CN112651314A - Automatic landslide disaster-bearing body identification method based on semantic gate and double-temporal LSTM - Google Patents
Automatic landslide disaster-bearing body identification method based on semantic gate and double-temporal LSTM Download PDFInfo
- Publication number
- CN112651314A CN112651314A CN202011497515.9A CN202011497515A CN112651314A CN 112651314 A CN112651314 A CN 112651314A CN 202011497515 A CN202011497515 A CN 202011497515A CN 112651314 A CN112651314 A CN 112651314A
- Authority
- CN
- China
- Prior art keywords
- semantic
- network
- gate
- lstm
- remote sensing
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/176—Urban or other man-made structures
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- 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
- G06V10/267—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 by performing operations on regions, e.g. growing, shrinking or watersheds
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/188—Vegetation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Multimedia (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Image Analysis (AREA)
- Emergency Alarm Devices (AREA)
Abstract
The invention discloses a landslide disaster-bearing body automatic identification method based on a semantic gate and a double-temporal LSTM, which comprises the steps of data preparation, network structure and parameter setting, comprehensive error calculation, network training and verification, network effect evaluation, landslide and disaster-bearing body prediction and the like. The method can timely and accurately identify the landslide and the disaster-bearing body thereof in the disaster-affected area, further quickly acquire information such as disaster grade, disaster-affected range and the like, and can perform disaster situation evaluation based on the information to guide emergency rescue work after disaster.
Description
Technical Field
The invention belongs to the technical field of geographic information systems, relates to a method for quickly and accurately identifying landslide disasters and disaster-bearing bodies thereof, and particularly relates to a landslide disaster-bearing body automatic identification method based on a semantic gate and a double-temporal long-short term cyclic network (SG-BiTLSTM network).
Background
The Long-Short Term circulation network based on semantic gate and double tenses is composed of a U-Net network and two mutually coupled Long-Short Term Memory (LSTM) networks, wherein the U-Net network is one of semantic segmentation networks and is used for outputting image features and semantic segmentation maps; the two LSTM networks are used for outputting sentences describing the spatial relationship among the remote sensing objects.
The U-Net is in a U-shaped structure and consists of a compression path and an expansion path, and in the compression path, the size of an input characteristic diagram is gradually reduced through convolution operation, so that the classification precision of various remote sensing objects is improved; and in the expanding process, the size of the feature map is gradually reduced through deconvolution operation. However, the convolution operation discards a large amount of spatial information between remote sensing objects, and although the information can be restored in the deconvolution process, the basic information of the deconvolution is less, so that the object cannot be accurately positioned only by the characteristics of the deconvolution restoration. In order to further improve the positioning accuracy of the objects, a skip layer link mode is adopted in the U-Net network, namely, the feature graph before convolution operation is carried out in each layer in the compression process is directly spliced with the feature graph in the corresponding deconvolution layer, so that richer space information between the objects is obtained, and the positioning of the objects is facilitated.
In conclusion, the U-Net network realizes accurate classification of various remote sensing objects by utilizing convolution operation on one hand; and on the other hand, accurate positioning of the object is realized by utilizing the jump layer connection.
The long-short term memory network LSTM is a kind of recurrent neural network, and both of the long-short term memory network LSTM and the long-short term memory network LSTM belong to the category of deep learning. LSTM has a chain structure of repeating neural network modules suitable for processing and predicting events with long intervals and delays in time series. The specific memory and forgetting mode of the LSTM enables the LSTM to effectively adapt to the time sequence characteristics in the network learning process, and makes full use of historical information to establish a time dependence relationship. The LSTM can effectively preserve history information compared to the conventional RNN, and thus can be more widely used.
Compared with the traditional RNN, the LSTM network has the advantages that the hidden layer of the LSTM is not a common neuron any more, but a memory unit with a single memory mode.
Disclosure of Invention
In order to realize the automatic identification of the landslide and the disaster-bearing body thereof, the invention provides a method for automatically identifying the landslide and the disaster-bearing body thereof based on a semantic gate and a double-temporal LSTM.
The technical scheme adopted by the invention is as follows: a landslide disaster-bearing body automatic identification method based on semantic gate and double-temporal LSTM is characterized by comprising the following steps:
step 1: cutting the whole remote sensing image and making a sample;
manufacturing a corresponding group Truth according to the original remote sensing image, wherein different colors correspond to different types of remote sensing objects in the original image, and the remote sensing objects comprise four types of remote sensing objects of landslide, agriculture, greenbelt and buildings; cutting the remote sensing object into a plurality of samples with preset sizes;
step 2: constructing a double-temporal long-short-term cyclic network based on a semantic gate, and setting parameters;
the double-temporal long-short term circulation network based on the semantic gate is composed of a semantic segmentation network U-Net network and two long-short term memory networks LSTM networks which are coupled with each other; the two LSTM networks are used for outputting sentences describing the spatial relationship between the remote sensing objects;
the semantic gate-based dual-temporal long-short-term circulation network is composed of a Language LSTM and a Prediction LSTM;
designing a semantic door mechanism;
the semantic gate mechanism adopts a multilayer perceptron structure, and the hidden layer information h at t moment predicted by Prediction LSTM at t-1 momentt 2As input at time t; the structure is activated by using a sigmoid and a custom activation function respectively;
setting a double-temporal long-short term cyclic network comprehensive error function based on a semantic gate;
the error of the semantic gate-based dual-temporal long-short-term circulation network is divided into three parts, namely an error Loss1 of the Languge LSTM network at the current moment, an error Loss 2 of the Prediction LSTM network at the previous moment at the current moment and a cross entropy Loss3 between the object mask and the attention area matrix; loss1 and Loss 2 can enable the Language LSTM network to comprehensively consider the output of the Language LSTM network and the Prediction LSTM network when generating words at the current moment; the Loss3 is used for improving the positioning precision of the remote sensing object; the information of two tenses is integrated through Loss1 and 2, and the location is corrected through Loss3, so that the location precision of the model is improved, and the capability of autonomously determining the attention of remote sensing image information or context information is improved;
and step 3: training a double-temporal long-short-term circulation network based on a semantic gate;
firstly, pre-training a semantic segmentation network U-Net network, then carrying out comprehensive training on the semantic segmentation network U-Net network and two mutually coupled long-short term memory network LSTM networks, wherein the input in the training process is a semantic segmentation graph output by the U-Net network, and the output is a sentence describing the spatial relationship between remote sensing objects; obtaining a trained dual-temporal long-short term cyclic network based on a semantic gate;
and 4, step 4: predicting a landslide disaster bearing body;
and scanning and splicing the samples obtained by cutting line by line, and inputting the samples into a trained semantic gate-based dual-temporal long-short-term cyclic network to predict the landslide and the disaster-bearing body thereof.
Compared with the prior art, the invention has the beneficial effects that: the double-temporal long-short-term circulation network based on the semantic gate can solve the problem of error accumulation in a prediction stage to a certain extent, and meanwhile, the automatic identification of a disaster-bearing body is realized by utilizing a spatial relationship through combining a focusing matrix of the double-temporal LSTM and a semantic segmentation graph output by U-Net. Compared with the traditional method for identifying the landslide disaster-bearing body by utilizing the GIS spatial analysis technology, the method provided by the invention has the characteristics of fewer manual intervention links and higher disaster sensing efficiency, and is beneficial to the rapid development of emergency rescue work after disasters. The semantic door mechanism can realize that the model is controlled to dynamically and adaptively select the dependent image information or the context semantic information according to the prediction result at the previous moment, thereby greatly improving the identification precision of the object and the spatial relationship thereof.
Drawings
FIG. 1 is a flow chart of an embodiment of the present invention;
FIG. 2 is a sample landslide image in an experimental area according to an embodiment of the present invention;
FIG. 3 is a block diagram of a semantic gate based dual temporal long short term cyclic network model according to an embodiment of the present invention;
FIG. 4 is a block diagram of a dual-temporal LSTM network according to an embodiment of the present invention;
fig. 5 is a diagram illustrating the effect of landslide and disaster-bearing body identification based on semantic gate and a dual-temporal long-short term cyclic network according to an embodiment of the present invention.
Detailed Description
In order to facilitate the understanding and implementation of the present invention for those of ordinary skill in the art, the present invention is further described in detail with reference to the accompanying drawings and examples, it is to be understood that the embodiments described herein are merely illustrative and explanatory of the present invention and are not restrictive thereof.
Referring to fig. 1, the method for automatically identifying a landslide disaster-bearing body based on a semantic gate and a dual-temporal LSTM provided by the invention comprises the following steps:
step 1: cutting the whole remote sensing image and making a sample;
in this embodiment, a corresponding group Truth is manufactured according to an original remote sensing image, wherein different colors correspond to different types of remote sensing objects in the original image, including four types of remote sensing objects of landslide, agriculture, greenbelt and building; and cutting the remote sensing object into a plurality of samples with preset sizes.
In the embodiment, the remote sensing image is scanned line by using a self-programming program, each 224 × 224 pixels are cut into a sample, and the spatial resolution of 0.5 m is kept for the pixels in the sample.
Please refer to fig. 2, which shows a sample block with an original image on the left side; the right side is GT (ground Truth) corresponding to the original image, wherein different colors correspond to different types of remote sensing objects in the original image. Four types of remote sensing objects of landslide, agriculture, greenbelt and buildings are included in the sample. The sample size is 224 x 224 pixels and the spatial resolution remains constant at 0.5 meters.
Step 2: constructing a double-temporal long-short-term cyclic network based on a semantic gate, and setting network parameters;
referring to fig. 3, the semantic gate-based dual-temporal long-short term cyclic network of the present embodiment is composed of a semantic segmentation network (U-Net network) and two long-short term memory networks (dual-temporal LSTM network) coupled to each other; the system comprises a U-Net network, a dual-temporal LSTM network and a remote sensing object, wherein the U-Net network is used for outputting image characteristics and a semantic segmentation graph, and the dual-temporal LSTM network is used for outputting sentences describing spatial relations among the remote sensing objects;
in the embodiment, in the experimental process, the attention matrix is used for fusing the spatial relationship between the object mask generated by the semantic segmentation network U-Net and the object output by the dual-temporal LSTM network, so that the landslide and the disaster-bearing body thereof can be automatically identified based on the spatial relationship between the objects. In addition, the semantic gate-based dual-temporal long-short term cyclic network can dynamically and adaptively select and rely on remote sensing image characteristics or semantic information in the process of generating semantic annotation sentences. The parameter quantity of the U-Net network in the double-temporal long-short-term cyclic network based on the semantic gate is 864 ten thousand, and the parameter quantity of the double-temporal LSTM network is 24 thousand.
Referring to fig. 4, the dual-temporal LSTM network of the present embodiment is composed of a Language LSTM and a Prediction LSTM; word is not generated in Language LSTM at time tRelying only on the hidden layer information h at the moment immediately before itt-1 1Meanwhile, the semantic gate refers to the t moment hidden layer information h predicted by the Prediction LSTM at the t-1 momentt 2Therefore, the semantic annotation of the Language LSTM at the time t integrates the effects of the two networks at different times, so that the problem of error accumulation in the prediction stage is solved.
The embodiment designs a semantic door mechanism;
in order to enable the double-temporal long-short-term cyclic network based on the semantic gate to be capable of adaptively processing remote sensing image information, a semantic gate mechanism is designed, vocabulary types are generated at different moments, and the training network has double-temporal and adaptive image feature extraction and processing capabilities. The semantic gate mechanism adopts a multilayer perceptron structure, and the hidden layer information h at t moment predicted by Prediction LSTM at t-1 momentt 2As input at time t; the structure is activated by using a sigmoid and a custom activation function respectively;
in order to improve the positioning accuracy of the landslide body and the disaster-bearing body thereof, the embodiment designs the GT manufacturing strategy, and synthesizes the GT function with the double-temporal Loss function, so that the model can comprehensively and accurately interpret the landslide body, the disaster-bearing body and the spatial relationship thereof.
The implementation sets a double-temporal long-short-term cyclic network comprehensive error function based on a semantic gate; the error of the double-tense LSTM network is divided into three parts, namely an error Loss1 of the Language LSTM network at the current moment, an error Loss 2 of the Prediction LSTM network at the previous moment at the current moment and a cross entropy Loss3 between the object mask and the attention area matrix; loss1 and Loss 2 can enable the LSTM network to comprehensively consider the output of the Language LSTM and the Prediction LSTM network when generating words at the current moment; the Loss3 is used for improving the positioning precision of the remote sensing object; through the integration of two tense information (double tense effect) by Loss1 and 2, Loss3 corrects positioning, thereby improving the positioning accuracy of the model and the ability of autonomously determining to focus on remote sensing image information or context information;
and step 3: training and verifying a double-temporal long-short-term cyclic network based on a semantic gate;
firstly, pre-training a semantic segmentation network U-Net network, then carrying out comprehensive training on the semantic segmentation network U-Net network and two mutually coupled long-short term memory networks (dual-temporal LSTM networks), verifying and evaluating a network model from multiple aspects such as semantic precision, stability and positioning precision of a remote sensing object, and analyzing and evaluating the action effect of a semantic gate; and stopping training when the result achieves the expected effect, thereby obtaining the well-trained dual-temporal long-short term cyclic network based on the semantic gate.
In this embodiment, the semantic segmentation network U-Net is pre-trained, and the input in the process is a remote sensing image sample and the output is a semantic segmentation map corresponding to the remote sensing image sample.
In the embodiment, the semantic segmentation network U-Net is pre-trained, and then is comprehensively trained with two mutually coupled long-short term memory network LSTM networks, wherein the input in the training process is a semantic segmentation graph output by the U-Net network, and the output is a sentence describing the spatial relationship between remote sensing objects.
In this embodiment, the number of iterations of the integrated training is 1600, and the learning rate of the dual-temporal LSTM is 0.001.
And 4, step 4: predicting a landslide disaster bearing body;
and scanning and splicing the samples line by line, and inputting the samples into a trained semantic gate-based dual-temporal long-short term cyclic network to predict the landslide and the disaster-bearing body thereof.
In the embodiment, the remote sensing image is scanned line by using a self-programming sequence, each 224 × 224 pixels are cut into a sample, and the pixels in the sample keep the original spatial resolution of 0.5 m. And then, inputting the sample into a trained semantic gate-based dual-temporal long-short term cycle network model to predict the landslide and the disaster-bearing body thereof, thereby providing basic geographic data support for emergency resource sharing after landslide disaster.
Please refer to fig. 5, which is a diagram of the recognition effect of landslide and disaster-bearing bodies of the semantic gate and the dual-temporal long-short term cyclic network of the present embodiment, wherein a is a remote sensing image in the whole research area, and b, c, and d all correspond to the area in the red square frame in a. b is the original image of the area; c is a semantic segmentation graph of the region; and d, the yellow curve coil in the step d is the landslide disaster carrier identified by the double-temporal long-short term cyclic network based on the semantic gate.
The embodiment also analyzes and evaluates the effect of the double-temporal long-short-term circulation network based on the semantic gate; the method comprises the following steps of performing an experiment on a double-temporal long-short-term circulation network based on a semantic gate, analyzing an experiment result by using a verification set randomly distributed from an image sample, explaining the contrast advantage of the experiment result with a traditional LSTM model, and verifying the stability of the result by using a Monte Carlo experiment; the results show that: compared with the traditional LSTM network, the semantic gate-based dual-temporal long-short-term cyclic network has a larger improvement in the two aspects of the identification precision of the remote sensing object and the positioning precision of the attention area.
It should be understood that the above description of the preferred embodiments is given for clarity and not for any purpose of limitation, and that various changes, substitutions and alterations can be made herein without departing from the spirit and scope of the invention as defined by the appended claims.
Claims (2)
1. A landslide disaster-bearing body automatic identification method based on semantic gate and double-temporal LSTM is characterized by comprising the following steps:
step 1: cutting the whole remote sensing image and making a sample;
manufacturing a corresponding group Truth according to the original remote sensing image, wherein different colors correspond to different types of remote sensing objects in the original image, and the remote sensing objects comprise four types of remote sensing objects of landslide, agriculture, greenbelt and buildings; cutting the remote sensing object into a plurality of samples with preset sizes;
step 2: constructing a double-temporal long-short-term cyclic network based on a semantic gate, and setting parameters;
the double-temporal long-short term circulation network based on the semantic gate is composed of a semantic segmentation network U-Net network and two long-short term memory networks LSTM networks which are coupled with each other; the two LSTM networks are used for outputting sentences describing the spatial relationship between the remote sensing objects;
the semantic gate-based dual-temporal long-short-term circulation network is composed of a Language LSTM and a Prediction LSTM;
designing a semantic door mechanism;
the semantic gate mechanism adopts a multilayer perceptron structure, and the hidden layer information h at t moment predicted by Prediction LSTM at t-1 momentt 2As input at time t; the structure is activated by using a sigmoid and a custom activation function respectively;
setting a double-temporal long-short term cyclic network comprehensive error function based on a semantic gate;
the error of the semantic gate-based dual-temporal long-short-term circulation network is divided into three parts, namely an error Loss1 of the Languge LSTM network at the current moment, an error Loss 2 of the Prediction LSTM network at the previous moment at the current moment and a cross entropy Loss3 between the object mask and the attention area matrix; loss1 and Loss 2 can enable the Language LSTM network to comprehensively consider the output of the Language LSTM network and the Prediction LSTM network when generating words at the current moment; the Loss3 is used for improving the positioning precision of the remote sensing object; the information of two tenses is integrated through Loss1 and 2, and the location is corrected through Loss3, so that the location precision of the model is improved, and the capability of autonomously determining the attention of remote sensing image information or context information is improved;
and step 3: training a double-temporal long-short-term circulation network based on a semantic gate;
firstly, pre-training a semantic segmentation network U-Net network, then carrying out comprehensive training on the semantic segmentation network U-Net network and two mutually coupled long-short term memory network LSTM networks, wherein the input in the training process is a semantic segmentation graph output by the U-Net network, and the output is a sentence describing the spatial relationship between remote sensing objects; obtaining a trained dual-temporal long-short term cyclic network based on a semantic gate;
and 4, step 4: predicting a landslide disaster bearing body;
and scanning and splicing the samples obtained by cutting line by line, and inputting the samples into a trained semantic gate-based dual-temporal long-short-term cyclic network to predict the landslide and the disaster-bearing body thereof.
2. The method for automatically identifying a landslide disaster-bearing body based on semantic gate and bi-temporal LSTM according to claim 1, wherein: and 3, verifying and evaluating the trained dual-temporal long-short-term circulation network based on the semantic gate from the semantic accuracy, the stability and the positioning accuracy of the remote sensing object, analyzing and evaluating the action effect of the semantic gate, and stopping training when the result reaches the expected effect, thereby obtaining the trained dual-temporal long-short-term circulation network based on the semantic gate.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011497515.9A CN112651314A (en) | 2020-12-17 | 2020-12-17 | Automatic landslide disaster-bearing body identification method based on semantic gate and double-temporal LSTM |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011497515.9A CN112651314A (en) | 2020-12-17 | 2020-12-17 | Automatic landslide disaster-bearing body identification method based on semantic gate and double-temporal LSTM |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112651314A true CN112651314A (en) | 2021-04-13 |
Family
ID=75354830
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011497515.9A Pending CN112651314A (en) | 2020-12-17 | 2020-12-17 | Automatic landslide disaster-bearing body identification method based on semantic gate and double-temporal LSTM |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112651314A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113378582A (en) * | 2021-07-15 | 2021-09-10 | 重庆交通大学 | Landslide displacement prediction model and method based on semantic information driving |
| CN115131684A (en) * | 2022-08-25 | 2022-09-30 | 成都国星宇航科技股份有限公司 | Landslide identification method and device based on satellite data UNet network model |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110490081A (en) * | 2019-07-22 | 2019-11-22 | 武汉理工大学 | A kind of remote sensing object decomposition method based on focusing weight matrix and mutative scale semantic segmentation neural network |
| WO2020244287A1 (en) * | 2019-06-03 | 2020-12-10 | 中国矿业大学 | Method for generating image semantic description |
-
2020
- 2020-12-17 CN CN202011497515.9A patent/CN112651314A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020244287A1 (en) * | 2019-06-03 | 2020-12-10 | 中国矿业大学 | Method for generating image semantic description |
| CN110490081A (en) * | 2019-07-22 | 2019-11-22 | 武汉理工大学 | A kind of remote sensing object decomposition method based on focusing weight matrix and mutative scale semantic segmentation neural network |
Non-Patent Citations (1)
| Title |
|---|
| WENQI CUI 等: "Landslide Image Captioning Method Based on Semantic Gate and Bi-Temporal LSTM", 《ISPRS INTERNATIONAL JOURNAL OF GEO-INFORMATION》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113378582A (en) * | 2021-07-15 | 2021-09-10 | 重庆交通大学 | Landslide displacement prediction model and method based on semantic information driving |
| CN113378582B (en) * | 2021-07-15 | 2022-04-26 | 重庆交通大学 | Landslide displacement prediction model construction and use method based on semantic information driving |
| CN115131684A (en) * | 2022-08-25 | 2022-09-30 | 成都国星宇航科技股份有限公司 | Landslide identification method and device based on satellite data UNet network model |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109492830B (en) | Mobile pollution source emission concentration prediction method based on time-space deep learning | |
| WO2022095682A1 (en) | Text classification model training method, text classification method and apparatus, device, storage medium, and computer program product | |
| CN112634292A (en) | Asphalt pavement crack image segmentation method based on deep convolutional neural network | |
| CN111914885B (en) | Multi-task personality prediction method and system based on deep learning | |
| CN110321361B (en) | Test question recommendation and judgment method based on improved LSTM neural network model | |
| CN104217216A (en) | Method and device for generating detection model, method and device for detecting target | |
| CN112651314A (en) | Automatic landslide disaster-bearing body identification method based on semantic gate and double-temporal LSTM | |
| CN115131627B (en) | Construction and training method of lightweight plant disease and pest target detection model | |
| CN112116137A (en) | Student class dropping prediction method based on mixed deep neural network | |
| CN111062411A (en) | Method, apparatus and device for identifying multiple compounds from mass spectrometry data | |
| CN115661505A (en) | Semantic perception image shadow detection method | |
| CN115659966A (en) | Rumor detection method and system based on dynamic heteromorphic graph and multi-level attention | |
| CN115908793A (en) | Coding and decoding structure semantic segmentation model based on position attention mechanism | |
| CN115561834A (en) | Meteorological short-term and temporary forecasting all-in-one machine based on artificial intelligence | |
| CN115294337A (en) | Method for training semantic segmentation model, image semantic segmentation method and related device | |
| CN116052419A (en) | Deep learning-based graph neural network traffic flow prediction method | |
| CN114565149A (en) | CGA fusion model-based time series data prediction method and device and computer equipment | |
| Sonawane et al. | ChatBot for college website | |
| CN112579777A (en) | Semi-supervised classification method for unlabelled texts | |
| CN114548382B (en) | Migration training method, device, equipment, storage medium and program product | |
| CN116883709A (en) | Carbonate fracture-cavity identification method and system based on channel attention mechanism | |
| CN116091763A (en) | Apple leaf disease image semantic segmentation system, segmentation method, device and medium | |
| CN113221993A (en) | Large-view-field small-sample target detection method based on meta-learning and cross-stage hourglass | |
| Wu | Sequential images prediction using convolutional LSTM with application in precipitation nowcasting | |
| KR102713582B1 (en) | Device and method for extraction of damage mechanism from bridge inspection reports |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210413 |