CN113610070A - Landslide disaster identification method based on multi-source data fusion - Google Patents
Landslide disaster identification method based on multi-source data fusion Download PDFInfo
- Publication number
- CN113610070A CN113610070A CN202111179257.4A CN202111179257A CN113610070A CN 113610070 A CN113610070 A CN 113610070A CN 202111179257 A CN202111179257 A CN 202111179257A CN 113610070 A CN113610070 A CN 113610070A
- Authority
- CN
- China
- Prior art keywords
- data
- landslide
- image
- remote sensing
- neural network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a landslide disaster identification method based on multi-source data fusion, which belongs to the technical field of intelligent geological disaster identification, and is characterized in that multi-source data are used as input, data are preprocessed according to characteristic attributes of the multi-source data, and a data consistency relation is constructed; and then, identifying by using a pre-trained landslide hazard identification network, and superposing an output result with the original image, thereby realizing the landslide hazard range identification at the pixel level. Practice proves that the method has a good recognition effect on landslide disasters, and the recognition accuracy can reach 75% on the basis of the existing data set. According to the method, multi-source heterogeneous data is adopted, all influence factors and judgment and identification modes of landslide disasters are comprehensively considered, a landslide disaster identification network is established in a targeted manner, fusion solving is carried out on three dimensions of data, characteristics and the network, pixel-level landslide prediction is achieved, and a method is provided for relevant researchers.
Description
Technical Field
The invention belongs to the technical field of geological disaster identification, relates to cross-field combination of deep learning and geological fields, and particularly relates to a landslide disaster identification method based on multi-source data fusion.
Background
The geological disaster of major landslide in China has three characteristics: the dots are many-sided and wide, and the concealment is strong; high speed and long distance, and great harm; the human cannot reach the target and is difficult to observe. The traditional landslide detection method needs to consume a large amount of manpower and material resources, has the defects of difficult data acquisition, complex information processing, untimely disaster early warning and the like, and is low in applicability. With the development of the times, the leading edge technology brings new possibility for the detection of high-order remote geological disasters.
With the rapid development of the aerospace remote sensing technology, the method is widely applied to monitoring, investigation and evaluation of various geological disasters. By processing and analyzing the information in the remote sensing image, the type attribute, the spatial distribution characteristic, the space-time transformation characteristic and the like of the ground feature can be distinguished.
On the other hand, due to the complex formation mechanism, evolution process and influence factors of the landslide disaster, the recognition and positioning of the landslide disaster only by means of single optical image data have certain limitations. The existing data resources are fully utilized, the knowledge base data of each region and each platform are integrated, and the method is a powerful data support for landslide disaster identification.
With the application of artificial intelligence technology represented by deep learning in the visual field, great progress is made in a plurality of fields, and the realization of feature extraction by a deep neural network with deep structure and most significant parameters is one of the most popular directions at present. The deep neural network is an efficient method for realizing intelligent recognition of landslide disasters.
At present, the high-order remote landslide disaster identification method based on the remote sensing technology needs to extract the characteristics of textures, geometry and the like from ground objects, a characteristic space is constructed through the textures, shapes, hues and the like of landslides on high-resolution images, and when the image characteristics of the landslide disaster in a research area are not obvious, the landslide identification precision is influenced to a certain extent.
Chinese patent No. CN111898419A discloses a system and method for detecting landslide in different zones based on a cascaded deep convolutional neural network, which combines artificial intelligence technology with landslide hazard identification, divides the territory into four zones according to the region where landslide hazard is likely to occur or frequently occurs, and generates different detection network models for different zones. However, in the training of the landslide detection network model, only images shot by the unmanned aerial vehicle are used, but influence factors of landslide formation are various, and influence of factors such as geology and rivers is not considered in the invention. Therefore, various possible influence factors need to be comprehensively considered, and a stable and reliable algorithm network model based on multi-source data is established.
Disclosure of Invention
The main problems to be solved by the invention are as follows: aiming at the problems of insufficient utilization of existing landslide disaster data, limited training samples, poor segmentation identification precision and the like, the improvement is realized based on a deep learning semantic segmentation method, a data consistency relation is established for multi-source heterogeneous data from relevant influence factors and an identification mode formed by landslide disasters, the landslide disaster feature extraction and identification based on the multi-source data are realized, the pixel-level landslide disaster identification precision is improved, and a technical means of normalized monitoring is provided for landslide disaster monitoring.
The technical scheme of the invention is as follows: a landslide disaster identification method based on multi-source data fusion is characterized in that multi-source data are used as input, data are preprocessed according to characteristic attributes of the multi-source data, and a data consistency relation is constructed; and then, identifying by using a pre-trained landslide hazard identification network, and superposing an output result with the original image, thereby realizing the landslide hazard range identification at the pixel level. Practice proves that the method has a good recognition effect on landslide disasters, and the recognition accuracy can reach 75% on the basis of the existing data set. According to the method, multi-source heterogeneous data is adopted, all influence factors and judgment and identification modes of landslide disasters are comprehensively considered, a landslide disaster identification network is established in a targeted manner, fusion solving is carried out on three dimensions of data, characteristics and the network, pixel-level landslide prediction is achieved, and a method is provided for relevant researchers.
The method is realized by the following steps:
(1) acquiring multi-source heterogeneous data influencing landslide and analysis thereof, wherein the multi-source heterogeneous data comprises optical remote sensing image data (TIF image) and a labeling image (PNG image) thereof, geological data (SHP grid format), an elevation map (TIF image), a river distribution map (SHP grid format), a landform map (TIF image) and a movable fracture zone (SHP grid format);
(2) preprocessing the multi-source heterogeneous data in the step (1), constructing a data consistency relation, standardizing and unifying the data, and realizing geographic position matching and ensuring the same image resolution by taking an optical remote sensing image as a reference;
(3) identifying the multi-source heterogeneous data processed in the step (2) by using a pre-trained landslide disaster identification neural network to obtain a prediction result (a binary image indicating the occurrence range of landslides, wherein each pixel point in the image is a preset landslide category value);
(4) and (3) superposing the prediction result in the step (2) with the original optical remote sensing image by using an OpenCV algorithm library to obtain a final landslide disaster prediction result, so that landslide disaster recognition is realized.
Preprocessing multi-source heterogeneous data in the step (2) and constructing a data consistency relation, and specifically comprises the following steps:
s1, cutting the optical remote sensing image to remove the occlusion;
s2, processing the raster format file in (1), mapping the region division factor of the raster data to a numerical value, and performing attribute transformation to obtain a label image composed of numerical values in the range of 0 to 255, specifically, the method includes the following steps:
1) selecting the most characteristic feature items of the raster format files of different data sources, and marking the feature items as identification items;
2) acquiring all possible values of the identification items, and respectively assigning unique numbers;
3) adding a numerical characteristic column into the raster file, wherein the value of the numerical characteristic column is the number corresponding to the identification item;
s3, carrying out geometric correction processing on the optical remote sensing image to eliminate image distortion generated in the image acquisition process, so that various ground objects on the image have accurate geographical positions and meet the set plane precision requirement;
s4, converting the raster data processed by the S2, the bitmap data in the label image format and the optical remote sensing image data processed by the S1 and the S3 into the same geographic longitude and latitude coordinate system to realize geographic position matching;
and S5, processing the raster data into a bitmap format, enabling the bitmap format to be matched with the geographic range of the optical remote sensing image data, and realizing the associative correspondence between multisource heterogeneous landslide hazard data even if the pixel points of the raster data and the optical remote sensing image data are in one-to-one correspondence.
In the step (3), the training of the landslide hazard recognition neural network comprises the following steps:
1) processing multi-source heterogeneous data according to the steps (1) and (2) to obtain input samples suitable for landslide disaster recognition neural network, and recording the input samples asX i ；
2) Constructing a deep convolutional neural network comprising convolutional layer, pooling layer and pyramid pooling module with holes (ASPP), wherein the input of the network is the one in the step 1)X i The output of the network is a binary map (indicating the occurrence range of the landslide);
3) identifying and modeling landslide as a semantic segmentation problem, using the convolutional neural network to identify and solve, solving the gradient of the convolutional neural network through a loss function so as to obtain a back propagation residual error, and then continuously updating a weight parameter;
4) by trainingSample(s)X i Training the weight parameters of the deep convolutional neural network, and obtaining multilayer convolutional kernel parameters after the convolutional neural network is stable after a certain number of iterations, thereby completing the training of the network.
In the step (3), the network structure design of the landslide hazard recognition neural network fully considers each influence factor and the discrimination mode of the landslide hazard, and the three dimensions of data, characteristics and a network model are fused, so that the pixel-level landslide prediction is realized, and the method specifically comprises the following modules:
1) the convolutional neural network module ResNet34 is used for learning landslide disaster influence characteristics shown by the non-optical remote sensing image data, including altitude and the like; where the input to ResNet34 isX i After convolution, the feature vector is output as the feature vector representing the influence characteristics of the landslide disaster and is recorded asS(X i )；
2) The convolutional neural network module ResNet50 is used for learning the obvious visual characteristics of the landslide disaster expressed in the optical remote sensing image data, including the edge, the geometric shape and the like of the landslide; where the input to ResNet50 isX i After convolution, the feature vector is output as a feature vector representing the significant visual features of the landslide hazard and is recorded asC(X i )；
3) The system comprises a holey spatial pyramid pooling module (ASPP), which comprises five parts: one convolutional layer with convolution kernel of 1x1, three void convolutional layers with convolution kernel of 3x3, one average pooling layer, followed by one convolutional layer with convolution kernel of 1x 1; the hole convolution adds a hole coefficient on the basis of the traditional discrete convolutionlThe calculation formula is
In the ASPP module, each convolution layer is followed by a BatchNorm layer; as shown in fig. 6, after the fifth part, i.e. the averaging pooling layer and the convolution layer, the output feature vector changes, and it is necessary to expand the feature by using the bilinear interpolation method, and then the feature vector is compared with the other convolution layers in the ASPP shown in fig. 6Splicing the processed features together by using a Concatenate method to form a multi-scale mixed feature, then accessing the multi-scale mixed feature into a convolution layer with a convolution kernel of 1x1, and outputting to obtain a feature vector with 256 dimensions; the ASPP convolves and samples the given input in parallel with holes of different sampling rates, which is equivalent to capturing the context of the image at multiple scales, and further extracts multi-scale landslide information.
The principle of the invention is as follows: by combining a deep learning and computer vision method, multi-source data such as optical remote sensing image data, digital elevation data, landform partition data, geological partition data, hydrological distribution data, fracture zone distribution data and the like are adopted, domain expert knowledge is introduced, landslide hazard influence factors and an identification mode are considered, fusion solving is carried out on three dimensions of data, characteristics and a network model, and therefore pixel-level landslide prediction is achieved, and the method for intelligently identifying landslide hazards is provided.
Compared with the prior art, the invention has the advantages that:
(1) firstly, multi-source heterogeneous data is adopted, different from other single-source data network models, optical remote sensing image data, digital elevation data, national landform regional data, national geological regional data, national hydrological distribution data, national fracture zone distribution data and other multi-source data are fused, and various influence factors and judgment and identification modes of landslide disasters are comprehensively considered;
(2) secondly, according to the respective characteristic attributes of the multi-source heterogeneous data, different convolutional neural network modules are set up in a targeted manner, so that the characteristics of different data can be extracted more fully;
the invention realizes the pixel-level landslide prediction by fusion solving on three dimensions of data, characteristics and a network model, and provides a method for related researchers.
Drawings
FIG. 1 is a schematic diagram of a process implementation of the present invention;
FIG. 2 is a schematic diagram showing comparison between a remote sensing image before and after image cropping; (a) remote sensing images with cloud layer shielding, (b) remote sensing images with cloud layer shielding removed;
FIG. 3 is a schematic view of a geometry calibration;
FIG. 4 is a ResNet34 network architecture diagram;
FIG. 5 is a ResNet50 network architecture diagram;
FIG. 6 is a schematic diagram of an ASPP module;
FIG. 7 is a schematic diagram of a general architecture of a landslide hazard identification network;
fig. 8 is a schematic diagram of a landslide disaster recognition result, wherein an original remote sensing image (a), a labeled image (b), and a network model prediction image are superimposed on an original image to obtain an effect (c).
Detailed Description
The technical solutions in the embodiments of the present invention are clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
As shown in fig. 1, the landslide disaster identification method based on multi-source data fusion of the present invention includes the following steps:
(1) acquiring multi-source heterogeneous data influencing landslide and analysis thereof, wherein the multi-source heterogeneous data comprises optical remote sensing image data (TIF image) and a labeling image (PNG image) thereof, geological data (SHP grid format), an elevation map (TIF image), a river distribution map (SHP grid format), a landform map (TIF image) and a movable fracture zone (SHP grid format);
(2) preprocessing the multi-source heterogeneous data in the step (1), constructing a data consistency relation, standardizing and unifying the data, and using an optical remote sensing image as a reference to realize geographic position matching and ensure that the image resolutions are the same, specifically comprising the following steps:
s1, as shown in figure 2, cutting the optical remote sensing image to remove the shielding therein, wherein (a) the remote sensing image with cloud layer shielding, (b) the remote sensing image with cloud layer shielding is removed;
s2, processing the raster format file in the step (1), mapping the region division factor of raster data into numerical values, and performing attribute conversion to obtain a marked image formed by the numerical values in the range of 0-255; specifically, the method comprises the following steps:
1) selecting the most characteristic feature items of the raster format files of different data sources, and marking the feature items as identification items;
2) acquiring all possible values of the identification items, and respectively assigning unique numbers;
3) adding a numerical characteristic column into the raster file, wherein the value of the numerical characteristic column is the number corresponding to the identification item;
s3, as shown in figure 3, geometric correction processing is carried out on the optical remote sensing image to eliminate image distortion generated in the image acquisition process, so that various ground objects on the image have accurate geographical positions and meet the set plane precision requirement;
s4, converting the raster data processed by the S2, the bitmap data in the format of the marked image and the optical remote sensing image data processed by the S1 and the S3 into the same geographic longitude and latitude coordinate system to realize geographic position matching, and specifically, the method comprises the following steps:
1) reading the coordinate system parameters of the optical remote sensing image by using a rasterio algorithm library;
2) using geopands geoseries to crs () function to grid the grid in S1 according to the coordinate system parameters
Projecting the data to the same coordinate system so as to realize geographic position matching;
s5, reading resolution information of the optical remote sensing image by using a rasterio algorithm library, and then calling a features.rasterize () function in rasterio to map the raster data, wherein the resolution setting is consistent with the optical remote sensing image, so that the raster data is matched with the optical remote sensing image data in geographic range, even if the pixel points of the raster data and the optical remote sensing image are in one-to-one correspondence, the correlation correspondence between multisource heterogeneous landslide hazard data is realized.
(3) Identifying the multi-source heterogeneous data processed in the step (2) by using a pre-trained landslide hazard identification neural network to obtain a prediction result (a binary image indicating the occurrence range of landslides);
(4) and (3) superposing the prediction result in the step (2) with the original optical remote sensing image by using an OpenCV algorithm library to obtain a final landslide disaster prediction result, so that landslide disaster recognition is realized.
As shown in fig. 4, the convolutional neural network module ResNet34 is used for learning landslide disaster influence features, including altitude and the like, represented by non-optical remote sensing image data, wherein the features cannot be represented in two-dimensional optical remote sensing image data, but have a significant influence on landslide identification; where the input to ResNet34 isX i After convolution, the feature vector is output as the feature vector representing the influence characteristics of the landslide disaster and is recorded asS(X i ) (ii) a Specifically, the resolution of the input image of ResNet34 is 224 × 224, and the input image is downsampled by a convolutional layer (convolution kernel is 7 × 7, step size is 2) to obtain a feature vector with the resolution of 112 × 112; the resolution is reduced and the complexity of calculation is reduced through a maximum pooling layer; and then, four groups of different residual blocks (the number of the residual blocks in each group is 3, 4, 6 and 3 respectively, and the input and output dimensions of the residual blocks in the same group are the same, namely 64, 128, 256 and 512 respectively) are processed to sequentially obtain the feature vectors with the sizes of 56 × 56, 28 × 28, 14 × 14 and 7 × 7.
As shown in fig. 5, the convolutional neural network module ResNet50 is configured to learn the significant visual features of the landslide disaster, which are shown in the optical remote sensing image data, including the edge and the geometric shape of the landslide; where the input to ResNet50 isX i After convolution, the feature vector is output as a feature vector representing the significant visual features of the landslide hazard and is recorded asC(X i ) (ii) a Specifically, the resolution of the input image of the ResNet50 is 224 × 224, and the input image is downsampled by a convolutional layer (convolution kernel is 7 × 7, step size is 2) to obtain a feature map with the resolution of 112 × 112; then passing through a maximum pooling layer; and then four groups of residual blocks are passed (the number of the residual blocks in each group is 3, 4, 6 and 3 respectively, and the output dimensions are 256, 512, 1024 and 2048).
As shown in fig. 6, the holed spatial pyramid pooling module (ASPP) includes five parts: one convolution kernel is a convolution layer with 1x1, and three convolution kernels are hollow convolutions with 3x3Layer, an average pooling layer followed by a convolution layer with a convolution kernel of 1x 1; the hole convolution adds a hole coefficient on the basis of the traditional discrete convolutionlThe calculation formula is
In the ASPP module, each convolution layer is followed by a BatchNorm layer; as shown in fig. 6, after the fifth part, that is, the average pooling layer and the convolutional layer are processed, the output feature vector changes, a bilinear interpolation method is required to expand the feature, then the feature is spliced with the features processed by the other convolutional layers in the ASPP shown in fig. 6 by using a Concatenate method to form a multi-scale mixed feature, and then the multi-scale mixed feature is accessed to the convolutional layer with a convolutional kernel of 1 × 1, and the feature vector with 256 dimensions is output; the ASPP convolves and samples the given input in parallel with holes of different sampling rates, which is equivalent to capturing the context of the image at multiple scales, and further extracts multi-scale landslide information.
As shown in fig. 7, the general architecture diagram of the landslide disaster identification network is shown, the network includes convolutional neural network modules ResNet34 and ResNet50, and a spatial pyramid pooling with holes module (ASPP), where the ASPP includes a plurality of convolutional kernels with different rates, the processed multi-source heterogeneous data is input into the landslide disaster identification network, then different features are extracted according to different convolutional neural network modules, and then the features are fused to obtain a fused feature vector, specifically, the fusion of the features is performed in the following manner:
wherein the content of the first and second substances, X i represents the input of the network obtained by the multi-source heterogeneous data after the data consistency relationship construction processing,C(X i ) To representX i The characteristic vector obtained after the ResNet50 module, S (S: (S))X i ) To representX i Through ResNet34 module to obtainThe feature vector of the received image is obtained,Irepresenting the feature vector obtained after fusion; the fused feature vector has better feature characterization capability. The fused feature vector is subjected to ASPP processing and then upsampled to recover detail information of the image lost by the downsampling operation of convolution, where the upsampling is linear interpolation. Restoring the image information to the original size after the upsampling processing, reconstructing the image information, and outputting the predicted value of each pixel point in the obtained landslide information to be a preset landslide category value;
part of experimental parameter settings:
setting during landslide hazard recognition network training (an optimization algorithm is SGD, an initial learning rate is 0.01, momentum is 0.9, evaluation indexes are selected IOU, Pixel Accuracy and F1-Score, wherein the IOU is used for measuring the overlapping rate of a predicted candidate frame and an artificial mark frame, the Pixel Accuracy is used for calculating the proportion of a predicted correct Pixel to a total Pixel, and the F1-Score is a harmonic mean of a network model Accuracy rate and a recall rate). As a result of the experiments shown in Table 1 below, the process of the present invention (Ours) exhibited about a 30% improvement in IOU performance, about a 14% improvement in Pixel Accuracy performance, and about a 34% improvement in F1-Score performance over the process described in the Table.
Table 1: comparison of Experimental results
Fig. 8 is a schematic diagram of a landslide disaster recognition result, where (a) is a remote sensing image, (b) is a corresponding labeled image, (c) is an effect of superimposing a network model prediction image and an original image, and the method can realize pixel-level landslide prediction by fusing three dimensions of data, features and a network model, and provides an intelligent recognition method of landslide disaster position and shape range, so as to provide reference for related researchers;
the above examples are provided only for the purpose of describing the present invention, and are not intended to limit the scope of the present invention. The scope of the invention is defined by the appended claims. Various equivalent substitutions and modifications can be made without departing from the spirit and principles of the invention, and are intended to be within the scope of the invention.
Claims (4)
1. A landslide disaster identification method based on multi-source data fusion is characterized by comprising the following steps:
(1) acquiring multi-source heterogeneous data influencing landslide and analysis thereof, wherein the multi-source heterogeneous data comprises optical remote sensing image data of a TIF (triangulated irregular field) image, an optical remote sensing image data annotation image of a PNG (portable network generator) image, geological data of an SHP (strapdown transform) grid format, an elevation map of the TIF image, a river distribution map of the SHP grid format, a landform map of the TIF image and a movable fracture zone of the SHP grid format;
(2) preprocessing the multi-source heterogeneous data in the step (1), constructing a data consistency relation, standardizing and unifying the data, and realizing geographic position matching and ensuring the same image resolution by taking an optical remote sensing image as a reference;
(3) identifying the multi-source heterogeneous data preprocessed in the step (2) by using a pre-trained landslide disaster identification neural network to obtain a prediction result, and indicating a binary image of a landslide occurrence range;
(4) and (3) superposing the prediction result in the step (2) with the original optical remote sensing image by using an OpenCV (open vehicle vision correction) algorithm library to obtain a final landslide disaster prediction result, so that landslide disaster recognition is realized.
2. The landslide disaster identification method based on multi-source data fusion of claim 1, wherein in the step (2), pre-processing and data consistency relationship construction of multi-source heterogeneous data comprises the following steps:
s1, cutting the optical remote sensing image to remove the occlusion;
s2, processing the geological data in the SHP grid format, the river distribution map in the SHP grid format and the active fracture zone in the SHP grid format in the step (1), mapping the region division factor of the grid data into numerical values, and performing attribute conversion to obtain a marked image formed by the numerical values in the range of 0-255, specifically, the method comprises the following steps:
(11) selecting the most characteristic feature items of the raster format files of different data sources, and marking the feature items as identification items;
(12) acquiring all possible values of the identification items, and respectively assigning unique numbers;
(13) adding a numerical characteristic column into the raster file, wherein the value of the numerical characteristic column is the number corresponding to the identification item;
s3, carrying out geometric correction processing on the optical remote sensing image to eliminate image distortion generated in the image acquisition process, so that various ground objects on the image have accurate geographical positions and meet the set plane precision requirement;
s4, converting the raster data processed in the S2, the bitmap data in the label image format and the optical remote sensing image data processed in the steps S1 and S3 into the same geographic longitude and latitude coordinate system to realize geographic position matching;
and S5, processing the raster data into a bitmap format, enabling the bitmap format to be matched with the geographic range of the optical remote sensing image data, and realizing the associative correspondence between multisource heterogeneous landslide hazard data even if the pixel points of the raster data and the optical remote sensing image data are in one-to-one correspondence.
3. The landslide disaster identification method based on multi-source data fusion according to claim 1, wherein the landslide disaster identification method comprises the following steps: in the step (3), the training of the landslide hazard recognition neural network comprises the following steps:
(21) processing multi-source heterogeneous data according to the steps (1) and (2) to obtain input samples suitable for landslide disaster recognition neural network, and recording the input samples asX i ；
(22) Constructing a deep convolutional neural network comprising convolutional layer, pooling layer and space pyramid pooling module (ASPP) with holes, wherein the input of the network isX i The output of the network is a binary image in a PNG format, which indicates the occurrence range of the landslide;
(23) identifying and modeling landslide as a semantic segmentation problem, using the convolutional neural network to identify and solve, and solving the gradient of the convolutional neural network through a loss function to obtain a back propagation residual error;
(24) by training the sampleX i And training the weight parameters of the deep convolutional neural network, and obtaining multilayer convolutional kernel parameters after the convolutional neural network is stable through a certain number of iterations to complete training.
4. The landslide disaster identification method based on multi-source data fusion as claimed in claim 2, wherein: the network structure design of the landslide hazard recognition neural network fully considers each influence factor and discrimination mode of the landslide hazard, fusion solving is carried out on three dimensions of data, characteristics and a network model, pixel-level landslide prediction is realized, and the landslide hazard recognition neural network specifically comprises the following modules:
1) the convolutional neural network module ResNet34 is used for learning landslide disaster influence characteristics shown by the non-optical remote sensing image data, including altitude and the like; where the input to ResNet34 isX i After convolution, the feature vector is output as the feature vector representing the influence characteristics of the landslide disaster and is recorded asS(X i )；
2) The convolutional neural network module ResNet50 is used for learning the obvious visual characteristics of the landslide disaster expressed in the optical remote sensing image data, including the edge, the geometric shape and the like of the landslide; where the input to ResNet50 isX i After convolution, the feature vector is output as a feature vector representing the significant visual features of the landslide hazard and is recorded asC(X i )；
3) The space pyramid pooling module ASPP with holes comprises five parts: one convolutional layer with convolution kernel of 1x1, three void convolutional layers with convolution kernel of 3x3 and one average pooling layer, followed by one convolutional layer with convolution kernel of 1x 1; the hole convolution adds a hole coefficient on the basis of the traditional discrete convolutionlThe calculation formula is
In the ASPP module, each convolution layer is followed by a BatchNoAn rm layer; the feature vectors output after the average pooling layer and the convolutional layer are processed are changed, a bilinear interpolation method is adopted to expand the features, then the feature vectors are spliced with the features processed by other convolutional layers in the ASPP by a Concatenate method to form multi-scale mixed features, then the multi-scale mixed features are accessed to the convolutional layer with a convolutional kernel of 1x1, and the feature vectors with 256 dimensions are output; the ASPP convolves and samples the given input in parallel with holes of different sampling rates, which is equivalent to capturing the context of the image at multiple scales, and further extracts multi-scale landslide information.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111179257.4A CN113610070A (en) | 2021-10-11 | 2021-10-11 | Landslide disaster identification method based on multi-source data fusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111179257.4A CN113610070A (en) | 2021-10-11 | 2021-10-11 | Landslide disaster identification method based on multi-source data fusion |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113610070A true CN113610070A (en) | 2021-11-05 |
Family
ID=78343538
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111179257.4A Pending CN113610070A (en) | 2021-10-11 | 2021-10-11 | Landslide disaster identification method based on multi-source data fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113610070A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114120122A (en) * | 2021-11-29 | 2022-03-01 | 中国平安财产保险股份有限公司 | Remote sensing image-based disaster damage identification method, device, equipment and storage medium |
| CN115588134A (en) * | 2022-09-08 | 2023-01-10 | 中国地质环境监测院(自然资源部地质灾害技术指导中心) | Landslide identification method and system based on landslide characteristic elements and spatial structure |
| CN115661681A (en) * | 2022-11-17 | 2023-01-31 | 中国科学院空天信息创新研究院 | Deep learning-based landslide disaster automatic identification method and system |
| CN116304991A (en) * | 2023-05-16 | 2023-06-23 | 广东省科学院广州地理研究所 | Multi-source heterogeneous species distribution data fusion method and device |
| CN117216641A (en) * | 2023-09-20 | 2023-12-12 | 四川省公路规划勘察设计研究院有限公司 | Intelligent identification method for high-level collapse geological disasters |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102819023A (en) * | 2012-07-27 | 2012-12-12 | 中国地质大学（武汉） | Method and system of landslide recognition of complicated geological background area based on LiDAR |
| CN106226779A (en) * | 2016-07-11 | 2016-12-14 | 铁道第三勘察设计院集团有限公司 | An easily Prediction of Landslide based on remote sensing technology |
| CN109740652A (en) * | 2018-12-24 | 2019-05-10 | 深圳大学 | A kind of pathological image classification method and computer equipment |
| CN109784350A (en) * | 2018-12-29 | 2019-05-21 | 天津大学 | In conjunction with the dress ornament key independent positioning method of empty convolution and cascade pyramid network |
| CN110781825A (en) * | 2019-10-25 | 2020-02-11 | 云南电网有限责任公司电力科学研究院 | Power grid landslide area identification system and method |
| AU2020100048A4 (en) * | 2020-01-10 | 2020-02-13 | Ding, Mengfang Mr | Method of object detection for vehicle on-board video based on RetinaNet |
| CN111310898A (en) * | 2020-02-14 | 2020-06-19 | 中国地质大学（武汉） | Landslide hazard susceptibility prediction method based on RNN |
| CN111968125A (en) * | 2020-06-28 | 2020-11-20 | 上海明寰科技有限公司 | High-resolution remote sensing image segmentation system and method based on information tracing |
| CN112101328A (en) * | 2020-11-19 | 2020-12-18 | 四川新网银行股份有限公司 | Method for identifying and processing label noise in deep learning |
| CN113076871A (en) * | 2021-04-01 | 2021-07-06 | 华南理工大学 | Fish shoal automatic detection method based on target shielding compensation |
| CN113469094A (en) * | 2021-07-13 | 2021-10-01 | 上海中科辰新卫星技术有限公司 | Multi-mode remote sensing data depth fusion-based earth surface coverage classification method |
-
2021
- 2021-10-11 CN CN202111179257.4A patent/CN113610070A/en active Pending
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102819023A (en) * | 2012-07-27 | 2012-12-12 | 中国地质大学（武汉） | Method and system of landslide recognition of complicated geological background area based on LiDAR |
| CN106226779A (en) * | 2016-07-11 | 2016-12-14 | 铁道第三勘察设计院集团有限公司 | An easily Prediction of Landslide based on remote sensing technology |
| CN109740652A (en) * | 2018-12-24 | 2019-05-10 | 深圳大学 | A kind of pathological image classification method and computer equipment |
| CN109784350A (en) * | 2018-12-29 | 2019-05-21 | 天津大学 | In conjunction with the dress ornament key independent positioning method of empty convolution and cascade pyramid network |
| CN110781825A (en) * | 2019-10-25 | 2020-02-11 | 云南电网有限责任公司电力科学研究院 | Power grid landslide area identification system and method |
| AU2020100048A4 (en) * | 2020-01-10 | 2020-02-13 | Ding, Mengfang Mr | Method of object detection for vehicle on-board video based on RetinaNet |
| CN111310898A (en) * | 2020-02-14 | 2020-06-19 | 中国地质大学（武汉） | Landslide hazard susceptibility prediction method based on RNN |
| CN111968125A (en) * | 2020-06-28 | 2020-11-20 | 上海明寰科技有限公司 | High-resolution remote sensing image segmentation system and method based on information tracing |
| CN112101328A (en) * | 2020-11-19 | 2020-12-18 | 四川新网银行股份有限公司 | Method for identifying and processing label noise in deep learning |
| CN113076871A (en) * | 2021-04-01 | 2021-07-06 | 华南理工大学 | Fish shoal automatic detection method based on target shielding compensation |
| CN113469094A (en) * | 2021-07-13 | 2021-10-01 | 上海中科辰新卫星技术有限公司 | Multi-mode remote sensing data depth fusion-based earth surface coverage classification method |
Non-Patent Citations (5)
| Title |
|---|
| ZHIPENG CUI ET AL: "Semantic segmentation with multi-path refinement and pyramid pooling dilated-resnet", 《2017ICIP》 * |
| 张丹丹: "《基于深度神经网络技术的高分遥感图像处理及应用》", 31 August 2020 * |
| 池长艳: "基于高分辨率遥感影像的滑坡灾害危险性评价研究", 《中国优秀博硕士学位论文全文数据库(硕士)基础科学辑》 * |
| 肖琳: "基于深度学习多模型融合的路面裂缝识别算法研究", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 * |
| 袁赣南(等)主编: "《导航定位***工程》", 28 February 2009 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114120122A (en) * | 2021-11-29 | 2022-03-01 | 中国平安财产保险股份有限公司 | Remote sensing image-based disaster damage identification method, device, equipment and storage medium |
| CN114120122B (en) * | 2021-11-29 | 2024-04-30 | 中国平安财产保险股份有限公司 | Disaster damage identification method, device, equipment and storage medium based on remote sensing image |
| CN115588134A (en) * | 2022-09-08 | 2023-01-10 | 中国地质环境监测院(自然资源部地质灾害技术指导中心) | Landslide identification method and system based on landslide characteristic elements and spatial structure |
| CN115661681A (en) * | 2022-11-17 | 2023-01-31 | 中国科学院空天信息创新研究院 | Deep learning-based landslide disaster automatic identification method and system |
| CN116304991A (en) * | 2023-05-16 | 2023-06-23 | 广东省科学院广州地理研究所 | Multi-source heterogeneous species distribution data fusion method and device |
| CN116304991B (en) * | 2023-05-16 | 2023-08-08 | 广东省科学院广州地理研究所 | Multi-source heterogeneous species distribution data fusion method and device |
| CN117216641A (en) * | 2023-09-20 | 2023-12-12 | 四川省公路规划勘察设计研究院有限公司 | Intelligent identification method for high-level collapse geological disasters |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108230329B (en) | Semantic segmentation method based on multi-scale convolution neural network | |
| CN110929607B (en) | Remote sensing identification method and system for urban building construction progress | |
| CN113610070A (en) | Landslide disaster identification method based on multi-source data fusion | |
| CN109493320B (en) | Remote sensing image road extraction method and system based on deep learning, storage medium and electronic equipment | |
| CN112712535B (en) | Mask-RCNN landslide segmentation method based on simulation difficult sample | |
| CN108596108B (en) | Aerial remote sensing image change detection method based on triple semantic relation learning | |
| CN111929683B (en) | Landslide deformation accumulation area prediction model generation method and landslide deformation accumulation area prediction method | |
| CN110991430B (en) | Ground feature identification and coverage rate calculation method and system based on remote sensing image | |
| Xu et al. | Feature-based constraint deep CNN method for mapping rainfall-induced landslides in remote regions with mountainous terrain: An application to Brazil | |
| CN114332473A (en) | Object detection method, object detection device, computer equipment, storage medium and program product | |
| CN115457396B (en) | Surface target ground object detection method based on remote sensing image | |
| CN115830471A (en) | Multi-scale feature fusion and alignment domain self-adaptive cloud detection method | |
| Lv et al. | ShapeFormer: A shape-enhanced vision transformer model for optical remote sensing image landslide detection | |
| CN117788296B (en) | Infrared remote sensing image super-resolution reconstruction method based on heterogeneous combined depth network | |
| CN116168246A (en) | Method, device, equipment and medium for identifying waste slag field for railway engineering | |
| Jing et al. | Island road centerline extraction based on a multiscale united feature | |
| CN114359709A (en) | Target detection method and device for remote sensing image | |
| CN113408540A (en) | Synthetic aperture radar image overlap area extraction method and storage medium | |
| CN109101977A (en) | A kind of method and device of the data processing based on unmanned plane | |
| Eken et al. | An automated technique to determine spatio-temporal changes in satellite island images with vectorization and spatial queries | |
| CN112800851A (en) | Water body contour automatic extraction method and system based on full convolution neuron network | |
| Qi et al. | JAED-Net: joint attention encoder–decoder network for road extraction from remote sensing images | |
| Li et al. | Learning to holistically detect bridges from large-size vhr remote sensing imagery | |
| Parape et al. | Optimization of structure elements for morphological hit-or-miss transform for building extraction from VHR airborne imagery in natural hazard areas | |
| Guili et al. | A man-made object detection algorithm based on contour complexity evaluation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20211105 |