CN113780439A - Multi-cloud identification system of different types of meteorological satellites based on unsupervised domain adaptation - Google Patents
Multi-cloud identification system of different types of meteorological satellites based on unsupervised domain adaptation Download PDFInfo
- Publication number
- CN113780439A CN113780439A CN202111081849.2A CN202111081849A CN113780439A CN 113780439 A CN113780439 A CN 113780439A CN 202111081849 A CN202111081849 A CN 202111081849A CN 113780439 A CN113780439 A CN 113780439A
- Authority
- CN
- China
- Prior art keywords
- satellite data
- domain
- satellite
- module
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/29—Geographical information databases
-
- 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
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Databases & Information Systems (AREA)
- Health & Medical Sciences (AREA)
- Remote Sensing (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a multi-cloud identification system of different types of meteorological satellites based on unsupervised domain adaptation, which adopts an unsupervised domain adaptation method to solve the difficulties of cross-satellite and detection channel difference and the like, extracts domain invariant features based on learning data distribution of two fields, reduces the domain difference in a high-level semantic feature space, further realizes multi-cloud identification on an unmarked target domain satellite and realizes the average class precision of a plurality of different cloud classes. The invention applies unsupervised domain adaptation to the satellite remote sensing field, carries out cloud classification on certain satellites under the condition that the satellites are not marked, and realizes the combination of the remote sensing meteorological field and deep learning. By adopting the system, the data marking work of weather professionals on the new satellite can be reduced, the dependence of weather on a numerical value calculation monitoring mode is eliminated, the new satellite can be quickly applied to the ground and product feedback is realized after the new satellite is used, and a foundation is laid for more different series of satellite applications.
Description
Technical Field
The invention relates to the technical field of meteorological monitoring, in particular to a multi-cloud identification system of different types of meteorological satellites based on unsupervised domain adaptation.
Background
Satellite technology has gained rapid development on the basis of the aerospace industry in recent years, and satellite-based meteorological monitoring is more closely related to the life of people. However, knowledge among different satellites is difficult to migrate and apply, so that the method has extremely important influence on research on domain adaptive migration among different meteorological satellites, particularly has a promoting effect on the development of aerospace industry and meteorological satellites in China, can greatly save time and resources of professional researchers of the meteorological satellites, and has great significance on deep and detailed research on the meteorological satellites.
Data monitored by the meteorological satellite in the space are transmitted back to the earth surface, and classification of different cloud types is completed through modes of manual labeling, numerical mode calculation and the like, wherein the data generally comprise a plurality of channels, such as a visible light channel, a near infrared channel, a thermal infrared channel and the like. Professional meteorological personnel can divide out ten kinds of cloud through the information that data were given through artifical mark or numerical mode operation, wherein include: clear rolling clouds, rolling cloud layer clouds, deep convection clouds, high-altitude accumulated clouds, high-altitude layered clouds, rain cloud layer clouds, accumulated clouds, layered clouds and the like, and inversion, forecast and early warning are carried out on the weather through the cloud types.
However, different types of meteorological satellites have different observation bands, the number of detection channels, and the like. For example, the wind cloud No. 4A satellite is a meteorological satellite independently developed in the last 70 th century of China, and has 14 channels in total, including visible light channels, short-wave infrared channels, medium-wave infrared channels, long-wave infrared channels and the like. However, the sunflower 8 satellite developed in japan is very different, and taking the detection band and frequency as an example, the sunflower 8 satellite has 16 channels in total, and has two more channels (a B channel and an infrared channel in a visible light channel) compared with the wind cloud 4 satellite.
Therefore, the different types of satellites have huge domain differences, and as the current deep learning data is mostly based on data driving type, and only one model can be better adapted to one type of data, the model migration between two types of satellites cannot be directly carried out, and the existing trained model cannot show better performance.
In the prior art, for example, CN108846334A discloses a cloud category automatic identification method and system, which includes a data acquisition module, a cloud image identification module, a cloud image result display module, and a model construction module, where the cloud image result display module is respectively in signal connection with the data acquisition module, the cloud image identification module, and the model construction module, and provides an improved Dense Net on the basis of a Dense connection convolutional network (Dense Net), and combines technologies such as mobile phone APP development and camera monitoring video processing, so as to solve the problems that due to the various kinds of clouds, partially extracted features have strong pertinence, effective features are difficult to extract from massive cloud image data, and internal relations among different cloud images cannot be fully exploited. However, this solution has the following drawbacks: (1) the tag cannot acquire: for target domain satellite data, usable labels cannot be acquired, so that the target domain data cannot be learned and trained by using a supervised deep learning method, and the application requirements cannot be met by using an unsupervised deep learning method; (2) the labor cost is high: for satellite data which is not perfect in cloud products, artificial labeling is mainly used, particularly, labeling of the satellite data belongs to pixel-level labeling, the artificial cost is high, and the labeling period is long; (3) existing models have low generalization on new data: the existing technology does not consider experience mutual reference among a plurality of satellites, and the existing trained deep learning model has the problems of poor generalization and the like on new data.
Disclosure of Invention
Aiming at a multi-cloud type identification task, the invention aims to solve the problem of how to transfer and apply experience knowledge acquired from a mature satellite (a source domain satellite) to a new unlabelled satellite (a target domain satellite), and further provides a multi-cloud identification system based on different types of meteorological satellites suitable for an unsupervised domain.
In order to achieve the above purpose, the invention provides the following technical scheme:
the invention firstly provides a multi-cloud identification system of different types of meteorological satellites based on unsupervised domain adaptation, and the multi-cloud identification system of different types of meteorological satellites based on unsupervised domain adaptation is characterized by comprising the following modules:
the input module is used for inputting the target domain satellite data and the source domain satellite data into the satellite data preprocessing module;
the satellite data preprocessing module is used for preprocessing satellite data, carrying out pixel level normalization processing, carrying out dimension processing and carrying out lowest spatial resolution processing of different channels of the same satellite according to the input satellite data;
the source domain satellite data feature extraction and satellite data segmentation module is used for performing feature extraction on satellite data of a source domain and performing semantic segmentation on the source domain data and target domain data;
the target domain satellite data feature extraction module is used for extracting features of satellite data of a target domain;
the target domain satellite data domain discriminator module is used for discriminating and feeding back high-dimensional semantic features of satellite data generated by the source domain and the target domain, and has an antagonistic relation with the target domain satellite data feature extraction module;
and the output module is used for outputting the marked file form of the target domain satellite.
Further, the input module comprises a source domain data input module and a target domain data input module, wherein the target domain data input module and the source domain data input module respectively use target domain satellite data and source domain satellite data as input and jointly input the target domain satellite data and the source domain satellite data into the satellite data preprocessing module for operation.
Furthermore, in the satellite data preprocessing module, the original format of the satellite data is a three-dimensional array, and the satellite data preprocessing comprises the unification of the resolution, the longitude and latitude directions, the unification of the data format, the normalization processing of the pixel level and the unification processing of the dimension.
Further, in the satellite data preprocessing module, the satellite data is converted into a data format which can be used by a deep learning framework after being preprocessed.
Furthermore, in the satellite data preprocessing module, dimension processing is unified into the size of a region with height multiplied by width multiplied by the number of satellite channels.
Furthermore, in the source domain satellite data feature extraction and satellite data segmentation module, the feature extraction of the satellite data of the source domain mainly comprises a deep neural network for feature extraction composed of a plurality of convolutional layers, the semantic segmentation of the satellite data of the source domain and the target domain mainly comprises a semantic segmentation neural network composed of a plurality of convolutional layers and an upper sampling layer, and finally, a segmentation result is output.
Furthermore, in the target domain satellite data feature extraction module, feature extraction is carried out on the satellite data of the target domain, the deep neural network mainly comprises a plurality of layers of convolution layers, and semantic features of the high-dimensional target domain satellite data are extracted.
The invention also provides a control method of the multi-cloud identification system based on different types of meteorological satellites suitable for the unsupervised domain, which comprises the following steps:
s1, obtaining labels of two satellite data and source domain satellite data in sufficient quantity and transmitting the labels into a satellite data preprocessing module;
s2, transmitting the preprocessed source domain satellite data into a source domain satellite data feature extraction and satellite data segmentation module for pre-training, and obtaining a segmentation training result of the source domain data;
s3, transmitting the preprocessed target domain satellite data into a target domain satellite data feature extraction module and a target domain satellite data domain discriminator module, and combining the segmentation training result in S2 to pre-train, namely only updating the target domain satellite data domain discriminator module;
and S4, continuing training after pre-training, namely, iteratively updating the target domain satellite data domain discriminator module and the target domain satellite data feature extraction module, and outputting a final prediction result.
Compared with the prior art, the invention has the beneficial effects that:
the invention provides a multi-cloud identification system of different types of meteorological satellites based on unsupervised domain adaptation, which adopts an unsupervised domain adaptation method to solve the difficulties of cross-satellite and detection channel difference and the like, extracts domain invariant features based on learning data distribution of two fields, reduces the domain difference in a high-level semantic feature space, further realizes multi-cloud identification on an unmarked target domain satellite, and realizes the average class precision of various different cloud classes.
The invention applies unsupervised domain adaptation to the satellite remote sensing field, carries out cloud classification on certain satellites under the condition that the satellites are not marked, and realizes the combination of the remote sensing meteorological field and deep learning.
By adopting the system, the data marking work of weather professionals on the new satellite can be reduced, the dependence of weather on a numerical value calculation monitoring mode is eliminated, the new satellite can be quickly applied to the ground and product feedback is realized after the new satellite is used, and a foundation is laid for more different series of satellite applications.
Drawings
In order to more clearly illustrate the embodiments of the present application or technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments described in the present invention, and other drawings can be obtained by those skilled in the art according to the drawings.
Fig. 1 is an architecture diagram of a multiple cloud identification system based on different types of meteorological satellites adapted by an unsupervised domain according to an embodiment of the present invention.
Fig. 2 is a flowchart of a control method of a multiple cloud identification system based on different types of meteorological satellites adapted in an unsupervised domain according to an embodiment of the present invention.
Detailed Description
For a better understanding of the present solution, the method of the present invention is described in detail below with reference to the accompanying drawings.
As shown in FIG. 1, the invention provides a multi-cloud identification system based on different types of meteorological satellites adapted by an unsupervised domain, which comprises the following modules:
(1) input module
And inputting the target domain satellite data and the source domain satellite data into a subsequent module for operation.
(2) Satellite data preprocessing module
The satellite data is preprocessed according to the input satellite data, and the preprocessing mainly comprises resolution, longitude and latitude direction unification, data format unification (processing is a data format available for a deep learning frame), pixel level normalization processing, dimension processing (unifying to h x w x c area size), and lowest spatial resolution processing of different channels of the same satellite.
Specifically, satellite data is used as input and enters a satellite data preprocessing module, the original format of the satellite data is a three-dimensional array, and the satellite data takes the sunflower satellite data No. 8 of the Zhejiang strait as an example, the satellite data comprises satellite data of 16 channels, and the satellite data is marked as marked data of 1 channel with the same size. The minimum spatial resolution processing of different channels of the same satellite refers to the area normalization processing of a region represented by one pixel of different channels on the same satellite, and the data format processing is to unify the data formats of different satellites and facilitate inputting the data formats into a deep learning model. The satellite data preprocessing module is mainly used for aligning data of a source domain and data of a target domain on the spatial resolution, and transmitting the preprocessed data serving as output to a subsequent module. Therefore, the data uniformity, integrity and accuracy in the subsequent data training process are ensured.
(3) Source domain satellite data feature extraction and satellite data segmentation module
The method comprises the steps of extracting features of satellite data of a source domain, mainly comprising a deep neural network for feature extraction and composed of a plurality of convolutional layers, semantically segmenting the satellite data of the source domain and a target domain, mainly comprising a semantically segmented deep neural network composed of a plurality of convolutional layers and an upper sampling layer, and finally outputting a segmentation result.
Specifically, in the source domain satellite data feature extraction and satellite data segmentation module, the main functions of the module are feature extraction of source domain satellite data and semantic segmentation of the source domain data and target domain data. The input of the method is preprocessed source domain satellite data and labels. In a specific structure, the first layer convolution module mainly plays a role in eliminating the influence caused by channel difference. For source domain data, the module extracts deep semantic information from the multi-layer residual error network module and restores the image through up-sampling. For target domain data, the target domain data is input into the output of a target domain satellite data feature extraction module at an intermediate stage, and partial feature extraction and up-sampling are carried out.
(4) Target domain satellite data feature extraction module
And extracting the features of the satellite data of the target domain, wherein the feature extraction mainly comprises a deep neural network consisting of a plurality of layers of convolution layers, and the semantic features of the satellite data of the high-dimensional target domain are extracted.
Specifically, in the target domain satellite data feature extraction module, the main function of the module is to extract the features of the target domain satellite data. Its input is the target domain satellite data. The first layer convolution module is mainly used for eliminating the influence caused by channel difference. The method obtains deep semantic information by extracting the characteristics of the satellite data of the target domain, and transmits the deep semantic information back to the source domain satellite data characteristic extraction and satellite data segmentation module. Which is one party in the countermeasure network, and the target domain satellite data discriminator module is confronted with each other.
(5) Target domain satellite data domain discriminator module
And distinguishing and feeding back high-dimensional semantic features of satellite data generated by the source domain and the target domain, wherein the feature extraction module is in an antagonistic relation with the satellite data feature extraction module of the target domain.
Specifically, in the target domain satellite data domain discriminator module, its main role is to discriminate whether the sample is from the source domain (actual distribution) or the target domain (learned distribution). The input of the method is semantic features of different depths from a source domain satellite data feature extraction and satellite data segmentation module. In the iterative learning process, if the discriminator module cannot separate the target domain features from the source domain features in the iterative process, the feature extractor module extracts features of which the source domain is more similar to the target domain. In the countertraining, the domain invariant features are extracted by learning the data distribution of two domains, and the domain difference is reduced in a high-level semantic feature space, so that unsupervised domain adaptation is realized. In addition to using the domain discriminator module, a tag discriminator or the like can be additionally inserted to optimize the model architecture.
(6) Output module
And outputting the marked file form of the target domain satellite.
The invention also provides a control method for a multi-cloud identification system based on different types of meteorological satellites adapted by an unsupervised domain, as shown in fig. 2, the method comprises the following steps:
s1: acquiring labels of two satellite data and source domain satellite data in sufficient quantity and transmitting the labels into a satellite data preprocessing module;
s2: transmitting the preprocessed source domain satellite data into a source domain satellite data feature extraction and satellite data segmentation module for pre-training, and obtaining a segmentation training result of the source domain data;
s3: transmitting the preprocessed target domain satellite data into a target domain satellite data feature extraction module and a target domain satellite data domain discriminator module, and combining the segmentation training result in S2 to perform pre-training, namely updating only the target domain satellite data domain discriminator module;
s4: and (4) continuing training after the pre-training is finished, namely, iteratively updating the target domain satellite data domain discriminator module and the target domain satellite data feature extraction module, and outputting a final prediction result.
Examples
The data of the sunflower 8 satellite in Japan is used as a source domain, and the Fengyun four-satellite A satellite in China is used as a target domain, so that unsupervised domain adaptive transfer learning is realized, and label-free labeling of the Fengyun four-satellite A satellite is realized.
In a data preprocessing module, two types of three-dimensional satellite data are input, the processing comprises the unification of resolution, the longitude and latitude directions are unified near the Zhejiang strait, the data format is unified into tfrecrd format of tensoflow, and the pixel level normalization processing and the dimension unification processing are carried out. The final preprocessed sunflower 8 satellite data is a 512 x 16 three-dimensional array, the sunflower 8 satellite label is a 512 x 1 three-dimensional array, the aeolian four satellite A satellite data is a 512 x 14 three-dimensional array, and channel differences exist between the two satellites.
Then, the preprocessed source domain data is input into a source domain satellite data feature extraction and satellite data segmentation module for pre-training, the pre-training is similar to a semantic segmentation algorithm, the distribution of the source domain data is mainly learned, and a segmentation training result of the sunflower No. 8 satellite data is obtained.
After pre-training is finished, the sunflower satellite data No. 8 is simultaneously input into a satellite data feature extraction and satellite data segmentation module, and the Fengyun satellite data A is input into a target domain satellite data feature extraction module. Through the target domain satellite data domain discriminator module, iterative updating based on the countermeasure training of the generated countermeasure network GAN is carried out, the domain gap is reduced, and cloud type distribution on the target domain data is learned.
And finally, outputting a cloud classification semantic segmentation result near the Zhejiang strait of the wind-cloud four-satellite type A by an output module, wherein different colors represent different types of cloud distribution.
The invention mainly relates to a non-supervision domain adaptive transfer learning mode, and belongs to the technical field of deep learning. Deep learning is to finally realize the learning and the discrimination of data by simulating a neural network of a human brain and building different convolution layers, pooling layers and the like to learn the internal rules and the representation levels of sample data.
The unsupervised domain adaptation is one of transfer learning, and is characterized in that under the condition that a target domain has no label, the feature with unchanged domain is extracted by means of generation of counterlearning and the like, and the domain difference between a source domain and the target domain is reduced in a high-level semantic feature space, so that unsupervised learning on the target domain is realized.
Compared with the prior art, the multi-cloud identification system based on different types of meteorological satellites suitable for the unsupervised domain and the control method thereof have the following advantages that:
(1) the labor cost is low: the invention can intelligently label satellite data without labels in an unsupervised learning and deep learning mode, thereby reducing the workload of meteorological workers and labor cost.
(2) The application period is short: compared with the manual labeling, the method has the advantages that the consumption time is long, after model training is completed, the cloud distribution on the target domain can be learned quickly, and the application period is short.
(3) The universality is strong: the invention can be conveniently transplanted to the transfer learning between any two satellites, has strong universality and simple and convenient application, and assists in manual identification.
The above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: it is to be understood that modifications may be made to the technical solutions described in the foregoing embodiments, or equivalents may be substituted for some of the technical features thereof, but such modifications or substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims (8)
1. A multi-cloud identification system based on different types of meteorological satellites adapted by an unsupervised domain is characterized by comprising the following modules:
the input module is used for inputting the target domain satellite data and the source domain satellite data into the satellite data preprocessing module;
the satellite data preprocessing module is used for preprocessing satellite data, carrying out pixel level normalization processing, carrying out dimension processing and carrying out lowest spatial resolution processing of different channels of the same satellite according to the input satellite data;
the source domain satellite data feature extraction and satellite data segmentation module is used for performing feature extraction on satellite data of a source domain and performing semantic segmentation on the source domain data and target domain data;
the target domain satellite data feature extraction module is used for extracting features of satellite data of a target domain;
the target domain satellite data domain discriminator module is used for discriminating and feeding back high-dimensional semantic features of satellite data generated by the source domain and the target domain, and has an antagonistic relation with the target domain satellite data feature extraction module;
and the output module is used for outputting the marked file form of the target domain satellite.
2. The unsupervised domain adaptation-based multi-cloud identification system for different types of meteorological satellites according to claim 1, wherein the input module comprises a source domain data input module and a target domain data input module, and the target domain data input module and the source domain data input module respectively use target domain satellite data and source domain satellite data as input and jointly input the target domain satellite data and the source domain satellite data into the satellite data preprocessing module for operation.
3. The unsupervised domain adaptation-based multi-cloud identification system for different types of meteorological satellites according to claim 1, wherein in the satellite data preprocessing module, the original format of the satellite data is a three-dimensional array, and the satellite data preprocessing comprises the unification of resolution, longitude and latitude directions, the unification of data formats and pixel-level normalization processing.
4. The unsupervised domain adaptation-based multi-cloud identification system for different types of meteorological satellites according to claim 1, wherein in the satellite data preprocessing module, the preprocessed satellite data is converted into a deep learning framework available data format.
5. The unsupervised domain adaptation-based multi-cloud identification system for different types of meteorological satellites according to claim 1, wherein dimension processing is unified into a region size of height x width x number of satellite channels in the satellite data preprocessing module.
6. The unsupervised domain adaptation-based multi-cloud identification system for different types of meteorological satellites according to claim 1, wherein in the source domain satellite data feature extraction and satellite data segmentation module, feature extraction is performed on satellite data of a source domain, the feature extraction mainly comprises a deep neural network for feature extraction, which is composed of a plurality of convolutional layers, semantic segmentation is performed on the satellite data of the source domain and a target domain, the semantic segmentation mainly comprises a semantic segmentation neural network, which is composed of a plurality of convolutional layers and an upper sampling layer, and a segmentation result is finally output.
7. The unsupervised domain adaptation-based multi-cloud identification system for different types of meteorological satellites according to claim 1, wherein in the target domain satellite data feature extraction module, feature extraction is performed on satellite data of a target domain, and the feature extraction module mainly comprises a deep neural network formed by a plurality of convolutional layers to extract semantic features of high-dimensional satellite data of the target domain.
8. The method for controlling the unsupervised domain adapted meteorological satellite based multi-cloud identification system for different types of meteorological satellites according to any one of claims 1-7, comprising the following steps:
s1, obtaining labels of two satellite data and source domain satellite data in sufficient quantity and transmitting the labels into a satellite data preprocessing module;
s2, transmitting the preprocessed source domain satellite data into a source domain satellite data feature extraction and satellite data segmentation module for pre-training, and obtaining a segmentation training result of the source domain data;
s3, transmitting the preprocessed target domain satellite data into a target domain satellite data feature extraction module and a target domain satellite data domain discriminator module, and combining the segmentation training result in S2 to pre-train, namely only updating the target domain satellite data domain discriminator module;
and S4, continuing training after pre-training, namely, iteratively updating the target domain satellite data domain discriminator module and the target domain satellite data feature extraction module, and outputting a final prediction result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111081849.2A CN113780439B (en) | 2021-09-15 | 2021-09-15 | Multi-cloud identification system of different types of meteorological satellites based on unsupervised domain adaptation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111081849.2A CN113780439B (en) | 2021-09-15 | 2021-09-15 | Multi-cloud identification system of different types of meteorological satellites based on unsupervised domain adaptation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113780439A true CN113780439A (en) | 2021-12-10 |
| CN113780439B CN113780439B (en) | 2023-09-22 |
Family
ID=78844034
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111081849.2A Active CN113780439B (en) | 2021-09-15 | 2021-09-15 | Multi-cloud identification system of different types of meteorological satellites based on unsupervised domain adaptation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113780439B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108062753A (en) * | 2017-12-29 | 2018-05-22 | 重庆理工大学 | The adaptive brain tumor semantic segmentation method in unsupervised domain based on depth confrontation study |
| KR101855652B1 (en) * | 2018-02-13 | 2018-06-25 | 대한민국 | System of total cloud cover retrieval from satellite observation using machine learning and method thereof |
| CN113052201A (en) * | 2020-12-22 | 2021-06-29 | 中国人民解放军61540部队 | Satellite cloud picture cloud classification method based on deep learning |
-
2021
- 2021-09-15 CN CN202111081849.2A patent/CN113780439B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108062753A (en) * | 2017-12-29 | 2018-05-22 | 重庆理工大学 | The adaptive brain tumor semantic segmentation method in unsupervised domain based on depth confrontation study |
| KR101855652B1 (en) * | 2018-02-13 | 2018-06-25 | 대한민국 | System of total cloud cover retrieval from satellite observation using machine learning and method thereof |
| CN113052201A (en) * | 2020-12-22 | 2021-06-29 | 中国人民解放军61540部队 | Satellite cloud picture cloud classification method based on deep learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113780439B (en) | 2023-09-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Du et al. | Pavement distress detection and classification based on YOLO network | |
| CN110879989B (en) | Ads-b signal target identification method based on small sample local machine learning model | |
| CN109033998B (en) | Remote sensing image ground object labeling method based on attention mechanism convolutional neural network | |
| CN104992223B (en) | Intensive population estimation method based on deep learning | |
| CN110245709B (en) | 3D point cloud data semantic segmentation method based on deep learning and self-attention | |
| CN107133569B (en) | Monitoring video multi-granularity labeling method based on generalized multi-label learning | |
| CN108399399B (en) | Urban range extraction method based on noctilucent remote sensing image | |
| CN108765404A (en) | A kind of road damage testing method and device based on deep learning image classification | |
| CN109063569A (en) | A kind of semantic class change detecting method based on remote sensing image | |
| CN115131627B (en) | Construction and training method of lightweight plant disease and pest target detection model | |
| CN114092832A (en) | High-resolution remote sensing image classification method based on parallel hybrid convolutional network | |
| CN112749663B (en) | Agricultural fruit maturity detection system based on Internet of things and CCNN model | |
| CN116051996A (en) | Two-stage crop growth prediction method based on multi-mode information | |
| CN112396635A (en) | Multi-target detection method based on multiple devices in complex environment | |
| CN113160150A (en) | AI (Artificial intelligence) detection method and device for invasion of foreign matters in wire network based on multi-type sample fusion and multi-complex network | |
| Huang et al. | Depth semantic segmentation of tobacco planting areas from unmanned aerial vehicle remote sensing images in plateau mountains | |
| CN111461006A (en) | Optical remote sensing image tower position detection method based on deep migration learning | |
| CN115131747A (en) | Knowledge distillation-based power transmission channel engineering vehicle target detection method and system | |
| CN113158891A (en) | Cross-camera pedestrian re-identification method based on global feature matching | |
| CN115585731A (en) | Space-air-ground integrated hydropower station space state intelligent monitoring management system and method thereof | |
| CN111242028A (en) | Remote sensing image ground object segmentation method based on U-Net | |
| Yuan et al. | Sensitivity examination of YOLOv4 regarding test image distortion and training dataset attribute for apple flower bud classification | |
| CN109426813B (en) | Remote sensing image user-defined interest point extraction method based on fuzzy clustering and neural network model | |
| CN111242134A (en) | Remote sensing image ground object segmentation method based on feature adaptive learning | |
| Andersson et al. | Combining street-level and aerial images for dengue incidence rate estimation |
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 |