chart of another embedded multi-channel spectral-topographic feature fusion coseismic landslide detection method provided by an embodiment of the present invention.

[0019] Figure 3 It is a structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0020] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention are described clearly and completely below. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort are also within the scope of protection of the present invention.

[0021] In the description of the present invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings and are intended solely to facilitate and simplify the description of the present invention. They are not intended to indicate or imply that the devices or components referred to must have, be constructed, or operate in a specific orientation, and therefore should not be construed as limitations on the present invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0022] In the description of the present invention, it should also be noted that, unless otherwise expressly specified or limited, the terms "mounted," "connected," and "connected" should be understood broadly. For example, they may refer to fixed, detachable, or integral connections; mechanical or electrical connections; direct or indirect connections through an intermediate medium; and internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on the specific circumstances.

[0023] Figure 1 This is a flow chart of an embedded multi-channel spectral-topographic feature fusion co-seismic landslide detection method provided by an embodiment of the present invention. This method is applicable to predicting co-seismic landslides in an earthquake-affected area after an earthquake occurs, and is executed by an electronic device. Figure 1 As shown, the specific steps include:

[0024] S110. Acquire high-resolution remote sensing image data and satellite stereo image pair data of the earthquake-affected area.

[0025] This embodiment uses high-resolution remote sensing image data and satellite stereo image data as data sources. The remote sensing image data is used to provide shallow image feature information, and the satellite stereo image data is used to provide high-level semantic information such as geological features, which together serve as the basis for detecting coseismic landslides.

[0026] Optionally, the high-resolution remote sensing image data and satellite stereo image pair data are obtained through an open source satellite platform. Exemplary high-resolution remote sensing image data are obtained through open source data, and satellite stereo image pair data are obtained using China Resources-3 (ZY-3) series stereo mapping satellite data.

[0027] S120 , processing the satellite stereo image pair data using a multi-image pair combination DSM extraction method and a median synthesis filtering method to generate terrain data information.

[0028] Landslides are influenced by many factors, including objective factors such as topography and geological conditions, as well as the influence of natural and human-induced forces. For coseismic landslides, a hazard associated with earthquakes, the micromorphological conditions of the slope before the landslide occurs are particularly important. In particular, the elevation, slope, and aspect of the natural slope have a significant impact on the triggering of coseismic landslides. Therefore, this step extracts topographic data from satellite stereo imagery to obtain high-level semantic information such as topographic features.

[0029] Specifically, the terrain data includes digital elevation model (DEM) data, slope data, and aspect data. DEM data is a discrete mathematical representation of the topography of the studied area, represented by a three-dimensional directed sequence (Xi, Yi, Zi), where i = 1, 2, ..., n represents the geographic location within the area, and Zi represents the altitude at the geographic location (Xi, Yi). Using offline DEM data to describe terrain variations can cover general terrain information.

[0030] Slope information is a numerical description of the slope of the studied area, recording the slope at each point. Existing technologies for detecting coseismic landslides do not incorporate slope information, making it easy to misidentify flat areas as landslide areas. Including slope information can avoid this misidentification of flat land landslides due to the lack of slope information.

[0031] Slope aspect information is a digital description of the slope direction of the study area, recording the slope direction at each point. Existing technologies for detecting coseismic landslides do not incorporate slope aspect information, making it easy to misidentify non-primary landslide directions as landslides. Including slope aspect information can avoid misidentification of non-primary landslide directions due to the lack of slope aspect information.

[0032] Optionally, the satellite stereo image pair data is processed using a multi-type image pair combination DSM extraction method to generate digital surface model data; the digital surface model data is processed using a median synthesis filtering method to remove surface attachment height information to generate digital elevation model data; and the digital elevation model data is used to generate slope data and aspect data.

[0033] In a specific embodiment, the slope data is calculated based on the formula: Slope = (elevation difference / horizontal distance) × 100%; the downhill direction with the largest rate of change from each pixel to its adjacent pixel value is used as the slope direction to calculate the slope direction data.

[0034] Optionally, after obtaining high-resolution remote sensing image data and terrain information data, these data are normalized based on the following formula to balance the use of image and terrain features that affect the extraction of coseismic landslides:

[0035]

[0036] Among them, Pix Value 、Pix Valuemax 、Pix Valuemin They respectively represent the data value to be normalized, the maximum data value, and the minimum data value in the same type of data.

[0037] S130 , performing feature fusion on the high-resolution remote sensing image data and the terrain information data, wherein the high-resolution remote sensing image data includes three channels of RGB, and the terrain information data includes three channels of digital elevation model data, slope data, and aspect data.

[0038] Optionally, a convolution kernel of a specified size is designed according to the number of feature channels and the model input requirements, and a convolution operation is performed on the normalized multi-channel spectral-topographic dataset formed by a combination of multiple bands such as optical images and geological data to obtain feature fusion data that meets the model's requirements for input channels and size.

[0039] In one embodiment, using Figure 2 The Feature Fusion module shown implements feature fusion. This module performs the following operations: It uses three 3×3 convolution kernels with six channels to perform convolution calculations on the three RGB channels of the high-resolution remote sensing image data and the three channels of the digital elevation model data, slope data, and aspect data of the terrain information data, achieving multi-source data feature fusion. For example, the convolution kernel parameters used are padding = 1 and stride = 1.

[0040] Optionally, before performing the convolution operation, consistent resampling, band registration and combination methods can be used to combine the spectral-topographic feature data bands including the optical remote sensing image and the DEM and its derived data (including slope data and broken line data) to form a feature data set as the object of the convolution operation.

[0041] This step makes full use of the geological knowledge related to the research object. By adding a convolution operation before the deep convolutional neural network, the multi-channel feature data or the multi-channel feature data after band combination is converted into channel and size sample data that meets the input requirements of the deep convolutional neural network, and the effective combination of the pre-convolution operator and the deep convolutional neural network is realized in an embedded manner.

[0042] S140. Analyze the feature-fused data using a wide-area co-seismic landslide semantic segmentation detection model based on deep learning to predict co-seismic landslides in the earthquake-prone area.

[0043] In this step, the feature fused data is input into the trained co-seismic landslide detection model. Optionally, the co-seismic landslide detection model is built based on DeepLab V3+. Figure 2 The co-seismic landslide detection model is composed of an encoder and a decoder, wherein the encoder is used to extract terrain features, including a residual network and a dilated convolution, and the decoder is used to restore target boundary details by interpolation. Optionally, the residual network uses a ResNet50 network, which serves as the backbone network of the co-seismic landslide detection model and can maintain a relatively fast model detection speed and a relatively high model detection accuracy. The output of the co-seismic landslide detection model is the area where the landslide occurs. In a specific embodiment, the output detection result is a grayscale image, in which the area with a pixel value of 0 represents the area where the landslide is predicted to occur, and the area with a pixel value of 1 represents the area where the landslide is predicted not to occur.

[0044] Specifically, the main body of DeepLab V3+'s Encoder is composed of a spatial pyramid pooling module (Atrous Spatial Pyramid Pooling, ASPP) with dilated convolution and a backbone network. Among them, dilated convolution (AtrousConvolution) is one of the keys to the model. Dilated convolution (AtrousConvolution) can reduce the downsampling rate while ensuring the receptive field. The final feature map obtained is not only semantically rich but also relatively fine, and the original resolution can be directly restored by interpolation. The ASPP module replaces the downsampling layer with dilated convolution by modifying the Block behind the backbone network, and uses dilated convolutions of different rates to control the receptive field without changing the size of the feature image, so as to extract multi-scale information. ASPP uses multiple parallel dilated convolutions, combined with image-level features (ie, global average pooling). By Figure 2As can be seen, the ASPP module mainly consists of a 1*1 convolutional layer, three 3*3 dilated convolutions, and an ImagePooling layer. The convolutional layer can extract local features, and the ImagePooling layer can extract global features. The concatenation method is then used to fuse the five features of different scales. After 1*1 convolution, high-level semantic features are obtained. The decoder structure is relatively simple. It first upsamples the features obtained by the encoder through bilinear interpolation to obtain 4 times the features. Then, it fuses the features with the low-level features of the corresponding size in the encoder through the concatenation method. Subsequently, 3*3 convolution is used to further fuse the features, and finally bilinear interpolation is used to obtain the segmentation prediction of the same size as the original image. All upsampling in the decoder uses the bilinear interpolation method.

[0045] After an earthquake occurs, an embodiment of the present invention uses a feature-fusion coseismic landslide detection model built based on DeepLab V3+ to detect landslides in the earthquake-affected area. First, high-resolution remote sensing imagery and satellite stereo image pair data are acquired from the earthquake-affected area. The satellite stereo image pair data is processed using a multi-class image pair combined DSM extraction method and a median synthesis filter to generate terrain data. The high-resolution remote sensing imagery and terrain data are then subjected to feature fusion through a convolution operation. The fused data is then used to detect landslides in the earthquake-affected area using a feature-fusion coseismic landslide model built based on DeepLab V3+. This method, implemented as a coseismic landslide detection model built based on DeepLab V3+ and using ResNet50 as its backbone network, offers fast detection speed and high accuracy. Furthermore, the method incorporates geoscientific constraints into the coseismic landslide detection model by leveraging multi-source optical imagery and digital elevation model data, along with derived geoscientific data such as slope and aspect. This prevents misjudgments of flat-surface landslides and misjudgments of non-landslide main directions, thereby improving the accuracy of coseismic landslide detection.

[0046] More specifically, the embodiments of the present invention focus on solving two problems: multi-source feature data fusion and high-level semantic feature learning. First, by utilizing data normalization and convolution operations, the optical remote sensing image features are fused with the micro-geomorphological features and geological knowledge data features that are closely related to co-seismic landslides. Second, based on the DeepLab V3+ semantic segmentation network, the advantages of the spatial pyramid module and the encoder-decoder structure are comprehensively utilized. On the one hand, the spatial pyramid module applies multi-sampling rate dilated convolution, multi-receptive field convolution or pooling to the input features to explore multi-scale contextual information. On the other hand, by introducing the decoder module, it further integrates the underlying detail features with the high-level semantic features. The encoder-decoder structure gradually recovers the spatial information to capture clearer target boundaries, thereby improving the accuracy of the segmentation boundaries.

[0047] Optionally, the training phase of the coseismic landslide detection model also includes acquiring high-resolution remote sensing imagery and satellite stereo image pair data for the earthquake-affected region, and using a cloud-free image synthesis algorithm and multi-temporal phase comparison method to generate cloud-free optical remote sensing imagery showing landslide changes before and after the earthquake as training data. It should be noted that the earthquake-affected region during the training phase can be the same as or different from the earthquake-affected region during the prediction phase; there is no necessary connection between the two.

[0048] The cloud-free image synthesis algorithm uses an optical remote sensing data cloud-free image synthesis algorithm to generate cloud-free image data, removing the impact of cloud obstruction on the training of the co-seismic landslide detection model. The multi-temporal comparison method uses a method of comparing high-resolution remote sensing image data before and after the earthquake to determine whether the landslide was triggered by the earthquake. If the landslide exists in the pre-earthquake image and its morphology does not change in the post-earthquake image, then the landslide is considered to be a landslide that existed before the earthquake. If the landslide does not exist in the pre-earthquake image, or exists in the pre-earthquake image but its morphology changes in the post-earthquake image, then the landslide is considered to be an earthquake-triggered landslide, i.e., a co-seismic landslide. Data showing changes in morphology between pre-earthquake and post-earthquake images is used as training data for the co-seismic landslide detection model.

[0049] Optionally, the data after the feature fusion is divided into a training set, a test set and a validation set in a ratio of 6:3:1 by a random method, and the accuracy evaluation indicators such as Precision, mIou, and F1 score are used to compare and analyze the results to optimize the iterative model parameters.

[0050] In one specific implementation, using the 2017 Jiuzhaigou Ms7.0 earthquake as an example, and using the Jiuzhaigou earthquake coseismic landslide catalog as a sample, this method, an embedded multi-channel spectral-topographic feature fusion coseismic landslide detection method, was used to perform wide-area coseismic landslide extraction. This method, using multi-source optical remote sensing imagery, digital elevation models (DEMs), and derived feature data reflecting microtopographic conditions, such as slope and aspect, was employed. Experimental results showed that this method completed coseismic landslide extraction over an area of ​​142.54 square kilometers in just 1.35 seconds, achieving a precision of 0.796203, a recall of 0.672002, and an F1 score of 0.728404. Compared to U-Net, DeepLab V3+, and traditional machine learning methods that utilize spectral features alone, this method improved the accuracy of wide-area coseismic landslide extraction by 5% to 13%. This shows that this technical solution has achieved, to a certain extent, the rapid and accurate extraction of wide-area co-seismic landslides under the "large background and small target" situation, greatly reducing labor and time costs while significantly improving accuracy.

[0051] In summary, the method of the present application generally provides a deep learning model that integrates geological knowledge. As a multi-channel feature data-driven method, its essence is a mathematical process that automatically obtains high-level function features through multiple low-level features. Compared with a single nonlinear function, the nonlinear composite process of multiple functions can fit more complex functional correlations than manual fitting, and extract more effective information related to landslide information extraction from low-value density data, that is, deep-level abstract features, without the need for manual design of high-level features, thereby greatly reducing the cost of manual errors in experiments and achieving efficient and accurate identification of wide-area co-seismic landslides under complex background conditions.

[0052] Figure 3 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention is shown in FIG. Figure 3 As shown, the device includes a processor 50, a memory 51, an input device 52 and an output device 53; the number of processors 50 in the device can be one or more. Figure 3 In the embodiment, a processor 50 is used as an example; the processor 50, the memory 51, the input device 52 and the output device 53 in the device can be connected by a bus or other means. Figure 3 The bus connection is taken as an example.

[0053] Memory 51, as a computer-readable storage medium, can be used to store software programs, computer-executable programs, and modules, such as the program instructions / modules corresponding to the embedded multi-channel spectral-topographic feature fusion coseismic landslide detection method in the embodiments of the present invention. Processor 50 executes the software programs, instructions, and modules stored in memory 51 to perform various functional applications and data processing of the device, thereby implementing the embedded multi-channel spectral-topographic feature fusion coseismic landslide detection method described above.

[0054] The memory 51 may primarily include a program storage area and a data storage area. The program storage area may store an operating system and at least one application required for a function; the data storage area may store data generated based on the use of the terminal, etc. Furthermore, the memory 51 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state memory device. In some instances, the memory 51 may further include memory remotely located relative to the processor 50, and these remote memories may be connected to the device via a network. Examples of the aforementioned networks include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0055] The input device 52 may be used to receive input digital or character information and generate key signal input related to user settings and function control of the device. The output device 53 may include a display device such as a display screen.

[0056] An embodiment of the present invention further provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the embedded multi-channel spectral-topographic feature fusion coseismic landslide detection method of any embodiment.

[0057] The computer storage medium of the embodiment of the present invention can adopt any combination of one or more computer-readable media. Computer-readable media can be computer-readable signal media or computer-readable storage media. Computer-readable storage media can be, for example, but not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or components, or any combination thereof. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by an instruction execution system, device or device or used in combination with it.

[0058] A computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0059] The program code included in the computer-readable medium can be transmitted with any appropriate medium, including but not limited to wireless, electric wire, optical cable, RF, etc., or any suitable combination thereof. The computer program code for performing the operation of the present invention can be written in one or more programming languages ​​or a combination thereof, and the programming language includes an object-oriented programming language such as Java, Smalltalk, C++, and also includes a conventional procedural programming language such as "C" language or similar programming language. The program code can be executed entirely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer and partially on a remote computer, or completely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (e.g., using an Internet service provider to connect via the Internet).

[0060] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the technical solutions of the embodiments of the present invention.

Claims

1. An embedded multi-channel spectral-topographic feature fusion coseismic landslide detection method, characterized in that: include: Obtain high-resolution remote sensing image data and satellite stereo image pair data in earthquake-affected areas; The satellite stereo image pair data are processed using a multi-image pair combination DSM extraction method to generate digital surface model data; the digital surface model data are processed using a median synthesis filtering method to remove surface appendage height information to generate digital elevation model data; the digital elevation model data are used to generate slope data and aspect data; and terrain data information is formed from the digital elevation model data, slope data, and broken line data; Using three convolution kernels of size 3×3 and number of channels 6, convolution calculation is performed on the three RGB channels of the high-resolution remote sensing image data and the three channels of digital elevation model data, slope data and aspect data of the terrain information data, a total of six channels, to achieve data feature fusion; The feature fusion data is analyzed using a wide-area co-seismic landslide semantic segmentation detection model based on deep learning to predict co-seismic landslides in the earthquake-prone area. The wide-area co-seismic landslide semantic segmentation detection model is built based on DeepLab V3+ and includes two parts: an encoder and a decoder. The encoder is used to extract terrain features, including a residual network and a dilated convolution. The decoder is used to restore target boundary details through interpolation.

2. The method according to claim 1, characterized in that The digital elevation model data in the terrain data information can cover general terrain information, the slope data can eliminate misjudgment of flat landslide detection, and the slope direction data can eliminate misjudgment of non-landslide main directions.

3. The method according to claim 1, characterized in that The step of generating slope data and aspect data by using the digital elevation model data includes: Slope data is calculated based on the formula: Slope = (elevation difference / horizontal distance) × 100%; The downslope direction with the largest rate of change from each pixel to its adjacent pixel value is taken as the aspect to calculate the aspect data.

4. The method according to claim 1, wherein Before the feature fusion of the high-resolution remote sensing image data and the terrain information data, the method further includes: performing normalization processing on the high-resolution remote sensing image data and the terrain information data based on the following formula to balance the use of image and terrain features that affect the coseismic landslide extraction effect: Among them, Pix Value 、Pix Valuemax 、Pix Valuemin They respectively represent the data value to be normalized, the maximum data value, and the minimum data value in the same type of data.

5. The method according to claim 1, wherein The residual network uses the ResNet50 network, which is used as the backbone network of the co-seismic landslide detection model to maintain a faster model detection speed and higher model detection accuracy.

6. An electronic device, characterized in that: include: one or more processors; a memory for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the embedded multi-channel spectrum-topography feature fusion co-seismic landslide detection method as described in any one of claims 1-5.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the embedded multi-channel spectrum-topography feature fusion coseismic landslide detection method as described in any one of claims 1 to 5 is implemented.

Images