CN110807400A - Twin network-based collapse hidden danger characteristic information extraction method - Google Patents
Twin network-based collapse hidden danger characteristic information extraction method Download PDFInfo
- Publication number
- CN110807400A CN110807400A CN201911034777.9A CN201911034777A CN110807400A CN 110807400 A CN110807400 A CN 110807400A CN 201911034777 A CN201911034777 A CN 201911034777A CN 110807400 A CN110807400 A CN 110807400A
- Authority
- CN
- China
- Prior art keywords
- network
- hidden danger
- twin network
- data
- collapse hidden
- 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
- 238000000605 extraction Methods 0.000 title claims abstract description 12
- 238000012549 training Methods 0.000 claims abstract description 9
- 238000013528 artificial neural network Methods 0.000 claims description 5
- 238000013527 convolutional neural network Methods 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 2
- 239000013598 vector Substances 0.000 abstract description 6
- 238000001514 detection method Methods 0.000 abstract description 4
- 238000000034 method Methods 0.000 description 5
- 230000007547 defect Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000012790 confirmation Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000013024 troubleshooting Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
- 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
- G06V10/443—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 by matching or filtering
- G06V10/449—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters
- G06V10/451—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters with interaction between the filter responses, e.g. cortical complex cells
- G06V10/454—Integrating the filters into a hierarchical structure, e.g. convolutional neural networks [CNN]
Abstract
The invention relates to a twin network-based collapse hidden danger feature information extraction method, which solves the problem of how to effectively extract collapse hidden danger features from multi-source information data. The technical scheme includes that gradient, remote sensing and other data are fused, then multi-source data are sent to a twin network for training, the twin network comprises two inputs and one output, pictures marked as containing collapse hidden danger and pictures marked as not containing the collapse hidden danger are randomly paired when the network is trained, then paired data pairs are sent to the network for training, whether two pictures are of the same type or not is output from the network, the network is considered to be trained after the network is converged, and finally the pictures to be detected are sent to the trained network to obtain feature vectors extracted from unknown pictures in the network. The invention can provide input characteristics for automatic detection of collapse hidden danger, and can effectively improve the accuracy and efficiency of automatic detection of collapse hidden danger.
Description
Technical Field
The invention belongs to the field of geographic remote sensing, and further relates to a twin network-based collapse hidden danger characteristic information extraction method
Background
At present, the collapse geological disaster becomes one of geological disasters with high occurrence frequency and large influence in mountain areas in China, so that the position information of hidden danger points of the collapse geological disaster can be effectively and quickly identified and predicted, and the method has important effects on reducing influence of related geological disasters, troubleshooting of dangerous cases and early warning. At present, the confirmation of geological disasters of geological collapse is carried out by combining high-resolution remote sensing information, topographic information, field inspection and verification and other means. The methods have the defects of long manual investigation time, high sample acquisition cost, untimely detection of hidden danger points and the like, and are difficult to rapidly investigate the hidden danger points of the geological disaster in a large area range.
The current main methods for extracting and sampling features are manually selected features and automatically extracted features, the manually selected features are widely applied to image processing, but manually selected visual features may not well distinguish data, the neural network based on deep learning can obtain rich features of images, and the characteristic of neural network layering is convenient for people to perform experiments on features of different layers, so that the feature performance of collapse on remote sensing images is further researched. On the region labeling in the identification of the actual task collapse hidden trouble points, only the region of the existing known collapse hidden trouble points can be labeled, and for other regions, whether collapse hidden troubles exist or not is not known, so that the other regions cannot be labeled as negative samples for carrying out classification training. Therefore, how to learn common features in similar objects is the primary requirement for identifying and detecting collapse hidden danger from a single-class sample in a scene with only a single-class object.
The twin network is a network which takes the twin network as a starting point and completes similarity learning among single-class data, and training using the single-class data is beneficial to extracting similar features among collapse hidden danger point areas by the neural network.
Disclosure of Invention
In order to overcome the defects of large manual demand, high sample acquisition cost, untimely detection of hidden danger points and the like in the prior art, the invention provides a twin network-based collapse hidden danger feature information extraction method.
The technical scheme adopted by the invention for solving the technical problems is as follows:
the twin network model comprises two convolutional networks sharing weight, the input of the two convolutional networks is a pair of images marked as the same type or different types, and the convolutional neural networks are respectively used for automatically extracting the characteristics in the two networks, so that the networks can accurately classify whether the data are the same type or not on the basis of a large number of characteristics.
The characteristic dimensionality of different network layers can be defined by defining the number of convolution kernels in the convolution layers, each convolution kernel can be adjusted in size according to needs, and the nodes of each layer in the convolution neural network are connected with part of or all of the nodes of the previous layer and used for acquiring the characteristics of the image area.
The twin network has two inputs, the output of which is the probability that two input pictures belong to the same category. The twin network optimization aims at high similarity between features of data of the same category and low similarity between data of different categories.
In actual training, the network is trained with photos labeled as collapsed and photos labeled as not collapsed. And then, sending the unknown sample into a trained twin network model, and extracting the finally output feature vector as the feature of the unknown sample on whether collapse hidden danger exists or not.
The method has the advantages that the generated area sample pair is used for training the twin network model, the convolution layer with the last characteristic extraction in the branch network is selected as a characteristic extractor, and the extracted characteristic is used for representing the collapse hidden danger common characteristic of the sample and is used for further hidden danger point identification.
Drawings
FIG. 1 is a flow chart of twin network extraction of collapse hidden trouble features.
Fig. 2 is a schematic diagram of a twin network structure for feature extraction of collapse hidden danger points.
Detailed Description
The invention is further described with reference to the following figures and examples.
(1) And fusing various types of data including elevation digital information (DEM), gradient data, slope data and high-resolution remote sensing images, and splicing the data in the channel dimension to obtain multi-source data.
(2) And carrying out grid division on the multi-source data, wherein each grid image is used as sample data.
(3) And (3) making an input data pair, wherein one sample with hidden danger and one sample without hidden danger are used as input data, and two samples with hidden danger or two samples without hidden danger are used as the other input data. A sample pair is marked as 1 when both data in the pair are from the same class, and 0 otherwise. The data set is denoted as S _ in.
(4) Training is fed S _ in into a twin network, with separate branches for each input, but with shared parameters between the branches. The final result is ensured to be irrelevant to the data input sequence, namely, the model meets the symmetry.
(5) The twin network outputs characteristic vectors in the training process, the characteristic vectors of the two branches are calculated to be combined in a distance calculation layer, the similarity degree of the characteristic vectors extracted by the two branches is calculated, the similarity measurement of the final output [0,1] is calculated by using a sigmoid layer, and the network is optimized by using a cross entropy loss function.
(6) And sending the unknown sample into a trained twin network model, and extracting the finally output feature vector as the feature of the unknown sample on whether collapse hidden danger exists or not.
Claims (4)
1. A collapse hidden danger characteristic information extraction method based on a twin network is characterized by comprising the following steps: the model comprises two paths of convolution networks sharing weight, the inputs of the two paths of the convolution networks are a pair of images marked as the same type or different types, and the characteristics of the two paths of the networks are automatically extracted by utilizing the convolution neural networks respectively, so that the networks can realize the accurate classification of whether the data are the same type or not on the basis of a large number of characteristics.
2. The twin network based collapse hidden danger feature information extraction method according to claim 1, characterized in that: the characteristic dimensions of different network layers can be defined in the convolutional layers of the convolutional neural network by defining the number of convolutional cores, each convolutional core can be adjusted in size according to needs, and the nodes of each layer in the convolutional neural network are connected with part of or all the nodes of the previous layer to acquire the characteristics of the image area.
3. The twin network based collapse hidden danger feature information extraction method according to claim 1 or 2, characterized in that: the twin network has two inputs, the output of which is the probability that two input pictures belong to the same category, and the goal of twin network optimization is that the similarity between features of data of the same category is high and the similarity between data of different categories is low.
4. The twin network based collapse hidden danger feature information extraction method according to claim 3, characterized in that: in actual training, the network is trained with photos labeled as collapsed and photos labeled as not collapsed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911034777.9A CN110807400A (en) | 2019-10-29 | 2019-10-29 | Twin network-based collapse hidden danger characteristic information extraction method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911034777.9A CN110807400A (en) | 2019-10-29 | 2019-10-29 | Twin network-based collapse hidden danger characteristic information extraction method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110807400A true CN110807400A (en) | 2020-02-18 |
Family
ID=69489374
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911034777.9A Pending CN110807400A (en) | 2019-10-29 | 2019-10-29 | Twin network-based collapse hidden danger characteristic information extraction method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110807400A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112491468A (en) * | 2020-11-20 | 2021-03-12 | 福州大学 | FBG sensing network node fault positioning method based on twin node auxiliary sensing |
| CN112508060A (en) * | 2020-11-18 | 2021-03-16 | 哈尔滨工业大学（深圳） | Landslide mass state judgment method and system based on graph convolution neural network |
| WO2022171067A1 (en) * | 2021-02-09 | 2022-08-18 | 北京有竹居网络技术有限公司 | Video processing method and apparatus, and storage medium and device |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108573276A (en) * | 2018-03-12 | 2018-09-25 | 浙江大学 | A kind of change detecting method based on high-resolution remote sensing image |
| CN110321859A (en) * | 2019-07-09 | 2019-10-11 | 中国矿业大学 | A kind of optical remote sensing scene classification method based on the twin capsule network of depth |
-
2019
- 2019-10-29 CN CN201911034777.9A patent/CN110807400A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108573276A (en) * | 2018-03-12 | 2018-09-25 | 浙江大学 | A kind of change detecting method based on high-resolution remote sensing image |
| CN110321859A (en) * | 2019-07-09 | 2019-10-11 | 中国矿业大学 | A kind of optical remote sensing scene classification method based on the twin capsule network of depth |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112508060A (en) * | 2020-11-18 | 2021-03-16 | 哈尔滨工业大学（深圳） | Landslide mass state judgment method and system based on graph convolution neural network |
| CN112508060B (en) * | 2020-11-18 | 2023-08-08 | 哈尔滨工业大学（深圳） | Landslide body state judging method and system based on graph convolution neural network |
| CN112491468A (en) * | 2020-11-20 | 2021-03-12 | 福州大学 | FBG sensing network node fault positioning method based on twin node auxiliary sensing |
| WO2022171067A1 (en) * | 2021-02-09 | 2022-08-18 | 北京有竹居网络技术有限公司 | Video processing method and apparatus, and storage medium and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108846835B (en) | Image change detection method based on depth separable convolutional network | |
| WO2020125216A1 (en) | Pedestrian re-identification method, device, electronic device and computer-readable storage medium | |
| CN110807400A (en) | Twin network-based collapse hidden danger characteristic information extraction method | |
| CN102385592B (en) | Image concept detection method and device | |
| CN106951863B (en) | Method for detecting change of infrared image of substation equipment based on random forest | |
| CN113284144B (en) | Tunnel detection method and device based on unmanned aerial vehicle | |
| CN111008576A (en) | Pedestrian detection and model training and updating method, device and readable storage medium thereof | |
| CN113469050A (en) | Flame detection method based on image subdivision classification | |
| CN111652846B (en) | Semiconductor defect identification method based on characteristic pyramid convolution neural network | |
| CN115131747A (en) | Knowledge distillation-based power transmission channel engineering vehicle target detection method and system | |
| CN111539456A (en) | Target identification method and device | |
| CN111241905A (en) | Power transmission line nest detection method based on improved SSD algorithm | |
| CN111931721B (en) | Method and device for detecting color and number of annual inspection label and electronic equipment | |
| Yan et al. | The research of building earthquake damage object-oriented change detection based on ensemble classifier with remote sensing image | |
| CN116823793A (en) | Device defect detection method, device, electronic device and readable storage medium | |
| CN110033038B (en) | Intelligent selection algorithm for aerodynamic heat test data | |
| CN111353412A (en) | End-to-end 3D-CapsNet flame detection method and device | |
| CN115937492A (en) | Transformer equipment infrared image identification method based on feature identification | |
| CN113609913B (en) | Pine wood nematode disease tree detection method based on sampling threshold interval weighting | |
| CN109117841A (en) | Scene text detection method based on stroke width transformation and convolutional neural networks | |
| CN115272882A (en) | Discrete building detection method and system based on remote sensing image | |
| CN114186784A (en) | Electrical examination scoring method, system, medium and device based on edge calculation | |
| CN112241694A (en) | Method for identifying unworn safety belt based on CenterNet | |
| Jing | Insulator Detection Study Based on Improved Faster-RCNN | |
| CN113378787B (en) | Intelligent traffic electronic prompting device detection method and system based on multi-feature vision |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200218 |
|
| RJ01 | Rejection of invention patent application after publication |