CN113095437A - Fire point detection method for Himapari-8 remote sensing data - Google Patents
Fire point detection method for Himapari-8 remote sensing data Download PDFInfo
- Publication number
- CN113095437A CN113095437A CN202110476630.6A CN202110476630A CN113095437A CN 113095437 A CN113095437 A CN 113095437A CN 202110476630 A CN202110476630 A CN 202110476630A CN 113095437 A CN113095437 A CN 113095437A
- Authority
- CN
- China
- Prior art keywords
- fire
- data
- remote sensing
- fire point
- sensing data
- 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.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic 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
- G06N3/048—Activation functions
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Probability & Statistics with Applications (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Fire Alarms (AREA)
Abstract
The invention discloses a fire point detection method for Himapari-8 remote sensing data, and belongs to the technical field of fire point detection of remote sensing data. Which comprises the following steps: preprocessing the Himapari-8 satellite remote sensing data to obtain pixels and neighborhoods thereof and manufacturing a data set; building a double-input convolutional neural network model, and respectively inputting a radiation temperature wave band and a reflectivity wave band into the network; training a deep learning model by using the data set, iteratively updating network model parameters according to the change condition of the loss function, and converging to obtain an optimal network model; the obtained model is used for testing the fire point detection of the centralized data. Compared with the traditional context relative threshold algorithm, the method can automatically judge the fire point and effectively improve the accuracy of the fire point detection of the remote sensing data.
Description
Technical Field
The invention relates to the technical field of remote sensing data fire point detection, in particular to a fire point detection method of Himapari-8 (namely, a sunflower-8 meteorological satellite) remote sensing data.
Background
In recent years, fires have been more frequently generated in places such as forests, farmlands, and pastures on the earth. More recent serious fire events include a california fire in the united states of 11 months in 2018, a brazilian amazon forest fire in 8 months in 2019, and an australian jungle forest fire that continues from 2019 to 2010. Furthermore, fire occurs periodically every year in parts of africa and europe.
The occurrence of a fire may be caused by the natural burning of objects in the ecological environment, or may be caused by human deliberate or inadvertent efforts. A fire disaster can generate a large amount of dense smoke and harmful gases, and release a large amount of carbon dioxide, thereby seriously polluting the air environment; meanwhile, the ecological environment is damaged, a large number of wild animals die, and residents living nearby are also affected.
The current fire detection method aiming at satellite remote sensing data mainly adopts various threshold algorithms, including a bispectrum algorithm, a fixed threshold algorithm, a dynamic threshold algorithm of space context and a multi-temporal detection algorithm. The threshold algorithm determines the threshold of each waveband data through a statistical method to eliminate non-fire pixels such as cloud, water and non-fire areas, and determines fire pixels according to a temperature threshold. However, the fire detection accuracy of the algorithms is low, and particularly for small-area fires or fires with low temperature, more false alarms and fire missing points exist, and the false alarm rate is high. In addition, this uncertainty results in the inability of the threshold algorithm to be applied to all regions due to inconsistent climate and geographic conditions across the globe.
Disclosure of Invention
The invention aims to provide a fire point detection method of Himapari-8 remote sensing data, which is based on deep learning, can learn characteristics hidden in data in a plurality of remote sensing data, automatically find out a proper threshold value, greatly improve the accuracy of fire point detection and timely acquire relevant information of a fire.
In order to achieve the purpose, the invention adopts the technical scheme that:
a fire point detection method for Himapari-8 remote sensing data comprises the following steps:
acquiring original remote sensing data of a Himapari-8 satellite and corresponding fire point data, wherein the original remote sensing data comprises longitude and latitude and data of each wave band, and the fire point data comprises longitude and latitude information of a fire point;
randomly extracting N data blocks containing fire point longitude and latitude from the original remote sensing data as positive samples, and randomly extracting N data blocks containing no fire point longitude and latitude from the original remote sensing data as negative samples;
step three, disordering the extracted positive and negative samples, and randomly dividing the positive and negative samples into a training set and a testing set;
establishing a double-input network model, wherein the network model comprises a fire detection network branch, a cloud water detection network branch and a fusion module; the fire detection network branch is used for processing a temperature radiation wave band related to fire detection, the cloud water detection network branch is used for processing a reflectivity wave band related to cloud water detection, the fire detection network branch and the cloud water detection network branch both comprise a wave band attention module and a pixel attention module, output data of the two branches are transmitted to a fusion module, and the fusion module is used for outputting a fire detection result;
step five, training the dual-input network model established in the step four by using the training set and the test set obtained in the step three to obtain a trained model;
and step six, dividing the original remote sensing data of the Himapwari-8 satellite to be detected into blocks with the same size as the positive and negative sample data blocks, inputting the blocks into a trained model, and detecting whether fire points exist in the blocks.
Further, the data block size of each of the positive and negative samples is 21 × 21 pixels, and the extraction bands of each of the positive and negative samples include albedo _03, albedo _04, albedo _06, tbb _07, tbb _14, tbb _15, SOZ, latitude, and longtude.
Further, the fire detection network branch comprises a pixel attention module, a wave band attention module, a convolution block and a convolution block which are connected in series in sequence; the cloud water detection network branch comprises a pixel attention module, a wave band attention module, a rolling block and a rolling block which are sequentially connected in series.
Further, in the waveband attention module, data input into the waveband attention module are respectively input into a global average pooling layer and a global maximum pooling layer, then the pooling results of the global average pooling layer and the global maximum pooling layer are subjected to element-level addition, then a 1 × 1 convolution and ReLU activation function are connected, and a 1 × 1 convolution block and a Sigmoid activation function are connected to obtain attention characteristics; and performing element-level multiplication on the attention feature and the data of the input waveband attention module to obtain the waveband attention.
Further, in the pixel attention module, inputting data input into the pixel attention module into two cascaded 1 × 1 convolutions to obtain a feature map with the number of channels being 1, wherein the first 1 × 1 convolution has an activation function, and the first 1 × 1 convolution has a Sigmoid activation function; and then, expanding the feature map with the channel number of 1 into the size of the data of the input pixel attention module through copy operation, and performing element-level multiplication on the expanded feature map and the data of the input pixel attention module to obtain the pixel attention.
Further, in the fusion module, output results of the fire detection network branch and the cloud water detection network branch are respectively input into the corresponding full connection layers, element-level multiplication is performed on outputs of the two full connection layers to obtain a fusion result, and the fusion result is input into one full connection layer and a subsequent SoftMax function to obtain a fire detection result.
Further, when training is carried out in the fifth step, cross entropy loss is selected as a loss function, an Adam optimizer is selected for optimization, the initial learning rate is set to be 0.0001, and a multistep LR adjustment strategy is adopted to adjust the learning rate; inputting the training set samples into a network model for training, and obtaining an optimal model for fire point detection according to the change condition of the loss value; then inputting the test set sample into the trained model to obtain the discrimination confidence of each fire point data, and if the confidence is greater than 0.5, judging that the sample belongs to the fire point; and (5) counting the accuracy of fire point detection and verifying the accuracy of the model.
The invention has the beneficial effects that:
1. the method is based on deep learning, can effectively improve the accuracy, has higher running speed, can acquire the relevant information of the fire in time, and can effectively assist the early warning, prevention and control of the forest fire.
2. The band attention and pixel attention structure provided by the invention can better extract the characteristics of fire data, has stronger expression capability compared with a common convolution block, and can better distinguish fire point characteristics from non-fire point characteristics by the constructed double-input network structure, thereby reducing misjudgment and false alarm.
3. The method has high accuracy and easy realization, and can be quickly expanded to the fire detection task of remote sensing data of other similar satellites.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used for describing the embodiments will be briefly introduced below.
FIG. 1 is a flow chart of a fire detection method according to an embodiment of the present invention.
Fig. 2 is a network structure diagram of a dual-input network model in the embodiment of the present invention.
FIG. 3 is a block diagram of a band attention module and a pixel attention module in an embodiment of the invention.
Fig. 4 is a graph of the loss variation of the training process in the embodiment of the present invention.
Detailed description of the invention
The technical solution of the present invention will be described more clearly and completely with reference to the accompanying drawings and the detailed description.
As shown in fig. 1, a method for detecting a fire point of himwari-8 remote sensing data comprises the following steps:
and step 1, extracting a data block.
Firstly, the acquired Himapari-8 satellite remote sensing data is subjected to wave band selection, wherein the selected wave bands are respectively the 3 rd, 4 th, 6 th, 7 th, 14 th and 15 th wave bands, and SOZ (solar zenith angle) and longitude and latitude data are obtained.
The 3 rd waveband is an albedo _03 attribute, is a reflectivity waveband and is sensitive to elements such as land, cloud and the like; the 4 th waveband is an albedo-04 waveband which is a reflectivity waveband and is sensitive to elements such as ocean, water and the like; the 6 th waveband is an albedo-06 waveband which is a reflectivity waveband and is sensitive to detection of the cloud layer; the 7 th, 14 th and 15 th wave bands are tbb _07, tbb _14 and tbb _15 wave bands respectively, are brightness temperature wave bands and are sensitive to the temperature of the earth surface and the temperature of the cloud layer.
The method reserves an SOZ wave band, wherein the SOZ is a Solar zenith angle (Solar zenith angle), represents an included angle between the incident Solar ray and the normal line of the ground plane of the pixel at the position of the pixel, has a value range of 0-90 degrees, and is mainly used for determining local time as day or night. The longitude and latitude wave bands are latitude wave bands and longtude wave bands in the data, and are mainly used for positioning by referring to longitude and latitude when a data set is manufactured, and corresponding fire point data blocks and non-fire point data blocks can be selected.
The size range of the manufactured data block is set to be 21 x 21, model calculation can be carried out when the size range is set to be any odd number, the larger the data block is, the more calculation time is needed, the result is relatively more accurate, but the problem that training is difficult exists. While smaller blocks operate faster, but the result is coarser, weighing both factors, suggesting that an odd number between 9-25 is chosen to construct the block.
And for the positive sample, all fire points on global fire point data at a certain time are obtained, and a corresponding fire point data block is obtained through longitude and latitude positioning. Meanwhile, for the acquisition of the non-fire point data blocks, the non-fire point data blocks with the same size are sampled and cut on the data file at the same time in a random sampling mode to serve as negative samples. The number of positive samples is chosen to be consistent with the negative samples.
And 2, constructing a data set.
After the fire point data blocks and the non-fire point data blocks are obtained, the fire point data blocks and the non-fire point data blocks need to be divided to obtain a training set and a test set. The division ratio of the training set to the test set is 7: and 3, selecting fire point data blocks and non-fire point data blocks by adopting a random sampling method during division. The random division can enable the data to be uniformly distributed in the training set and the testing set, so that the model learns more distribution, and the accuracy of the model is improved.
The training set serves to train the model so that the model can learn useful classes of parameters from the data to fit the neural network model. The test set is used for verifying the quality of the model obtained by training, evaluating the performance of the learning model, and obtaining the accuracy through calculation to judge the quality of the model training.
As shown in fig. 2, the dual-input network model is composed of three parts, the upper half is a fire point evaluation branch, radiation temperature band data is input, the lower half is a cloud water detection branch, corresponding reflectivity bands are input, and the final fusion module fuses the results. For the fire point evaluation branch and the cloud water detection branch, a wave band attention module and a pixel attention module are respectively designed, as shown in fig. 3, the two attention modules can effectively extract the characteristics of fire point data, and the accuracy of fire point detection is improved.
Step 4, training and parameter setting
Model training was performed using a GPU, specifically the machine model number invida (NVIDIA) TITAN V. The programming uses the Python language and the Pytorch deep learning framework.
The learning rate is the magnitude of the update of the control weights during each iteration. During the training, the initial learning rate was set to 0.001 and the multistep lr method was used to adjust the learning rate. As training progresses, the learning rate needs to be reduced to better reach the optimum value as the network approaches convergence.
The method sets the epoch to be 150, the loss function adopts cross entropy loss, and the optimizer adopts an Adam optimization algorithm. During the training process, the mini-batch mode is selected, wherein the batch _ size is set to 8, that is, 8 data blocks are simultaneously sent into the network each time. When too many feeds are made, the network will not converge, while when too few feeds are made, the network will converge too slowly and may fall into a local extreme.
As shown in fig. 4, during the training process, a tensorbard is selected to record the change of the loss value, so as to determine whether the network is converged by training. After the network training is finished, the network model parameters are reserved for testing and application.
And 6, testing.
The test uses the accuracy as an evaluation index, and TP is used for predicting positive classes into positive class numbers, TN is used for predicting negative classes into negative class numbers, FP is used for predicting negative classes into positive class numbers, and FN is used for predicting positive classes into negative class numbers. The calculation formula of the accuracy is as follows: accuracy = (TP + TN)/(TP + TN + FP + FN).
Through tests, the prediction accuracy of the network model for the test set of the fire point and the non-fire point obtained through random sampling is 99%.
The method is based on deep learning, has higher running speed and can acquire the relevant information of the fire in time. Compared with the traditional context relative threshold algorithm, the method can automatically judge the fire point and effectively improve the accuracy of the fire point detection of the remote sensing data.
Claims (7)
1. A fire point detection method for Himapari-8 remote sensing data is characterized by comprising the following steps:
acquiring original remote sensing data of a Himapari-8 satellite and corresponding fire point data, wherein the original remote sensing data comprises longitude and latitude and data of each wave band, and the fire point data comprises longitude and latitude information of a fire point;
randomly extracting N data blocks containing fire point longitude and latitude from the original remote sensing data as positive samples, and randomly extracting N data blocks containing no fire point longitude and latitude from the original remote sensing data as negative samples;
step three, disordering the extracted positive and negative samples, and randomly dividing the positive and negative samples into a training set and a testing set;
establishing a double-input network model, wherein the network model comprises a fire detection network branch, a cloud water detection network branch and a fusion module; the fire detection network branch is used for processing a temperature radiation wave band related to fire detection, the cloud water detection network branch is used for processing a reflectivity wave band related to cloud water detection, the fire detection network branch and the cloud water detection network branch both comprise a wave band attention module and a pixel attention module, output data of the two branches are transmitted to a fusion module, and the fusion module is used for outputting a fire detection result;
step five, training the dual-input network model established in the step four by using the training set and the test set obtained in the step three to obtain a trained model;
and step six, dividing the original remote sensing data of the Himapwari-8 satellite to be detected into blocks with the same size as the positive and negative sample data blocks, inputting the blocks into a trained model, and detecting whether fire points exist in the blocks.
2. The method of claim 1, wherein the size of the data blocks of the positive and negative samples is 21 x 21 pixels, and the extraction bands of the positive and negative samples each include albedo _03, albedo _04, albedo _06, tbb _07, tbb _14, tbb _15, SOZ, latitude, and longtude.
3. The fire detection method of Hiwari-8 remote sensing data according to claim 1, wherein the fire detection network branch comprises a pixel attention module, a wave band attention module, a rolling block, a convolution block which are connected in series in sequence; the cloud water detection network branch comprises a pixel attention module, a wave band attention module, a rolling block and a rolling block which are sequentially connected in series.
4. The fire detection method of Hiwari-8 remote sensing data according to claim 3, wherein in the waveband attention module, the data input into the waveband attention module are respectively input into a global average pooling layer and a global maximum pooling layer, then the pooling results of the global average pooling layer and the global maximum pooling layer are subjected to element-level addition, then a 1 x 1 convolution and ReLU activation function are connected, and a 1 x 1 convolution block and a Sigmoid activation function are connected to obtain the attention feature; and performing element-level multiplication on the attention feature and the data of the input waveband attention module to obtain the waveband attention.
5. The fire detection method of Himapari-8 remote sensing data according to claim 3, wherein in the pixel attention module, the data input into the pixel attention module is input into two cascaded 1 x 1 convolutions to obtain a feature map with the channel number of 1, wherein the first 1 x 1 convolution has an activation function, and the first 1 x 1 convolution has a Sigmoid activation function; and then, expanding the feature map with the channel number of 1 into the size of the data of the input pixel attention module through copy operation, and performing element-level multiplication on the expanded feature map and the data of the input pixel attention module to obtain the pixel attention.
6. The fire detection method of Hiwari-8 remote sensing data according to claim 3, wherein in the fusion module, output results of the fire detection network branch and the cloud water detection network branch are respectively input into respective corresponding full connection layers, then element-level multiplication is performed on outputs of the two full connection layers to obtain a fusion result, and then the fusion result is input into one full connection layer and a subsequent SoftMax function to obtain the fire detection result.
7. The fire point detection method for the Himapari-8 remote sensing data according to claim 1, wherein during training in the fifth step, cross entropy loss is selected as a loss function, an Adam optimizer is selected for optimization, an initial learning rate is set to be 0.0001, and a MultiStepLR adjustment strategy is adopted to adjust the learning rate; inputting the training set samples into a network model for training, and obtaining an optimal model for fire point detection according to the change condition of the loss value; then inputting the test set sample into the trained model to obtain the discrimination confidence of each fire point data, and if the confidence is greater than 0.5, judging that the sample belongs to the fire point; and (5) counting the accuracy of fire point detection and verifying the accuracy of the model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110476630.6A CN113095437B (en) | 2021-04-29 | 2021-04-29 | Fire point detection method for Himapari-8 remote sensing data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110476630.6A CN113095437B (en) | 2021-04-29 | 2021-04-29 | Fire point detection method for Himapari-8 remote sensing data |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113095437A true CN113095437A (en) | 2021-07-09 |
| CN113095437B CN113095437B (en) | 2022-03-01 |
Family
ID=76680764
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110476630.6A Active CN113095437B (en) | 2021-04-29 | 2021-04-29 | Fire point detection method for Himapari-8 remote sensing data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113095437B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113724239A (en) * | 2021-09-09 | 2021-11-30 | 中国石油大学(华东) | Time-space fire point detection method based on Himapari-8 remote sensing image |
| CN114398595A (en) * | 2021-12-29 | 2022-04-26 | 青岛星科瑞升信息科技有限公司 | Fire point remote sensing identification method based on BP neural network |
| CN115512236A (en) * | 2022-10-13 | 2022-12-23 | 昆明理工大学 | Himarwari-8 multispectral cloud detection method and system based on K-means + |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180285716A1 (en) * | 2017-03-31 | 2018-10-04 | Ecole Polytechnique Federale De Lausanne (Epfl) | Simplification of neural network models |
| CN109118001A (en) * | 2018-08-09 | 2019-01-01 | 成都天地量子科技有限公司 | A kind of mountain fire monitoring method and system based on satellite remote sensing date |
| CN110516596A (en) * | 2019-08-27 | 2019-11-29 | 西安电子科技大学 | Empty spectrum attention hyperspectral image classification method based on Octave convolution |
| CN111090715A (en) * | 2019-12-26 | 2020-05-01 | 象辑知源（武汉）科技有限公司 | Satellite remote sensing fire monitoring system |
| CN112232156A (en) * | 2020-09-30 | 2021-01-15 | 河海大学 | Remote sensing scene classification method based on multi-head attention generation countermeasure network |
-
2021
- 2021-04-29 CN CN202110476630.6A patent/CN113095437B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180285716A1 (en) * | 2017-03-31 | 2018-10-04 | Ecole Polytechnique Federale De Lausanne (Epfl) | Simplification of neural network models |
| CN109118001A (en) * | 2018-08-09 | 2019-01-01 | 成都天地量子科技有限公司 | A kind of mountain fire monitoring method and system based on satellite remote sensing date |
| CN110516596A (en) * | 2019-08-27 | 2019-11-29 | 西安电子科技大学 | Empty spectrum attention hyperspectral image classification method based on Octave convolution |
| CN111090715A (en) * | 2019-12-26 | 2020-05-01 | 象辑知源（武汉）科技有限公司 | Satellite remote sensing fire monitoring system |
| CN112232156A (en) * | 2020-09-30 | 2021-01-15 | 河海大学 | Remote sensing scene classification method based on multi-head attention generation countermeasure network |
Non-Patent Citations (2)
| Title |
|---|
| BIN WANG ET AL.: "Estimate hourly PM2.5 concentrations from Himawari-8 TOA reflectance directly using geo-intelligent long short-term memory network", 《ENVIRONMENTAL POLLUTION》 * |
| 徐 迅等: "基于VIIRS 影像的秸秆焚烧火点监测方法", 《地理空间信息》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113724239A (en) * | 2021-09-09 | 2021-11-30 | 中国石油大学(华东) | Time-space fire point detection method based on Himapari-8 remote sensing image |
| CN113724239B (en) * | 2021-09-09 | 2024-03-29 | 中国石油大学(华东) | Space-time fire detection method based on Himaware-8 remote sensing image |
| CN114398595A (en) * | 2021-12-29 | 2022-04-26 | 青岛星科瑞升信息科技有限公司 | Fire point remote sensing identification method based on BP neural network |
| CN114398595B (en) * | 2021-12-29 | 2022-10-28 | 青岛星科瑞升信息科技有限公司 | Fire point remote sensing identification method based on BP neural network |
| CN115512236A (en) * | 2022-10-13 | 2022-12-23 | 昆明理工大学 | Himarwari-8 multispectral cloud detection method and system based on K-means + |
| CN115512236B (en) * | 2022-10-13 | 2023-04-28 | 昆明理工大学 | Himaware-8 multispectral cloud detection method and system based on K-means++ |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113095437B (en) | 2022-03-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113095437B (en) | Fire point detection method for Himapari-8 remote sensing data | |
| US20220076055A1 (en) | Systems, methods, and computer readable media for predictive analytics and change detection from remotely sensed imagery | |
| CN110335270B (en) | Power transmission line defect detection method based on hierarchical regional feature fusion learning | |
| CN110533631A (en) | SAR image change detection based on the twin network of pyramid pondization | |
| CN113553764B (en) | Mountain fire prediction method based on deep learning network | |
| Stojanova et al. | Estimating the risk of fire outbreaks in the natural environment | |
| CN113496104A (en) | Rainfall forecast correction method and system based on deep learning | |
| CN110134907B (en) | Rainfall missing data filling method and system and electronic equipment | |
| CN116543346A (en) | Deep learning-based transmission line video mountain fire detection method | |
| Tayal et al. | To Predict the Fire Outbreak in Australia using Historical Database | |
| Fan et al. | A comparative study of four merging approaches for regional precipitation estimation | |
| CN117556197B (en) | Typhoon vortex initialization method based on artificial intelligence | |
| Wang et al. | Nowcasting multicell short-term intense precipitation using graph models and random forests | |
| CN116229380A (en) | Method for identifying bird species related to bird-related faults of transformer substation | |
| Clements et al. | Applying multiple testing procedures to detect change in East African vegetation | |
| KR102496876B1 (en) | Apparatus and method for forecasting the seasonal forest fire risk | |
| Ramdani et al. | An artificial neural network approach to predict the future land use land cover of great Malang Region, Indonesia | |
| CN116976227A (en) | Storm water increasing forecasting method and system based on LSTM machine learning | |
| CN112509285A (en) | CNN-based global typhoon message collection method and system | |
| Di et al. | Research on secondary analysis method of synchronous satellite monitoring data of power grid wildfire | |
| CN108510097B (en) | Power transmission line corridor snow disaster monitoring method and system based on satellite remote sensing | |
| CN114998719A (en) | Forest fire prediction method based on deep learning and multi-source remote sensing data | |
| CN114758237A (en) | Construction method, detection method and construction device of automatic water delivery tunnel defect identification model, computer and storage medium | |
| Nakagawa et al. | Application of random forest to classify weather observation into rainfall using GNSS receiver | |
| CN111950343A (en) | Automatic transmission tower identification method and system based on satellite remote sensing technology |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |