CN111223183A - Landslide terrain detection method based on deep neural network - Google Patents
Landslide terrain detection method based on deep neural network Download PDFInfo
- Publication number
- CN111223183A CN111223183A CN201911112833.6A CN201911112833A CN111223183A CN 111223183 A CN111223183 A CN 111223183A CN 201911112833 A CN201911112833 A CN 201911112833A CN 111223183 A CN111223183 A CN 111223183A
- Authority
- CN
- China
- Prior art keywords
- landslide
- terrain
- remote sensing
- model
- sensing image
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 27
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 21
- 230000011218 segmentation Effects 0.000 claims abstract description 49
- 238000000605 extraction Methods 0.000 claims abstract description 18
- 238000007781 pre-processing Methods 0.000 claims abstract description 5
- 238000000034 method Methods 0.000 claims description 29
- 230000006870 function Effects 0.000 claims description 15
- 238000010586 diagram Methods 0.000 claims description 11
- 238000009825 accumulation Methods 0.000 claims description 9
- 238000005516 engineering process Methods 0.000 claims description 8
- 238000001228 spectrum Methods 0.000 claims description 7
- 238000002372 labelling Methods 0.000 claims description 4
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims description 4
- 238000005286 illumination Methods 0.000 claims description 3
- 230000009286 beneficial effect Effects 0.000 claims description 2
- 238000012549 training Methods 0.000 claims description 2
- 230000007547 defect Effects 0.000 description 4
- 238000006243 chemical reaction Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- 230000018109 developmental process Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000004927 fusion Effects 0.000 description 3
- 238000010606 normalization Methods 0.000 description 3
- 230000003287 optical effect Effects 0.000 description 3
- 230000003321 amplification Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 238000003199 nucleic acid amplification method Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 230000003595 spectral effect Effects 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- -1 grassland Substances 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 239000011435 rock Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
- G06T17/05—Geographic models
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Geometry (AREA)
- Software Systems (AREA)
- Remote Sensing (AREA)
- Computer Graphics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention provides a landslide terrain detection method based on a deep neural network, which comprises the steps of collecting landslide remote sensing images and digital elevation model data, obtaining specific range coordinates of landslides and generating a landslide marking graph; preprocessing data of the landslide remote sensing image, the digital elevation model and the landslide annotation map; constructing a geological feature extraction model with deep Lab V3+ as a framework and extracting abundant geological features from the remote sensing terrain data set; the landslide remote sensing image characteristics and the digital elevation model characteristics are fused, landslide field knowledge is combined, geological characteristic parameters are loaded, a terrain segmentation model is realized on the basis of a DeepLab V3+ architecture, landslide terrain segmentation of pixel-level granularity is completed, and the purpose of landslide terrain detection is achieved.
Description
Technical Field
The invention belongs to the technical field of geological disaster prediction, and particularly relates to a landslide terrain detection method based on a deep neural network of remote sensing images and digital elevation model data.
Background
The 'geological disaster hidden danger' is mostly distributed in vast western regions, conditions such as traffic, communication, electric power and the like are inconvenient, and manual checking is very difficult. 1 million casualties and 90 more than ten thousand of people suffering from disasters are caused by landslide every year in China, and the direct economic loss is as high as 20-60 billion yuan. How to effectively monitor and timely early warn geological disaster hidden danger points, especially geological disaster zones such as massif landslides with great harm for a long time, protect life safety of people, reduce property loss of people, and is a key difficult problem which needs to be solved urgently by geological monitoring personnel and related geological disaster emergency management departments.
The currently popular landslide detection methods are mainly divided into three types, namely a traditional landslide identification method, a synthetic aperture radar-based identification method and an optical remote sensing data identification method. The traditional landslide identification method monitors the development change condition of the landslide through field investigation, visual interpretation, observation point setting and the like, consumes a large amount of manpower and material resources, has the defects of complex information processing, untimely disaster early warning and the like, and cannot meet the wide area identification requirement. The synthetic aperture radar (InSAR) -based identification method mainly monitors the size and the speed of the surface deformation of a landslide body in real time, and has certain bottleneck because the technical development of China is relatively lagged, landslides are mostly distributed in remote areas and landslide basic data is scarce. With the rapid development of aerospace technology in recent years, high-resolution remote sensing images can provide abundant ground feature information, and the remote sensing technology becomes an effective means for landslide monitoring and identification. The landslide identification method based on the remote sensing image takes pixels as an analysis unit in the early stage, and has the defect that sufficient shape and terrain characteristics cannot be provided for landslide identification only by spectral characteristics of a single pixel; the object is gradually developed to be used as an analysis unit, and the defect is that the landslide is regarded as a general graph detection object, the characteristic of the landslide is ignored, the landslide existence judgment is emphasized, and the accuracy of identifying the landslide range at the pixel level is not high.
Disclosure of Invention
The main problems to be solved by the invention are as follows: the landslide terrain detection method based on the deep neural network is provided, an advanced artificial intelligence technology is utilized, accurate satellite remote sensing data is adopted, domain expert knowledge is combined, the result accuracy is high, and the investment of manpower, material resources and financial resources is greatly reduced while the result accuracy is ensured.
The technical scheme of the invention is as follows: a landslide terrain detection method based on a deep neural network is implemented according to the following steps:
a landslide terrain detection method based on a deep neural network is realized through the following steps:
step 2, carrying out data preprocessing on the landslide remote sensing image, the digital elevation model and the landslide specific range coordinate file obtained in the step 1: superposing the landslide remote sensing image and a coordinate file to generate a landslide labeling graph, wherein the comparison between the landslide labeling graph and a landslide terrain segmentation result is used for evaluating the landslide terrain detection accuracy; adopting a selective search segmentation method to increase the data sample size of the landslide remote sensing image, ensuring the integrity of landslide data, and simultaneously ensuring that the landslide remote sensing image has the same segmentation boundary with a landslide annotation diagram and a digital elevation model; the data enhancement technology is adopted to increase the training data volume, improve the generalization performance of the model and improve the robustness of the model;
step 3, constructing a deep Lab V3+ geological feature extraction model: the method comprises the following steps of adopting a deep Lab V3+ network encoder framework, using a remote sensing terrain data set with rich land coverage types as input data, and extracting general geological features in a remote sensing image: texture, color, illumination, vegetation and water body to obtain characteristic parameters for the robustness of geological features;
step 4, generating a feature vector: fusing the landslide remote sensing image and the digital elevation model data to obtain a feature vector containing the remote sensing image feature and the digital elevation feature, and using the feature vector as a basis for landslide terrain segmentation;
step 5, constructing a deep Lab V3+ terrain segmentation model: by adopting a DeepLab V3+ network encoder and decoder framework, the encoder loads characteristic parameters extracted by a geological characteristic extraction model, which is beneficial to rapid convergence of a terrain segmentation model; the decoder restores the characteristic diagram obtained by the encoder to the input size to generate a landslide terrain segmentation result;
and 6, combining the domain knowledge to construct a landslide characteristic relationship-based method: the method comprises the steps of considering the sufficiency of a local landslide area on landslide composition, learning local landslide area characteristics, and fusing the local area characteristics serving as an attention key area with integral landslide characteristics; and (3) setting constraint conditions and calculating a loss function by utilizing a method of combining spectrum, shape and context information, considering the relation among the features of the local landslide region, feeding back to act in the step (5), and finally judging the classification result that the pixels belong to the landslide and the non-landslide in the landslide terrain segmentation result according to a threshold value to generate a final landslide terrain segmentation result.
In the step 4, the landslide remote sensing image and the digital elevation model data are fused to generate a four-dimensional feature vector containing the remote sensing image feature and the digital elevation feature:
I(Xi)＝[SR(Xi)SG(Xi)SB(Xi)D(Xi)]
wherein, I (X)i) For landslide sample XiGenerated four-dimensional feature vector, SR(Xi)、SG(Xi)、SB(Xi) Are respectively landslide samples XiCorresponding red wave band, green wave band, blue wave band space and spectrum characteristics, D (X) of remote sensing imagei) For landslide sample XiCorresponding elevation features.
In the step 6, the local landslide area characteristics comprise rivers, a landslide trailing edge, a landslide accumulation body and cracks;
the local region features and the global features are fused as follows:
F(Xi)＝[I(Xi)LR(Xi)LT(Xi)LD(Xi)LF(Xi)]
wherein, F (X)i) Is a landslide sample XiPost-fusion feature vector, I (X)i) Is a landslide sample XiIntegral characteristics, LR(Xi)，LT(Xi)，LD(Xi)，LF(Xi) Are respectively landslide samples XiLocal area rivers, landslide trailing edges, landslide accumulation bodies and cracks;
in step 6, the loss function based on the landslide characteristic relationship is calculated as follows:
C＝C0+0.5C1+0.5C2+0.2C3+0.2C4
wherein C is the final loss function of the model, and the corresponding loss functions of landslide, river, landslide trailing edge, landslide accumulation body and crack are respectively C0、C1、C2、C3、C4From C to CjCalculated, n is the number of X input landslide samples, yiMarking the value, x, for a single sampleiPredicting a value for a single sample;
in the step 6, the value range of the threshold value delta T is 0.4-0.7, and the optimal threshold value is selected according to the landslide terrain detection accuracy and is 0.5. Marking the pixel points with the prediction probability larger than the threshold value as 1, judging the pixel points to be landslide composition pixel points, marking the pixel points with the prediction probability smaller than the threshold value as 0, judging the pixel points to be non-landslide areas, and finally obtaining a landslide terrain segmentation result:
wherein, O (x)j) For landslide sample XiA certain pixel point xjThe final determination result of (1), O' (x)j) And identifying an initial result for the terrain segmentation model, wherein delta T is a threshold value for determining the pixel point category.
The principle of the invention is as follows:
the method comprises the steps of constructing a geological feature extraction model and a terrain segmentation model based on a DeepLab V3+ network by adopting remote sensing image data and digital elevation data, extracting geological features of a landslide area, combining domain expert knowledge, considering the relation of features of a local area of a landslide, and obtaining a landslide terrain segmentation result by using a method of combining spectrum, shape, context information and the like.
Compared with the prior art, the invention has the advantages that:
(1) according to the landslide terrain detection method based on the deep neural network, the geological features of the landslide area in the data set are extracted and analyzed by using the remote sensing image data and the digital elevation data, and the landslide terrain in the large-area unmanned area can be segmented by means of satellite data.
(2) The detection result of the method has pixel level accuracy, and is different from the prior art that the landslide occurrence probability is only predicted within a certain range.
(3) The invention can automatically iterate, and the iterative prediction effect is better and better under the condition of enough data quantity. With the increase of data quantity, landslide features extracted by the detection method are more and more sufficient, the learning effect tends to be perfect, and the landslide feature is positively acted on the landslide segmentation effect in unknown input data.
(4) The method combines the field knowledge, fully considers a plurality of factors such as landslide optical image characteristics, landslide field knowledge, landslide characteristic relevance and the like, provides more accurate basis for making a prediction result, and overcomes the defect that the traditional detection method only depends on single data.
Drawings
FIG. 1 is a schematic diagram of an implementation of the present invention;
FIG. 2 is a schematic diagram of landslide remote sensing image data;
FIG. 3 is a schematic diagram of the preprocessed landslide remote sensing image data;
FIG. 4 is a schematic diagram of a DeepLab V3+ network architecture;
FIG. 5 is a schematic diagram of a landslide feature relationship-based method;
fig. 6 is a schematic diagram of a landslide terrain segmentation result.
Detailed Description
The following describes in detail a specific embodiment of a deep neural network-based landslide terrain detection method according to the present invention with reference to the accompanying drawings.
< example >
In this embodiment, a landslide remote sensing image and a digital elevation model are taken as an example to explain a landslide terrain detection method based on a deep neural network.
As shown in fig. 1, the landslide terrain detection method based on the deep neural network provided in this embodiment is specifically implemented according to the following steps:
setting a uniform tile level by using an open source satellite image downloading platform, and downloading a landslide remote sensing image and a digital elevation model; and marking the concrete range of the landslide by using a mask marking function on the platform, and downloading and acquiring a longitude and latitude coordinate file of the marking point. And the landslide remote sensing image, the digital elevation model and the marked point coordinate file form sample data.
The data source adopted in this embodiment is Google Earth satellite image, the spatial resolution is 0.54 meter, and the image is 3 bands, as shown in fig. 2);
step 2, data preprocessing:
realizing the projection conversion of the coordinates of the marking points and the coordinates of the images: and converting the coordinates of the marking points and the landslide remote sensing image data into the same coordinate reference system, calculating the image size and affine transformation parameters after projection conversion, cutting a marking area after registering the landslide remote sensing image, and obtaining a landslide marking image.
The landslide remote sensing image is segmented by adopting a selective search segmentation method, the segmentation scale is set to be random size, the segmentation shape is square, the segmentation result only comprises two types of all landslide areas and no landslide area, and the integrity of landslide data is guaranteed. And meanwhile, the digital elevation model and the landslide annotation image are segmented by adopting the same coordinates, so that the landslide remote sensing image, the data elevation model and the landslide annotation image have the same segmentation boundary.
And adopting a data enhancement technology for the picture type data generated after cutting, wherein the data enhancement technology comprises geometric amplification methods such as constrained random cutting, expansion, overturning, size adjustment with interpolation and the like, and optical amplification methods such as random brightness disturbance, hue disturbance, saturation disturbance, contrast disturbance and the like. The remote sensing image of the landslide after the data preprocessing is adopted is shown in fig. 3. Data normalization is realized on the digital elevation model, a Min-max standardization (Min-max normalization) method is adopted, the elevation result after linear transformation falls into a [0, 1] interval, and the conversion function is as follows:
wherein Z is*Is the elevation value after the normalization of the point Z in the digital elevation model, wherein Z is the initial value of the point Z, ZminIs at a minimum elevation value of 8maxIs the maximum elevation value.
And step 3: constructing a deep Lab V3+ geological feature extraction model:
geological feature extraction model input data based on a deep neural network is a large remote sensing terrain data set with a large number of annotation images so as to learn an effective model with a plurality of different parameters; the remote sensing terrain data set adopts 10, 20 and 60 meter image bands related to pixel sizes of 120 x 120, 60 x 60 and 20 x 20, and has a plurality of land coverage categories such as woodland, water, grassland, rock and the like. The remote sensing terrain data set with abundant data can promote the geological feature extraction model to obtain more sufficient geological features.
The geological feature extraction model adopts the encoder architecture of the existing deep neural network deep Lab V3 +. A schematic diagram of the DeepLabV3+ network is shown in FIG. 4. The geological feature extraction model specifically adopts the encoder part in fig. 4, and the remote sensing terrain data set is used as input data of the geological feature extraction model. Then, a mapping relation between input and output is learned through a backbone network Xceptation network and an ASPP (advanced Spatial Pyramid Pooling) technology so as to achieve the purpose of feature extraction. An Entry flow module in an Xcenter network is mainly used for continuously sampling and reducing space dimension, wherein Middle flow is continuously learning association relation and optimizing characteristics, and Exit flow is summarizing and sorting characteristics; the ASPP technique can capture multiple levels of context, giving pixels a larger field of view. And finally integrating the characteristics obtained by the hidden layer through the full connection layer and judging the characteristic extraction capability of the model. The deep neural network can comprehensively operate landslide characteristics at a higher level on the basis of perceiving each characteristic in the remote sensing terrain data set, so that global information of landslide is obtained.
And universal geological features such as texture, color, illumination, vegetation, water body and the like in the remote sensing image are obtained through the geological feature extraction model, and feature parameters with certain robustness for the geological features are obtained.
Step 4, generating a feature vector:
fusing the landslide remote sensing image and the digital elevation model data to generate a four-dimensional feature vector containing the remote sensing image features and the digital elevation features:
I(Xi)＝[SR(Xi)SG(Xi)SB(Xi)D(Xi)]
wherein, I (X)i) For landslide sample XiGenerated four-dimensional feature vector, SR(Xi)、SG(Xi)、SB(Xi) Are respectively landslide samples XiCorresponding red wave band, green wave band, blue wave band space and spectrum characteristics, D (X) of remote sensing imagei) For landslide sample XiCorresponding elevation features.
Step 5, constructing a deep Lab V3+ terrain segmentation model:
and (4) inputting data into the terrain segmentation model based on the deep neural network as the four-dimensional feature vector obtained in the step (4).
The terrain segmentation model adopts an encoder-decoder framework of the existing neural network DeepLab V3+, and an encoder part adopts an input layer, a hidden layer framework and parameter setting which are the same as those of the geological feature extraction model on the basis of the DeepLab V3+ geological feature extraction model. And the decoder part restores the feature map obtained by the encoder to the size equal to the size of the input data through operations such as up-sampling, low-layer and high-layer feature map fusion, convolution and the like to obtain a landslide terrain segmentation result.
And the deep Lab V3+ terrain segmentation model loads the geological feature parameters of the remote sensing image obtained by the geological feature extraction model, and inputs a four-dimensional feature vector containing the features of the remote sensing image and the digital elevation features to obtain a landslide terrain segmentation result combining the space, the spectral features and the elevation features.
And 6, combining the domain knowledge to construct a landslide characteristic relationship-based method:
a method based on the landslide feature relationship is constructed by considering the sufficiency of local areas such as rivers, landslide trailing edges, landslide deposits, cracks and the like with respect to landslide composition in combination with the domain knowledge of landslide form, scale, boundary, surface features, as shown in fig. 5.
Learning and extracting landslide overall features and local region features on shallow image features obtained by a terrain segmentation model, and fusing the local region features as attention key regions with the overall features:
F(Xi)＝[I(Xi)LR(Xi)LT(Xi)LD(Xi)LF(Xi)]
wherein, F (X)i) Is a fusion feature, I (X)i) Is an integral feature of landslide, LR(Xi)，LT(Xi)，LD(Xi)，LF(Xi) The characteristics are respectively corresponding to rivers, trailing edges of landslides, landslide accumulation bodies and cracks in local areas.
The method of combining spectrum, shape and context information is utilized, reasoning is carried out on the basis of the relation among the characteristics, and the relevance of landslide characteristics is considered: the concomitance and the proximity of the river and the landslide, the obvious difference of the rear edge of the landslide, the special texture characteristic of a landslide accumulation body and the like are set, the distance constraint condition of the river and the landslide is set, and the result is initially and accurately segmented. And calculating the loss functions of the integral features of the landslide and the features of the local regions of the landslide respectively, and feeding back the loss functions to the features of the local regions of the landslide.
Wherein, CjIs a landslide machineThe loss amount of the body characteristics and the characteristics of each landslide local area, n is the number X of the input landslide samples, yiMarking the value for a single sample, and a is a single sample prediction value;
and calculating a fusion loss function based on the landslide characteristic relation by combining the landslide overall characteristics and the loss of the characteristics of the plurality of landslide local areas, and feeding back the shallow image characteristics obtained by acting on the terrain segmentation model, so that the back propagation is realized, the parameters of the terrain segmentation model are continuously updated, and a more accurate segmentation result is favorably obtained. The loss function based on the landslide feature relationship is calculated as follows:
C＝C0+0.5C1+0.5C2+0.2C3+0.2C4
wherein C is a model loss function, and the corresponding loss amounts of landslide, river, landslide trailing edge, landslide accumulation body and crack are respectively C0、C1、C2、C3、C4；
And judging whether the pixels in the landslide terrain segmentation result belong to the classification result of landslide and non-landslide according to the threshold value, and generating a final landslide terrain segmentation result. The value range of the threshold value delta T is 0.4-0.7, the optimal threshold value is selected according to the landslide prediction probability, the optimal threshold value is 0.5, the pixel points with the prediction probability larger than the threshold value are marked as 1 and are judged as landslide forming pixel points, the pixel points with the prediction probability smaller than the threshold value are marked as 0 and are judged as non-landslide areas, and finally, the landslide terrain segmentation result is obtained:
wherein, O (x)j) For landslide sample XiA certain pixel point xjThe final determination result of (1), O' (x)j) And identifying an initial result for the terrain segmentation model, wherein delta T is a threshold value for determining the pixel point category.
The result of the landslide terrain segmentation is shown in fig. 6, wherein the area outlined by the black lines is the landslide terrain segmentation area obtained by the landslide terrain detection method.
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 (5)
1. A landslide terrain detection method based on a deep neural network is characterized by comprising the following steps:
step 1, acquiring data: according to the landslide coordinate points, acquiring a landslide remote sensing image and a Digital Elevation Model (DEM) by utilizing an open source satellite image downloading platform, marking a landslide specific range and acquiring a landslide specific range coordinate file;
step 2, carrying out data preprocessing on the landslide remote sensing image, the digital elevation model and the landslide specific range coordinate file obtained in the step 1: superposing the landslide remote sensing image and a coordinate file to generate a landslide labeling graph, wherein the comparison between the landslide labeling graph and a landslide terrain segmentation result is used for evaluating the landslide terrain detection accuracy; adopting a selective search segmentation method to increase the data sample size of the landslide remote sensing image, ensuring the integrity of landslide data, and simultaneously ensuring that the landslide remote sensing image has the same segmentation boundary with a landslide annotation diagram and a digital elevation model; the data enhancement technology is adopted to increase the training data volume, improve the generalization performance of the model and improve the robustness of the model;
step 3, constructing a deep Lab V3+ geological feature extraction model: the method comprises the following steps of adopting a deep Lab V3+ network encoder framework, using a remote sensing terrain data set with rich land coverage types as input data, and extracting general geological features in a remote sensing image: texture, color, illumination, vegetation and water body to obtain characteristic parameters for the robustness of geological features;
step 4, generating a feature vector: fusing the landslide remote sensing image and the digital elevation model data to obtain a feature vector containing the remote sensing image feature and the digital elevation feature, and using the feature vector as a basis for landslide terrain segmentation;
step 5, constructing a deep Lab V3+ terrain segmentation model: by adopting a DeepLab V3+ network encoder and decoder framework, the encoder loads characteristic parameters extracted by a geological characteristic extraction model, which is beneficial to rapid convergence of a terrain segmentation model; the decoder restores the characteristic diagram obtained by the encoder to the input size to generate a landslide terrain segmentation result;
and 6, combining the domain knowledge to construct a landslide characteristic relationship-based method: the method comprises the steps of considering the sufficiency of a local landslide area on landslide composition, learning local landslide area characteristics, and fusing the local area characteristics serving as an attention key area with integral landslide characteristics; and (3) setting constraint conditions and calculating a loss function by utilizing a method of combining spectrum, shape and context information, considering the relation among the features of the local landslide region, feeding back to act in the step (5), and finally judging the classification result that the pixels belong to the landslide and the non-landslide in the landslide terrain segmentation result according to a threshold value to generate a final landslide terrain segmentation result.
2. The landslide terrain detection method based on the deep neural network of claim 1, wherein: in the step 4, the landslide remote sensing image and the digital elevation model data are fused to generate a four-dimensional feature vector containing the remote sensing image feature and the digital elevation feature:
I(Xi)＝[SR(Xi)SG(Xi)SB(Xi)D(Xi)]
wherein, I (X)i) For landslide sample XiGenerated four-dimensional feature vector, SR(Xi)、SG(Xi)、SB(Xi) Are respectively landslide samples XiCorresponding red wave band, green wave band, blue wave band space and spectrum characteristics, D (X) of remote sensing imagei) For landslide sample XiCorresponding elevation features.
3. The landslide terrain detection method based on the deep neural network of claim 1, wherein: in the step 6, the local landslide area characteristics comprise rivers, a landslide trailing edge, a landslide accumulation body and cracks;
the local region features and the global features are fused as follows:
F(Xi)＝[I(Xi)LR(Xi)LT(Xi)LD(Xi)LF(Xi)]
wherein, F (X)i) Is a landslide sample XiPost-fusion feature vector, I (X)i) Is a landslide sample XiIntegral characteristics, LR(Xi)，LT(Xi)，LD(Xi)，LF(Xi) Are respectively landslide samples XiLocal area rivers, landslide trailing edges, landslide accumulation bodies and cracks.
4. The landslide terrain detection method based on the deep neural network of claim 1, wherein: in step 6, the loss function based on the landslide characteristic relationship is calculated as follows:
C＝C0+0.5C1+0.5C2+0.2C3+0.2C4
wherein C is the final loss function of the model, and the corresponding loss functions of landslide, river, landslide trailing edge, landslide accumulation body and crack are respectively C0、C1、C2、C3、C4From C to CjCalculated, n is the number of X input landslide samples, yiMarking the value, x, for a single sampleiIs predicted for a single sample.
5. The landslide terrain detection method based on the deep neural network of claim 1, wherein: in the step 6, the value range of the threshold value delta T is 0.4-0.7, and the optimal threshold value is selected according to the landslide terrain detection accuracy and is 0.5. Marking the pixel points with the prediction probability larger than the threshold value as 1, judging the pixel points to be landslide composition pixel points, marking the pixel points with the prediction probability smaller than the threshold value as 0, judging the pixel points to be non-landslide areas, and finally obtaining a landslide terrain segmentation result:
wherein, O (x)j) For landslide sample XiA certain pixel point xjThe final determination result of (1), O' (x)j) And identifying an initial result for the terrain segmentation model, wherein delta T is a threshold value for determining the pixel point category.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911112833.6A CN111223183A (en) | 2019-11-14 | 2019-11-14 | Landslide terrain detection method based on deep neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911112833.6A CN111223183A (en) | 2019-11-14 | 2019-11-14 | Landslide terrain detection method based on deep neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111223183A true CN111223183A (en) | 2020-06-02 |
Family
ID=70825805
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911112833.6A Pending CN111223183A (en) | 2019-11-14 | 2019-11-14 | Landslide terrain detection method based on deep neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111223183A (en) |
Cited By (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111784082A (en) * | 2020-08-04 | 2020-10-16 | 安徽亿纵电子科技有限公司 | GIS mountain torrent prevention early warning system based on big data |
| CN111882664A (en) * | 2020-07-08 | 2020-11-03 | 武汉武大卓越科技有限责任公司 | Multi-window accumulated difference crack extraction method |
| CN111929683A (en) * | 2020-07-28 | 2020-11-13 | 中国自然资源航空物探遥感中心 | Landslide deformation accumulation area prediction model generation method and landslide deformation accumulation area prediction method |
| CN112068097A (en) * | 2020-07-15 | 2020-12-11 | 北京百度网讯科技有限公司 | Method and device for marking radar remote sensing data, electronic equipment and storage medium |
| CN112149920A (en) * | 2020-10-17 | 2020-12-29 | 河北省地质环境监测院 | Regional geological disaster trend prediction method |
| CN112347960A (en) * | 2020-11-13 | 2021-02-09 | 成都理工大学 | Landslide positioning method |
| CN112418033A (en) * | 2020-11-11 | 2021-02-26 | 广州数鹏通科技有限公司 | Landslide slope surface segmentation and identification method based on mask rcnn neural network |
| CN112489180A (en) * | 2020-10-30 | 2021-03-12 | 完美世界(北京)软件科技发展有限公司 | Data processing method, system, electronic device and computer readable medium |
| CN112560716A (en) * | 2020-12-21 | 2021-03-26 | 浙江万里学院 | High-resolution remote sensing image water body extraction method based on low-level feature fusion |
| CN112686192A (en) * | 2021-01-06 | 2021-04-20 | 电子科技大学 | Landslide stability classification method based on fine terrain features |
| CN112923904A (en) * | 2021-01-26 | 2021-06-08 | 四川融鑫信息科技有限公司 | Geological disaster hidden danger detection method for multi-unmanned aerial vehicle collaborative photogrammetry |
| CN112966765A (en) * | 2021-03-18 | 2021-06-15 | 四川虹美智能科技有限公司 | Service information processing method, device and computer readable medium |
| CN113192086A (en) * | 2021-05-11 | 2021-07-30 | 中国自然资源航空物探遥感中心 | Generation method of geological disaster hidden danger deformation intensity distribution diagram and storage medium |
| CN113378766A (en) * | 2021-06-25 | 2021-09-10 | 南通大学 | Marine large-scale wind power station monitoring system based on synthetic aperture radar |
| CN113591016A (en) * | 2021-09-30 | 2021-11-02 | 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 labeling contour generation method based on multi-user cooperation |
| CN113887515A (en) * | 2021-10-28 | 2022-01-04 | 中国自然资源航空物探遥感中心 | Remote sensing landslide identification method and system based on convolutional neural network |
| CN113936204A (en) * | 2021-11-22 | 2022-01-14 | 安徽师范大学 | High-resolution remote sensing image cloud and snow identification method and device fusing terrain data and deep neural network |
| CN114067221A (en) * | 2022-01-14 | 2022-02-18 | 成都数联云算科技有限公司 | Remote sensing image woodland extraction method, system, device and medium |
| CN114743096A (en) * | 2022-03-21 | 2022-07-12 | 东华理工大学 | Regional landslide identification method, device and system |
| CN115578845A (en) * | 2022-11-24 | 2023-01-06 | 西南交通大学 | Slope trailing edge crack early warning method, device, equipment and readable storage medium |
| 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 |
| CN112560716B (en) * | 2020-12-21 | 2024-05-28 | 浙江万里学院 | High-resolution remote sensing image water body extraction method based on low-level feature fusion |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| CN109613513A (en) * | 2018-12-20 | 2019-04-12 | 长安大学 | A kind of potential landslide automatic identifying method of optical remote sensing for taking InSAR deformation into account |
| CN110060273A (en) * | 2019-04-16 | 2019-07-26 | 湖北省水利水电科学研究院 | Remote sensing image landslide plotting method based on deep neural network |
-
2019
- 2019-11-14 CN CN201911112833.6A patent/CN111223183A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| CN109613513A (en) * | 2018-12-20 | 2019-04-12 | 长安大学 | A kind of potential landslide automatic identifying method of optical remote sensing for taking InSAR deformation into account |
| CN110060273A (en) * | 2019-04-16 | 2019-07-26 | 湖北省水利水电科学研究院 | Remote sensing image landslide plotting method based on deep neural network |
Non-Patent Citations (1)
| Title |
|---|
| 陈天博;胡卓玮;魏铼;胡顺强;: "无人机遥感数据处理与滑坡信息提取", 地球信息科学学报 * |
Cited By (32)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111882664A (en) * | 2020-07-08 | 2020-11-03 | 武汉武大卓越科技有限责任公司 | Multi-window accumulated difference crack extraction method |
| CN112068097B (en) * | 2020-07-15 | 2023-10-17 | 北京百度网讯科技有限公司 | Radar remote sensing data labeling method and device, electronic equipment and storage medium |
| CN112068097A (en) * | 2020-07-15 | 2020-12-11 | 北京百度网讯科技有限公司 | Method and device for marking radar remote sensing data, electronic equipment and storage medium |
| CN111929683A (en) * | 2020-07-28 | 2020-11-13 | 中国自然资源航空物探遥感中心 | Landslide deformation accumulation area prediction model generation method and landslide deformation accumulation area prediction method |
| CN111784082B (en) * | 2020-08-04 | 2021-02-19 | 安徽亿纵电子科技有限公司 | GIS mountain torrent prevention early warning system based on big data |
| CN111784082A (en) * | 2020-08-04 | 2020-10-16 | 安徽亿纵电子科技有限公司 | GIS mountain torrent prevention early warning system based on big data |
| CN112149920A (en) * | 2020-10-17 | 2020-12-29 | 河北省地质环境监测院 | Regional geological disaster trend prediction method |
| CN112489180A (en) * | 2020-10-30 | 2021-03-12 | 完美世界(北京)软件科技发展有限公司 | Data processing method, system, electronic device and computer readable medium |
| CN112418033A (en) * | 2020-11-11 | 2021-02-26 | 广州数鹏通科技有限公司 | Landslide slope surface segmentation and identification method based on mask rcnn neural network |
| CN112418033B (en) * | 2020-11-11 | 2024-05-03 | 广州数鹏通科技有限公司 | Landslide slope surface segmentation recognition method based on mask rcnn neural network |
| CN112347960A (en) * | 2020-11-13 | 2021-02-09 | 成都理工大学 | Landslide positioning method |
| CN112347960B (en) * | 2020-11-13 | 2021-09-21 | 成都理工大学 | Landslide positioning method |
| CN112560716A (en) * | 2020-12-21 | 2021-03-26 | 浙江万里学院 | High-resolution remote sensing image water body extraction method based on low-level feature fusion |
| CN112560716B (en) * | 2020-12-21 | 2024-05-28 | 浙江万里学院 | High-resolution remote sensing image water body extraction method based on low-level feature fusion |
| CN112686192B (en) * | 2021-01-06 | 2022-05-31 | 电子科技大学 | Landslide stability classification method based on fine terrain features |
| CN112686192A (en) * | 2021-01-06 | 2021-04-20 | 电子科技大学 | Landslide stability classification method based on fine terrain features |
| CN112923904A (en) * | 2021-01-26 | 2021-06-08 | 四川融鑫信息科技有限公司 | Geological disaster hidden danger detection method for multi-unmanned aerial vehicle collaborative photogrammetry |
| CN112966765A (en) * | 2021-03-18 | 2021-06-15 | 四川虹美智能科技有限公司 | Service information processing method, device and computer readable medium |
| CN113192086B (en) * | 2021-05-11 | 2022-01-28 | 中国自然资源航空物探遥感中心 | Generation method of geological disaster hidden danger deformation intensity distribution diagram and storage medium |
| CN113192086A (en) * | 2021-05-11 | 2021-07-30 | 中国自然资源航空物探遥感中心 | Generation method of geological disaster hidden danger deformation intensity distribution diagram and storage medium |
| CN113378766B (en) * | 2021-06-25 | 2022-04-05 | 南通大学 | Marine large-scale wind power station monitoring system based on synthetic aperture radar |
| CN113378766A (en) * | 2021-06-25 | 2021-09-10 | 南通大学 | Marine large-scale wind power station monitoring system based on synthetic aperture radar |
| CN113591016A (en) * | 2021-09-30 | 2021-11-02 | 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 labeling contour generation method based on multi-user cooperation |
| CN113591016B (en) * | 2021-09-30 | 2022-01-11 | 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 labeling contour generation method based on multi-user cooperation |
| CN113887515A (en) * | 2021-10-28 | 2022-01-04 | 中国自然资源航空物探遥感中心 | Remote sensing landslide identification method and system based on convolutional neural network |
| CN113936204A (en) * | 2021-11-22 | 2022-01-14 | 安徽师范大学 | High-resolution remote sensing image cloud and snow identification method and device fusing terrain data and deep neural network |
| CN114067221B (en) * | 2022-01-14 | 2022-04-15 | 成都数联云算科技有限公司 | Remote sensing image woodland extraction method, system, device and medium |
| CN114067221A (en) * | 2022-01-14 | 2022-02-18 | 成都数联云算科技有限公司 | Remote sensing image woodland extraction method, system, device and medium |
| CN114743096B (en) * | 2022-03-21 | 2023-04-18 | 东华理工大学 | Regional landslide identification method, device and system |
| CN114743096A (en) * | 2022-03-21 | 2022-07-12 | 东华理工大学 | Regional landslide identification method, device and system |
| CN115578845A (en) * | 2022-11-24 | 2023-01-06 | 西南交通大学 | Slope trailing edge crack early warning method, device, equipment and readable storage medium |
| 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 |
|---|---|---|
| CN111223183A (en) | Landslide terrain detection method based on deep neural network | |
| CN107527352B (en) | Remote sensing ship target contour segmentation and detection method based on deep learning FCN network | |
| CN112287807B (en) | Remote sensing image road extraction method based on multi-branch pyramid neural network | |
| CN110208880B (en) | Sea fog detection method based on deep learning and satellite remote sensing technology | |
| Chen et al. | Bottom-up image detection of water channel slope damages based on superpixel segmentation and support vector machine | |
| CN113469278B (en) | Strong weather target identification method based on deep convolutional neural network | |
| CN112037244B (en) | Landsat-8 image culture pond extraction method combining index and contour indicator SLIC | |
| WO2024021225A1 (en) | High-resolution true-color visible light model generation method, high-resolution true-color visible light model inversion method, and system | |
| CN110889840A (en) | Effectiveness detection method of high-resolution 6 # remote sensing satellite data for ground object target | |
| CN112836614A (en) | High-resolution remote sensing image classification method based on residual error network and transfer learning | |
| CN115512247A (en) | Regional building damage grade assessment method based on image multi-parameter extraction | |
| Wu et al. | Asphalt pavement crack detection based on multi-scale full convolutional network | |
| Huang et al. | An improved deep learning approach for retrieving outfalls into rivers from UAS imagery | |
| Wei et al. | Using object-oriented coupled deep learning approach for typical object inspection of transmission channel | |
| CN113469097B (en) | Multi-camera real-time detection method for water surface floaters based on SSD network | |
| Li et al. | Hybrid cloud detection algorithm based on intelligent scene recognition | |
| CN113269028B (en) | Water body change detection method and system based on deep convolutional neural network | |
| Dong et al. | Pixel-level intelligent segmentation and measurement method for pavement multiple damages based on mobile deep learning | |
| Xie et al. | Land use classification using improved U-Net in remote sensing images of urban and rural planning monitoring | |
| CN113158770A (en) | Improved mining area change detection method of full convolution twin neural network | |
| CN112365508A (en) | SAR remote sensing image water area segmentation method based on visual attention and residual error network | |
| Yang et al. | Automatic Identification Method of Construction and Demolition Waste Based on Deep Learning and GAOFEN-2 Data | |
| CN116341932A (en) | Tidal flat change monitoring method and system based on double remote sensing time sequence indexes | |
| CN115797792A (en) | Distance vertex angle rule-based Huang Hailu tidal distribution area and drift prediction initial field manufacturing method thereof | |
| CN115965838A (en) | Semi-supervised learning-based solid waste detection model construction method and solid waste detection method |
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: 20200602 |