CN115631162A - Landslide hidden danger identification method, system, medium and equipment - Google Patents
Landslide hidden danger identification method, system, medium and equipment Download PDFInfo
- Publication number
- CN115631162A CN115631162A CN202211293435.0A CN202211293435A CN115631162A CN 115631162 A CN115631162 A CN 115631162A CN 202211293435 A CN202211293435 A CN 202211293435A CN 115631162 A CN115631162 A CN 115631162A
- Authority
- CN
- China
- Prior art keywords
- layer
- landslide
- sub
- image
- hidden danger
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20021—Dividing image into blocks, subimages or windows
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
Abstract
The invention provides a landslide hidden danger identification method, relates to the technical field of landslide disaster hidden danger identification, and comprises the following steps: acquiring multi-source image data of a target area, wherein the multi-source image data comprises an optical remote sensing image, an SAR image, DEM data, a normalized vegetation index and a ground surface coverage type; performing channel superposition processing on multi-source image data to obtain a multi-channel image; dividing the multi-channel image into a plurality of sub-images, and inputting the plurality of sub-images into a landslide hidden danger prediction model one by one to perform landslide prediction to obtain the probability of landslide hidden danger in a sub-area corresponding to each sub-image; identifying a sub-image with the probability greater than a preset probability threshold, wherein the sub-area corresponding to the sub-image is an area with landslide hidden dangers in a target area; identifying a subimage with the probability greater than a preset probability threshold, wherein the subimage is an area with landslide hidden danger in a target area; the problem of low precision of landslide hidden danger identification results in the prior art is solved.
Description
Technical Field
The invention relates to the technical field of landslide hazard hidden danger identification, in particular to a landslide hidden danger identification method, a system, a medium and equipment.
Background
Landslide refers to a geological disaster in which rock-soil mass on a slope slides downward along the slope under the action of gravity under the influence of inducing factors such as river erosion, underground water activity, rainfall, earthquake, artificial slope cutting and the like. The proportion of landslide in all geological disasters in the country reaches more than 60%, and according to statistics, about 80% of major geological disaster events are not in known hidden danger points. Therefore, the comprehensive, large-range and efficient landslide hidden danger identification is of great practical significance for the urgent demand wood for landslide disaster prevention and control.
The traditional manual work is examined the landslide calamity hidden danger on the spot and is inefficient, with high costs, very consume manpower and materials. In recent years, with the development of computer technology, deep learning methods have attracted wide attention in the field of landslide identification, and can effectively extract deep features and improve the precision of landslide identification. Based on the assumption of spatial autocorrelation, landslide is closely related to the surrounding environment, the traditional machine learning method can only express the linear relationship among data, and the information of the surrounding environment of a landslide point is difficult to consider; the method based on the landslide hidden danger prediction model can consider the contextual environment information of landslide, has stronger feature extraction capability, and can express the complex nonlinear relation among landslide information. At present, a deep learning method is applied to landslide identification to a certain extent, a landslide identification method on a data layer is based on optical remote sensing images mostly, multi-modal data such as Digital Elevation Model (DEM) data and Synthetic Aperture Radar (SAR) images are fused, however, landslide hidden danger information is difficult to completely depict only the DEM or SAR images, and the method is more important to extracting landslide occurring areas. Deep learning models based on Convolutional Neural Networks (CNNs), such as VGGNet, UNet, resNet, etc., have achieved certain effects in landslide recognition. However, the network depth of the models is large, the parameter quantity is large, overfitting is easy to perform when the hidden landslide danger of the region with small data quantity is identified, and the calculation force requirement is high.
At present, SAR images, DEMs, optical remote sensing and the like are main data sources for landslide identification. The common means of applying the SAR image to landslide identification is to obtain a ground surface deformation rate map by utilizing an Interferometric Synthetic Aperture Radar (InSAR) technology, distinguish landslides by a threshold division method, but the method is subjective and has the phenomena of missing judgment and error judgment; because the landslide has unique topographic geometric expression and the DEM data can fully reflect topographic geometric characteristics, the DEM data has an important auxiliary effect on landslide identification and is widely used in landslide identification research, but the high-resolution DEM data has higher acquisition cost; the optical remote sensing image reflects the characteristics of the landslide such as color, shape, texture and the like, but is limited to information of the landslide, and the potential landslide cannot be identified. However, it is difficult to fully describe the landslide hazard only by using the above data, resulting in low precision of the landslide hazard identification result.
Disclosure of Invention
The invention provides a landslide hidden danger identification method, a system, a medium and equipment, and aims to solve the problem that in the prior art, landslide hidden danger identification results are low in accuracy.
In order to achieve the above object, the present invention provides a landslide hazard identification method, including:
step 1, acquiring multi-source image data of a target area, wherein the multi-source image data comprises an optical remote sensing image, an SAR image, DEM data, a normalized vegetation index and a ground surface coverage type;
and 4, identifying the sub-image with the probability greater than a preset probability threshold, wherein the sub-area corresponding to the sub-image is an area with landslide hidden danger in the target area.
The landslide hazard prediction model comprises: the device comprises a first convolution layer, a second convolution layer, a third convolution layer, a channel attention module, a first pooling layer, a second pooling layer, a Dropout layer, a full-connection layer and a softmax layer;
the output end of the first convolution layer is connected with the input end of the channel attention module, the output end of the channel attention module is connected with the input end of the first pooling layer, the output end of the first pooling layer is connected with the input end of the second convolution layer, the output end of the second pooling layer is connected with the input end of the second pooling layer, the output end of the second pooling layer is connected with the input end of the third convolution layer, the output end of the third convolution layer is connected with the input end of the Dropout layer, the output end of the Dropout layer is connected with the input end of the full-connection layer, and the output end of the full-connection layer is connected with the input end of softmax.
And further, inputting the sub-images into the landslide hidden danger prediction model one by one in a sliding window prediction mode to perform landslide prediction.
Further, after the sub-images are input into the first convolution layer one by one to be subjected to convolution processing, landslide prediction is carried out through the channel attention module, the first pooling layer, the second convolution module, the second pooling layer, the Dropout layer, the full connection layer and the softmax layer which are connected in sequence, and the probability that landslide hidden dangers exist in the sub-region corresponding to each sub-image is calculated.
Further, the channel attention module includes:
the global average pooling layer, the fourth convolution layer and the fifth convolution layer are connected in sequence;
and aiming at each sub-image in the plurality of sub-images, inputting the sub-image after passing through the first convolutional layer into the global averaging pooling layer, the fourth convolutional layer and the fifth convolutional layer, acquiring the number of channels, and carrying out weight multiplication on the sub-image after passing through the first convolutional layer and the number of channels to endow each channel in the sub-image with weight.
Furthermore, a first batch of normalization layers are arranged behind the first coiling layer, the input ends of the first batch of normalization layers are connected with the output ends of the first coiling layer, and the output ends of the first batch of normalization layers are connected with the input ends of the channel attention module.
And a second batch of normalization layers are arranged behind the second convolution layer, the input ends of the second batch of normalization layers are connected with the output end of the second convolution layer, and the output ends of the second batch of normalization layers are connected with the input end of the second pooling layer.
And a third batch of normalization layers are arranged behind the third convolution layer, the input ends of the third batch of normalization layers are also connected with the output end of the third convolution layer, and the output end of the third batch of normalization layers is connected with the input end of the Dropout layer.
Further, after step 4, the method further comprises:
and in the target area, carrying out visual processing on the landslide hidden danger area.
The invention also provides a landslide hidden danger identification device, which comprises:
the system comprises an acquisition module, a data processing module and a data processing module, wherein the acquisition module is used for acquiring multi-source image data of a target area, and the multi-source image data comprises an optical remote sensing image, an SAR image, DEM data, a normalized vegetation index NDVI and a ground surface coverage type;
the processing module is used for carrying out channel superposition processing on the multi-source image data to obtain a multi-channel image;
the prediction module is used for dividing the multi-channel image into a plurality of sub-images, inputting the plurality of sub-images into the landslide hidden danger prediction model one by one for landslide prediction, and obtaining the probability that the sub-area corresponding to each sub-image has landslide hidden danger; the sub-region is a part of the target region;
the identification module is used for identifying the sub-image with the probability greater than the preset probability threshold, and the sub-area corresponding to the sub-image is the area with the landslide hidden danger in the target area.
The invention also provides a computer readable storage medium for storing a computer program, and the computer program is executed to realize the landslide hazard identification method.
The invention also provides equipment for identifying the landslide hidden danger, which is used for realizing the landslide hidden danger identification method and comprises the following steps:
a memory and a processor;
the memory is used for storing a computer program;
the processor is for executing the computer program stored by the memory.
The scheme of the invention has the following beneficial effects:
the method comprises the steps of obtaining multi-source image data comprising an optical remote sensing image, an SAR image, DEM data, a normalized vegetation index NDVI and a ground surface coverage type, carrying out channel superposition processing on the multi-source image data to obtain a multi-channel image, dividing the multi-channel image into a plurality of sub-images according to lines and columns, inputting the sub-images into a landslide hidden danger prediction model one by one for landslide prediction, and obtaining the probability that the corresponding sub-area of each sub-image has landslide hidden dangers; identifying a sub-image with the probability larger than a preset probability threshold, and taking a sub-area corresponding to the sub-image as an area with landslide hidden dangers in a target area; compared with the conventional convolutional neural network, the landslide hidden danger prediction model adopted by the invention has the advantages of relatively less parameter quantity, lighter network, high training convergence speed, small dependence on computational power, higher efficiency, low acquisition difficulty of used data, suitability for large-range landslide hidden danger identification and capability of solving the problem of low precision of landslide hidden danger identification results caused by insufficient consideration of multi-source data characteristics of landslide regions in the prior art.
Other advantages of the present invention will be described in detail in the detailed description that follows.
Drawings
FIG. 1 is a schematic flow chart of an embodiment of the present invention;
FIG. 2 is a structural diagram of a landslide hazard prediction model in an embodiment of the invention;
FIG. 3 is a block diagram of a channel attention module in an embodiment of the present invention;
FIG. 4 is a ROC curve obtained by an embodiment of the present invention;
FIG. 5 is a PR curve obtained according to an embodiment of the present invention;
fig. 6 shows a landslide hazard identification visualization result according to an embodiment of the present invention.
Detailed Description
In order to make the technical problems, technical solutions and advantages of the present invention more apparent, the following detailed description is given with reference to the accompanying drawings and specific embodiments. It is to be understood that the embodiments described are only a few embodiments of the present invention, and not all embodiments. All other embodiments, which can be obtained by a person skilled in the art without making any creative effort based on the embodiments in the present invention, belong to the protection scope of the present invention.
In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicate orientations or positional relationships based on the orientations or positional relationships shown in the drawings, and are only for convenience of description and simplicity of description, but do not indicate or imply that the device or element being referred to must have a particular orientation, be constructed and operated in a particular orientation, and thus, should not be construed as limiting the present invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance.
In the description of the present invention, it is to be noted that, unless otherwise explicitly specified or limited, the terms "mounted", "connected" and "connected" are to be understood broadly, for example, as being either a locked connection, a detachable connection, or an integral connection; can be mechanically or electrically connected; they may be connected directly or indirectly through intervening media, or they may be interconnected between two elements. The specific meanings of the above terms in the present invention can be understood in a specific case to those of ordinary skill in the art.
Furthermore, the technical features involved in the different embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
The invention provides a landslide hidden danger identification method, a system, a medium and equipment aiming at the existing problems.
As shown in fig. 1, an embodiment of the present invention provides a landslide hazard identification method, including:
step 1, acquiring multi-source image data of a target area, wherein the multi-source image data comprises an optical remote sensing image, an SAR image, DEM data, a normalized vegetation index and a ground surface coverage type;
and 4, identifying the sub-image with the probability larger than a preset probability threshold, wherein the sub-area corresponding to the sub-image is an area with landslide hidden dangers in the target area.
The method for identifying a potential landslide hazard according to the present invention will be specifically described below with reference to the example of the town Yuan county in Yunnan province.
The method is characterized in that a Yuan county in Yunnan province is taken as a research area, the Yuan county in Yunnan province is located in the southwest of Yunnan province in China and between a Shalao mountain and a non-quantitative mountain, the area of a mountain area accounts for more than ninety percent of the total area of the county, the mountain area is a frequent landslide hazard area, 231 samples of the gentle slope hidden danger points in the area are taken, and the samples come from landslides which occur historically or are interpreted and labeled by experts.
In the embodiment of the invention, an optical remote sensing image (Landsat), a Sentinel-1SAR image, a DEM, an NDVI and a ground surface coverage type of a research area are obtained, the data can be obtained on a public website, and the unified resolution is set to be 30m in the embodiment. And acquiring an InSAR deformation rate map by utilizing the SAR image, setting the spatial resolution as 30m, and removing a nan value. Unifying an optical remote sensing image (RGB three wave bands), an InSAR deformation rate chart, a DEM, an NDVI and a ground surface coverage type into a geographical coordinate system, and generating a multichannel image in a channel superposition mode; and the channel superposition processing refers to splicing the pixel matrix of the multi-source image data to obtain a multi-channel image.
In the embodiment of the invention, existing landslide hidden danger points in a research area are used as positive samples, and in order to estimate the balance of a sample set, the positive samples and the negative samples are randomly sampled according to the proportion of 1. In practical application, landslide occupies a very small part of the whole map, and the rest parts are considered to be non-landslide regions, so that negative sample sampling may occur in a landslide hidden danger region or a landslide region may occur in an experiment, the situation belongs to a small probability event, and a small part of noise samples can also improve the robustness of the model.
The embodiment of the invention adopts a multi-scale strategy, samples are carried out based on a batch mode, three scales of 16 × 16, 20 × 20 and 24 × 24 are selected, and the sampling is carried out again to the smaller scale of 16 × 16. Based on the spatial autocorrelation assumption, the occurrence of landslide is closely related to the surrounding environment, and the multi-scale strategy can fully estimate the multi-scale context information around the landslide.
Specifically, before the landslide hazard prediction model is input, image data needs to be cleaned and data enhanced, and samples containing abnormal values nan are removed.
The embodiment of the invention adopts five modes of horizontal symmetrical turning, vertical symmetrical turning, 90-degree rotation, 180-degree rotation and 270-degree rotation to expand the number of samples by 5 times.
Specifically, as shown in fig. 2, the landslide hazard prediction model constructed in the embodiment of the present invention includes:
a first convolution layer, a second convolution layer and a third convolution layer for convolution processing;
a channel attention module for channel weighting;
a first pooling layer, a second pooling layer for performing downsampling operations;
a Dropout layer and a full connection layer for preventing the landslide hidden danger prediction model from being over-fitted;
a softmax layer for classification;
the output end of the first convolution layer is connected with the input end of the channel attention module, the output end of the channel attention module is connected with the input end of the first pooling layer, the output end of the first pooling layer is connected with the input end of the second convolution layer, the output end of the second pooling layer is connected with the input end of the second pooling layer, the output end of the second pooling layer is connected with the input end of the third convolution layer, the output end of the third convolution layer is connected with the input end of the Dropout layer, the output end of the Dropout layer is connected with the input end of the full-connection layer, and the output end of the full-connection layer is connected with the input end of softmax.
Specifically, after the sub-images are input into the first convolution layer one by one to be subjected to convolution processing, landslide prediction is performed through the channel attention module, the first pooling layer, the second convolution module, the second pooling layer, the Dropout layer, the full-link layer and the softmax layer which are connected in sequence, and the probability that landslide hidden dangers exist in the sub-region corresponding to each sub-image is calculated.
Specifically, in the embodiment of the present invention, the first convolution layer, the second convolution layer, and the third convolution layer all use convolution kernels with a size of 3 × 3, and the convolution step size stride and the feature map filling width padding are all set to 1; setting the window size and the step length of the first pooling layer and the second pooling layer as 2, and performing double down-sampling; dropout layer parameter is set to 0.5; finally, classification is carried out by using a softmax layer.
Specifically, as shown in fig. 3, the channel attention module includes:
the global average pooling layer, the fourth convolution layer and the fifth convolution layer are connected in sequence;
and aiming at each sub-image in the plurality of sub-images, inputting the sub-image after passing through the first convolution layer into the global average pooling layer, the fourth convolution layer and the fifth convolution layer to obtain the number of channels, multiplying the sub-image after passing through the first convolution layer by the number of channels in a weighting manner, and giving a weight to each channel in the sub-image so as to establish correlation among multiple channels and strengthen important characteristics.
Specifically, a first batch of normalization layers are arranged behind the first coiling layer, the input ends of the first batch of normalization layers are connected with the output end of the first coiling layer, and the output ends of the first batch of normalization layers are connected with the input end of the channel attention module;
a second batch of standardized layers are arranged behind the second convolution layer, the input ends of the second batch of standardized layers are connected with the output end of the second convolution layer, and the output ends of the second batch of standardized layers are connected with the input end of the second pooling layer;
a third batch of standardized layers are arranged behind the third convolution layer, the input ends of the third batch of standardized layers are also connected with the output end of the third convolution layer, and the output end of the third batch of standardized layers is connected with the input end of the Dropout layer;
and carrying out batch normalization processing through the first batch of normalization layers, the second batch of normalization layers and the third batch of normalization layers, wherein the function of the batch normalization processing is to accelerate the convergence of the model.
Inputting a sample set into a landslide hidden danger prediction model for training, wherein the size of an input sample is 16 × 7 (16 × 16 pixels, 7 channels), and a binary cross entropy is adopted as a loss function, and the formula is as follows:
wherein N is the total number of samples, y i True label 0 (non-landslide) or 1 (landslide), p (y) for ith binary sample i ) The probability of belonging to the y-label is output for the ith binary sample.
The landslide hazard prediction model obtained by training is tested on a test set, and the test result is shown in fig. 4 and 5.
According to the embodiment of the invention, a plurality of 16 × 16 sub-images are obtained by dividing the sub-images into 16 × 16 sub-images according to the rows and columns, the multi-channel image is cut into 16 × 16 sub-images by adopting a sliding window prediction mode, the 16 × 16 sub-images are input into a first convolution layer one by one to be subjected to convolution processing, landslide prediction is carried out through a channel attention module, a first pooling layer, a second convolution module, a second pooling layer, a third convolution layer, a Dropout layer, a full connection layer and a softmax layer which are connected in sequence, the probability of landslide hidden danger of each sub-image is calculated, and the full-map prediction result is obtained by splicing.
The row and column division mode is as follows: let the size of the sub-image be n x n (16 x 16 in the present invention), the number of rows and columns of the multi-channel image a x b, calculate how many sub-images can be divided, make up for 0 where there is not enough rows and columns, and then traverse the divided sub-images from row to column.
Specifically, the preset probability threshold is 0.8, a sub-image with the probability greater than 0.8 is identified, the sub-area corresponding to the sub-image is considered to be an area with landslide potential hazards in the target area, and the sub-area corresponding to the sub-image is considered to be an area without landslide potential hazards in the target area if the probability is less than 0.8;
specifically, after step 4, the method further comprises:
in the target area, the landslide hazard potential area is subjected to visualization processing, and a visualization result graph for identifying the landslide hazard potential is obtained and is shown in fig. 6.
According to the embodiment of the invention, a multi-scale strategy and multi-source data are used for carrying out an ablation test, in the training process of the landslide hazard prediction model, the number of the adopted positive samples and the number of the adopted negative samples are both 213, the batch size of batch _ size is set to be 16, and the training batch epoch is set to be 40. And evaluating the model by adopting four indexes of Accuracy, recall, precision and F1-Score. And calculating a confusion matrix of the model prediction result, and calculating the four indexes.
In the confusion matrix, TP represents positive samples with correct prediction, FP represents positive samples with incorrect prediction, FN represents negative samples with correct prediction, and TN represents negative samples with incorrect prediction.
Accuracy measures the Accuracy of the classification, which is the ratio of the number of correctly classified samples to the total number of samples. The calculation formula is as follows:
recall is the Recall rate and represents the probability of correct prediction in the positive sample, and is calculated by the formula:
precision is the Precision, which represents the probability of correct prediction in samples predicted to be positive, and is calculated by the formula:
F1-Score is the harmonic mean of recall ratio and accuracy ratio and measures the comprehensive performance of the model. The calculation formula is as follows:
the ablation experimental results are shown in the following table 1, wherein the multi-source represents that NDVI and the earth surface coverage type are added in addition to the basic optical remote sensing image, inSAR deformation rate and DEM; multiscale representation uses a multiscale sampling strategy; attention represents the joining channel attention module.
| Accuracy | Recall | Precision | F1-Score | Reference number (M) | |
| Multi-source + multi-scale + attention | 0.942 | 0.941 | 0.943 | 0.942 | 7.36 |
| Multi-source + multi-scale | 0.921 | 0.938 | 0.902 | 0.920 | 7.29 |
| Multiple sources | 0.788 | 0.881 | 0.667 | 0.759 | 7.29 |
| / | 0.727 | 0.674 | 0.892 | 0.768 | 7.29 |
TABLE 1
Experimental results show that the additional multi-source data and the multi-scale strategy adopted by the embodiment of the invention can effectively improve the precision of landslide identification. Ablation experiments show that by adopting a multi-scale sampling strategy, a sample set can be expanded to a certain degree, and the problem that overfitting is easy to happen due to few landslide sample points is solved.
Compared with the conventional convolutional neural network, the landslide hidden danger prediction model in the embodiment of the invention has the advantages that the parameter quantity is smaller and is only 7.37M, the network is lighter, the training convergence speed is high, the dependence on calculation power is smaller, the efficiency is higher, the acquisition difficulty of used data is low, and the method is suitable for large-range landslide hidden danger recognition. Under the condition of only depending on more than 200 positive sample points, higher identification precision can be obtained; the problem of low precision of landslide hidden danger identification results in the prior art is solved.
The embodiment of the invention also provides a landslide hazard identification device, which comprises:
the system comprises an acquisition module, a data processing module and a data processing module, wherein the acquisition module is used for acquiring multi-source image data of a target area, and the multi-source image data comprises an optical remote sensing image, an SAR image, DEM data, a normalized vegetation index NDVI and a ground surface coverage type;
the processing module is used for carrying out channel superposition processing on the multi-source image data to obtain a multi-channel image;
the prediction module is used for dividing the multi-channel image into a plurality of sub-images, inputting the plurality of sub-images into the landslide hidden danger prediction model one by one for landslide prediction, and obtaining the probability that the sub-area corresponding to each sub-image has landslide hidden danger;
the identification module is used for identifying a sub-image with the probability larger than a preset probability threshold, and the sub-region corresponding to the sub-image is a region with landslide hidden dangers in the target region.
The embodiment of the invention also provides a computer-readable storage medium for storing a computer program and implementing the landslide hazard identification method by executing the computer program.
The embodiment of the invention also provides equipment for identifying the landslide hidden danger, which is used for realizing the method for identifying the landslide hidden danger and comprises the following steps:
a memory and a processor;
the memory is used for storing a computer program;
the processor is for executing the computer program stored by the memory.
While the foregoing is directed to the preferred embodiment of the present invention, it will be appreciated by those skilled in the art that various changes and modifications may be made therein without departing from the principles of the invention as set forth in the appended claims.
Claims (9)
1. A landslide hidden danger identification method is characterized by comprising the following steps:
step 1, obtaining multi-source image data of a target area, wherein the multi-source image data comprises an optical remote sensing image, an SAR image, DEM data, a normalized vegetation index and a ground surface coverage type;
step 2, channel superposition processing is carried out on the multi-source image data to obtain a multi-channel image;
step 3, dividing the multi-channel image into a plurality of sub-images, and inputting the sub-images into a landslide hidden danger prediction model one by one to predict landslide, so as to obtain the probability of landslide hidden danger in the sub-area corresponding to each sub-image; the sub-region is a portion of the target region;
step 4, identifying the subimages with the probability larger than a preset probability threshold, wherein the subareas corresponding to the subimages are areas with landslide hidden dangers in the target area;
the landslide hazard prediction model comprises: the system comprises a first convolution layer, a second convolution layer, a third convolution layer, a channel attention module, a first pooling layer, a second pooling layer, a Dropout layer, a full-link layer and a softmax layer;
the output of first convolution layer with the input of passageway attention module is connected, the output of passageway attention module with the input of first pooling layer, the output of first pooling layer with the input of second pooling layer is connected, the output of second pooling layer with the input of third pooling layer is connected, the output of third pooling layer with the input of Dropout layer is connected, the output of Dropout layer with the input of full connecting layer is connected, the output of full connecting layer with softmax's input is connected.
2. The landslide hazard identification method according to claim 1,
and inputting the sub-images into the landslide hidden danger prediction model one by one in a sliding window prediction mode to perform landslide prediction.
3. The landslide hazard identification method according to claim 1,
after the sub-images are input into the first convolution layer one by one to be subjected to convolution processing, landslide prediction is carried out through the channel attention module, the first pooling layer, the second convolution layer, the second pooling layer, the third convolution layer, the Dropout layer, the fully-connected layer and the softmax layer which are sequentially connected, and the probability that landslide hidden dangers exist in the sub-region corresponding to each sub-image is calculated.
4. The landslide hazard identification method of claim 3 wherein the lane attentiveness module comprises:
the global average pooling layer, the fourth convolution layer and the fifth convolution layer are connected in sequence;
for each sub-image in the plurality of sub-images, inputting the sub-image subjected to the first convolutional layer into the global average pooling layer, the fourth convolutional layer and the fifth convolutional layer, acquiring the number of channels, and multiplying the sub-image subjected to the first convolutional layer by the number of channels to give a weight to each channel in the sub-image.
5. The landslide hazard identification method according to claim 3,
a first coiling layer is arranged behind the first coiling layer, the input end of the first coiling layer is connected with the output end of the first coiling layer, and the output end of the first coiling layer is connected with the input end of the channel attention module;
a second batch of standardized layers are arranged behind the second convolution layer, the input ends of the second batch of standardized layers are connected with the output end of the second convolution layer, and the output ends of the second batch of standardized layers are connected with the input end of the second pooling layer;
and a third batch of standardized layers are arranged behind the third convolution layer, the input ends of the third batch of standardized layers are also connected with the output end of the third convolution layer, and the output end of the third batch of standardized layers is connected with the input end of the Dropout layer.
6. The landslide hazard identification method according to claim 3, further comprising after step 4:
and in the target area, carrying out visualization processing on the landslide hidden danger area.
7. The utility model provides a landslide hidden danger recognition device which characterized in that includes:
the system comprises an acquisition module, a data processing module and a data processing module, wherein the acquisition module is used for acquiring multi-source image data of a target area, and the multi-source image data comprises an optical remote sensing image, an SAR image, DEM data, a normalized vegetation index NDVI and a ground surface coverage type;
the processing module is used for carrying out channel superposition processing on the multi-source image data to obtain a multi-channel image;
the prediction module is used for dividing the multi-channel image into a plurality of sub-images, inputting the sub-images into a landslide hidden danger prediction model one by one for landslide prediction, and obtaining the probability that the sub-area corresponding to each sub-image has landslide hidden danger; the sub-region is a portion of the target region;
and the identification module is used for identifying the sub-image with the probability greater than a preset probability threshold, and the sub-area corresponding to the sub-image is the area with the landslide hidden danger in the target area.
8. A computer-readable storage medium for storing a computer program, wherein the computer program is executed to implement the landslide hazard identification method according to any one of claims 1-6.
9. A landslide hazard identification apparatus for implementing the landslide hazard identification method according to any one of claims 1 to 6, comprising:
a memory and a processor;
the memory is used for storing a computer program;
the processor is configured to execute the computer program stored by the memory.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211293435.0A CN115631162A (en) | 2022-10-21 | 2022-10-21 | Landslide hidden danger identification method, system, medium and equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211293435.0A CN115631162A (en) | 2022-10-21 | 2022-10-21 | Landslide hidden danger identification method, system, medium and equipment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115631162A true CN115631162A (en) | 2023-01-20 |
Family
ID=84907461
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211293435.0A Pending CN115631162A (en) | 2022-10-21 | 2022-10-21 | Landslide hidden danger identification method, system, medium and equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115631162A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116030353A (en) * | 2023-03-29 | 2023-04-28 | 成都大学 | Landslide hazard automatic identification method based on convolutional neural network |
| CN116108758A (en) * | 2023-04-10 | 2023-05-12 | 中南大学 | Landslide susceptibility evaluation method |
| CN116561536A (en) * | 2023-07-11 | 2023-08-08 | 中南大学 | Landslide hidden danger identification method, terminal equipment and medium |
-
2022
- 2022-10-21 CN CN202211293435.0A patent/CN115631162A/en active Pending
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116030353A (en) * | 2023-03-29 | 2023-04-28 | 成都大学 | Landslide hazard automatic identification method based on convolutional neural network |
| CN116030353B (en) * | 2023-03-29 | 2023-05-30 | 成都大学 | Landslide hazard automatic identification method based on convolutional neural network |
| CN116108758A (en) * | 2023-04-10 | 2023-05-12 | 中南大学 | Landslide susceptibility evaluation method |
| CN116561536A (en) * | 2023-07-11 | 2023-08-08 | 中南大学 | Landslide hidden danger identification method, terminal equipment and medium |
| CN116561536B (en) * | 2023-07-11 | 2023-11-21 | 中南大学 | Landslide hidden danger identification method, terminal equipment and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109446992B (en) | Remote sensing image building extraction method and system based on deep learning, storage medium and electronic equipment | |
| CN113298818B (en) | Remote sensing image building segmentation method based on attention mechanism and multi-scale features | |
| CN109493320B (en) | Remote sensing image road extraction method and system based on deep learning, storage medium and electronic equipment | |
| CN115631162A (en) | Landslide hidden danger identification method, system, medium and equipment | |
| CN112668494A (en) | Small sample change detection method based on multi-scale feature extraction | |
| CN112183432B (en) | Building area extraction method and system based on medium-resolution SAR image | |
| CN110111345B (en) | Attention network-based 3D point cloud segmentation method | |
| Ling et al. | Object-based sub-pixel mapping of buildings incorporating the prior shape information from remotely sensed imagery | |
| CN113780149B (en) | Remote sensing image building target efficient extraction method based on attention mechanism | |
| CN111274918B (en) | River dryout and cutoff monitoring method and device based on multi-source remote sensing image | |
| Tian et al. | Large-scale deep learning based binary and semantic change detection in ultra high resolution remote sensing imagery: From benchmark datasets to urban application | |
| CN113343563B (en) | Landslide susceptibility evaluation method based on automatic sample selection and surface deformation rate | |
| CN111667187B (en) | Highway landslide risk evaluation method based on multi-source remote sensing data | |
| CN113312993B (en) | Remote sensing data land cover classification method based on PSPNet | |
| CN113223042B (en) | Intelligent acquisition method and equipment for remote sensing image deep learning sample | |
| CN114612664A (en) | Cell nucleus segmentation method based on bilateral segmentation network | |
| CN115331102A (en) | Remote sensing image river and lake shoreline intelligent monitoring method based on deep learning | |
| CN110569733B (en) | Lake long time sequence continuous water area change reconstruction method based on remote sensing big data platform | |
| CN113688172B (en) | Landslide susceptibility evaluation model training method, landslide susceptibility evaluation device and medium | |
| CN114943902A (en) | Urban vegetation unmanned aerial vehicle remote sensing classification method based on multi-scale feature perception network | |
| CN113989287A (en) | Urban road remote sensing image segmentation method and device, electronic equipment and storage medium | |
| CN113628180A (en) | Semantic segmentation network-based remote sensing building detection method and system | |
| CN112836614A (en) | High-resolution remote sensing image classification method based on residual error network and transfer learning | |
| CN116844053A (en) | Wheat planting area identification method, system, electronic equipment and storage medium | |
| CN115527113A (en) | Bare land classification method and device for remote sensing image |
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 |