CN113393466A - Semantic segmentation network model for MODIS sea fog detection - Google Patents
Semantic segmentation network model for MODIS sea fog detection Download PDFInfo
- Publication number
- CN113393466A CN113393466A CN202110676781.6A CN202110676781A CN113393466A CN 113393466 A CN113393466 A CN 113393466A CN 202110676781 A CN202110676781 A CN 202110676781A CN 113393466 A CN113393466 A CN 113393466A
- Authority
- CN
- China
- Prior art keywords
- semantic segmentation
- feature
- segmentation network
- sea fog
- seafog
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- 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
- 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
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (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 Analysis (AREA)
Abstract
The invention discloses a semantic segmentation network model for MODIS sea fog detection, which comprises the following steps: 1) constructing a sea fog sample set SeaFog _ Dataset; 2) constructing a semantic segmentation network for MODIS sea fog detection, adding feature maps obtained by completing encoding and decoding operations in each layer of network, and obtaining a final classification result through a softmax function layer; 3) and training and verifying the constructed semantic segmentation network to obtain a semantic segmentation network model finally used for sea fog detection. The method comprises the steps of downloading an MODIS satellite remote sensing image, labeling different marine ground objects through ArcGIS, and taking a labeling result as a training label of a semantic segmentation model; a new semantic segmentation network model is constructed for sea fog detection, a semantic segmentation model is trained on a GPU by using a back propagation algorithm, a deep semantic segmentation network is used for MODIS sea fog detection, and ground object type discrimination is carried out pixel by pixel, so that the sea fog detection precision is improved, and the method has the advantages of being scientific, reasonable, high in automation degree and the like.
Description
Technical Field
The invention relates to the technical field of marine environment monitoring, in particular to a semantic segmentation network model for MODIS sea fog detection.
Background
The low visibility of the sea fog seriously influences the production and life of people along the shore and the safety of marine navigation, in recent years, with the development of marine transportation industry and marine economy, the sea fog disaster has great influence on the development of society and economy, and the effective identification and monitoring of the sea fog are very important. Compared with coastal observation stations and buoy station observation, the satellite remote sensing technology provides a data base for long-time and large-range sea fog observation, and becomes a main means for sea fog monitoring.
The multilayer network in the deep learning method can mine data characteristics as much as possible, identification results are directly obtained, the method is efficient and convenient, and some researches show that the deep learning method can be used for marine environment monitoring and the effect is superior to that of the traditional method. The semantic segmentation network can be used for sea fog detection, the detection precision is greatly improved compared with a threshold value method and a machine learning method, and sea fog can be extracted and identified more accurately. However, in recent years, most of remote sensing image semantic segmentation networks are based on a full convolution neural network, the extracted fog area is rough, phenomena such as missing judgment, erroneous judgment and the like exist at cloud area/sea fog boundaries, and the semantic segmentation networks for sea fog detection still have a space for improvement.
Disclosure of Invention
Based on the background technology, the invention provides a semantic segmentation network model for MODIS sea fog detection, which makes full use of remote sensing image channels and spatial features and realizes accurate sea fog detection by establishing the semantic segmentation model.
The invention adopts the following technical scheme:
a semantic segmentation network model for MODIS sea fog detection is obtained and comprises the following steps:
(1) constructing an ocean fog sample set SeaFog _ Dataset, wherein the ocean fog sample set comprises an ocean fog data set SeaFog _ Image and an ocean fog Label set SeaFog _ Label;
(2) constructing a semantic segmentation network for MODIS sea fog detection, adding feature maps obtained by completing encoding and decoding operations in each layer of network, and obtaining a final classification result through a softmax function layer;
further, the semantic segmentation network in the step (2) is 4 layers.
(2.1) performing an encoding operation:
completing coding operation by using a coding module, and performing convolution operation and downsampling operation on an input SeaFog _ Dataset data set by using the coding module to obtain a Feature map E _ Feature _ 2;
further, the encoding operation in step (2.1) includes the steps of:
firstly, carrying out 1 convolution operation on an input SeaFog _ Dataset data set to obtain a Feature map E _ Feature _ 1;
and then 2 times of coding is carried out on the E _ Feature _1 through 2 residual modules to obtain a Feature map E _ Feature _2, each coding comprises 2 convolutional layers and 1 downsampling layer, a nonlinear activation function layer is accompanied behind each convolutional layer, and the calculation formula of the activation function is as follows:
wherein, x is the input characteristic, α is the adjustable factor, and the value α is 1.0.
Further, the convolution kernel size of the 1 convolution operation in step (2.1) is 7 × 7.
Further, the convolution kernel size in the 2 convolution layers in step (2.1) is 3 × 3.
(2.2) performing a decoding operation:
carrying out convolution operation and up-sampling operation on the obtained Feature image E _ Feature _2 by using a decoding module, and gradually recovering the size of an original image;
firstly, obtaining a Feature map D _ Feature _ front through 1 convolution operation with the convolution kernel size of 1 multiplied by 1, 1 deconvolution operation and 1 convolution operation with the convolution kernel size of 1 multiplied by 1;
then the Feature map D _ Feature _ front obtains a Feature map D _ Feature _ cSE through 1 global pooling layer, 2 convolution operations with convolution kernel size of 1 × 1 × 1 and 1 nonlinear activation function layer;
meanwhile, D _ Feature _ front obtains a Feature map D _ Feature _ sSE through 1 convolution operation with the convolution kernel size of 1 multiplied by 1 and 1 nonlinear activation function layer;
the decoding module performs an adding operation on the Feature map D _ Feature _ cSE and the Feature map D _ Feature _ sSE to obtain a Feature map D _ Feature _ scSE;
finally, adding the coding Feature diagram and the decoding Feature diagram of each layer to obtain a Feature diagram Feature _ 1;
further, the convolution kernel size of 1 deconvolution operation in step (2.2) is 3 × 3.
(2.3) carrying out post-processing operation on the finally obtained Feature map Feature _1 by the semantic segmentation network, and obtaining a final classification result through a softmax function layer;
further, in the step (2.3), specifically:
firstly, obtaining a Feature map Feature _2 through 1 deconvolution operation and 1 nonlinear activation function layer;
secondly, obtaining a Feature map Feature _3 through 1 convolution operation and 1 deconvolution operation;
and finally, obtaining a final classification result through a softmax function layer.
Further, the convolution kernel size of 1 deconvolution operation in step (2.3) is 3 × 3.
Of the 1 convolution operation and the 1 deconvolution operation, the convolution kernel size of the 1 convolution operation is 3 × 3; the convolution kernel size for 1 deconvolution operation is 2 × 2.
(3) And training and verifying the constructed semantic segmentation network to obtain a semantic segmentation network model finally used for sea fog detection.
Further, in step (3), the training and verification of the semantic segmentation network comprises the following steps:
randomly dividing the sea fog sample set SeaFog _ Dataset obtained in the step (1) into a training set SeaFog _ Train and a testing set SeaFog _ Test according to a proportion; training a semantic segmentation network through a SeaFog _ Train, and verifying the detection precision of the semantic segmentation network through the SeaFog _ Test; inputting a sea fog training set SeaFog _ Train and outputting a sea ground feature classification result predicted by a semantic segmentation network;
inputting a training set in batches, setting a loss function and an initial learning rate, taking ResNet18 model parameters as pre-training parameters, taking cross entropy loss of sea ground feature classification results predicted by a semantic segmentation network and sea fog Label sets SeaFog _ Label (ground feature real labels) as loss, and calculating the formula as follows:
FL(pt)＝-αt(1-pt)γlog(pt)
wherein alpha istFor the weighting factor, γ is used to adjust the sample weights, and α is sett＝0.25，γ＝2，ptReflecting the probability that the predicted value is the true value, ptThe calculation formula is as follows:
wherein p represents the probability that the prediction sample class is 1, and y is a real label;
and performing error back propagation by using an Adam optimization algorithm according to the loss error loss, updating the weight parameters of the network layer, verifying by using a cross verification method after each batch of training sets are trained, wherein the cross verification rate is 0.1, and training until the loss value of the loss function is lower than 0.1 to obtain the semantic segmentation network model finally used for sea fog detection.
Further, the training set SeaFog _ Train is larger than the Test set SeaFog _ Test.
Further, the ratio of the training set SeaFog _ Train to the Test set SeaFog _ Test was 8: 2.
Further, the classification of the semantic segmentation network includes a clear sky sea surface, sea fog and cloud region.
Further, the batch size of the training set is 8.
The invention has the following beneficial effects:
the method comprises the steps of downloading an MODIS satellite remote sensing image, labeling different marine ground objects through ArcGIS, and taking a labeling result as a training label of a semantic segmentation model; a new semantic segmentation network model is constructed for sea fog detection, a semantic segmentation model is trained on a GPU by using a back propagation algorithm, a deep semantic segmentation network is used in MODIS sea fog detection, and ground object type discrimination is carried out pixel by pixel, so that the sea fog detection precision is improved, and the method has the advantages of being scientific, reasonable, high in automation degree and the like;
compared with the prior art, the method is based on the multi-channel satellite remote sensing image, and the deep semantic segmentation network is applied to sea fog detection. The semantic segmentation model has strong universality and is suitable for MODIS remote sensing images in any range and different time periods; the sea fog detection accuracy is high, ground object sample channels and spatial features are extracted through different network layers and an attention mechanism, pixel global information is integrated, the sea fog boundary range is extracted more accurately, and the possibility is provided for improving the sea fog detection accuracy.
Drawings
FIG. 1 is a flow chart of the steps performed in the practice of the present invention;
FIG. 2 is a schematic structural diagram of a semantic segmentation model for MODIS sea fog detection according to the present invention;
FIG. 3 is a schematic diagram of the encoding-decoding structure of the semantic segmentation model for MODIS sea fog detection according to the present invention.
Detailed Description
In order to make the objects, contents and advantages of the present invention clearer, the following detailed description of the embodiments of the present invention is provided with the accompanying drawings and examples:
referring to fig. 1, the specific implementation steps of the semantic segmentation network model for MODIS sea fog detection are as follows:
(1) the method comprises the steps of constructing a sea fog sample set SeaFog _ Dataset, downloading an MODIS satellite remote sensing Image, labeling different sea ground objects in an Image through ArcGIS, wherein the sea ground objects comprise a clear air sea surface, sea fog and clouds, the clouds comprise medium-high clouds and low clouds, constructing a sea fog Label set SeaFog _ Label by using a labeling result, using a data set corresponding to the Label set SeaFog _ Label as a sea fog data set SeaFog _ Image, and jointly forming the sea fog sample set SeaFog _ Dataset by using the SeaFog _ Label and the sea fog Label.
(2) The method comprises the steps of constructing a semantic segmentation network for MODIS sea fog detection, wherein the overall structure of the constructed semantic segmentation model is shown in figure 2, so that a sea fog semantic segmentation model is obtained, the sea fog semantic segmentation model comprises 4 layers of networks, each layer of network comprises 2 parts, the first part is an encoding module, the second part is a decoding module, and the detailed structures of the two modules are shown in figure 3. And each layer of network performs summation operation on the feature maps obtained by completing the encoding and decoding operation, and obtains a final classification result through the softmax function layer.
2.1) the encoding module performs convolution operation and downsampling operation on the input SeaFog _ Dataset data set:
firstly, performing convolution operation with convolution kernel of 7 × 7 on an input SeaFog _ Dataset data set to obtain a Feature map E _ Feature _1, then performing 2-time coding on the E _ Feature _1 through 2 residual error modules to obtain a Feature map E _ Feature _2, wherein each coding comprises 2 convolution layers with convolution kernel size of 3 × 3 and 1 down-sampling layer, each convolution layer is accompanied by a non-linear activation function layer, and the calculation formula of the activation function is as follows:
wherein, x is an input characteristic, alpha is an adjustable factor, and the value alpha is 1.0;
2.2) the decoding module performs convolution operation and up-sampling operation on the Feature map E _ Feature _2 obtained by the 4-layer network coding module, and gradually restores the size of the original image:
firstly, generating a Feature map D _ Feature _1 through convolution operation with the convolution kernel size of 1 multiplied by 1, secondly obtaining a Feature map D _ Feature _2 through deconvolution operation with the convolution kernel size of 1 multiplied by 3, and then obtaining a Feature map D _ Feature _ front through convolution operation with the convolution kernel size of 1 multiplied by 1; the Feature map D _ Feature _ front obtains a Feature map D _ Feature _ cSE through 1 global pooling layer, 2 convolution operations with convolution kernel size of 1 × 1 × 1 and 1 nonlinear activation function layer; meanwhile, D _ Feature _ front obtains a Feature map D _ Feature _ sSE through 1 convolution operation with the convolution kernel size of 1 multiplied by 1 and 1 nonlinear activation function layer;
the decoding module performs an adding operation on the Feature map D _ Feature _ cSE and the Feature map D _ Feature _ sSE to obtain a Feature map D _ Feature _ scSE;
finally, adding the coding Feature diagram and the decoding Feature diagram of each layer to obtain a Feature diagram Feature _ 1;
2.3) carrying out post-processing operation on the finally obtained Feature map Feature _1 by the semantic segmentation network:
firstly, obtaining a Feature map Feature _2 through 1 deconvolution operation with convolution kernel size of 3 multiplied by 3 and 1 nonlinear activation function layer; secondly, obtaining a Feature map Feature _3 through convolution operation with 1 convolution kernel of which the size is 3 multiplied by 3 and deconvolution operation with 1 convolution kernel of which the size is 2 multiplied by 2; and finally, obtaining a final classification result through a softmax function layer.
(3) Training and verifying the constructed semantic segmentation network, and randomly dividing the sea fog sample set SeaFog _ Dataset obtained in the step (1) into a training set SeaFog _ Train and a testing set SeaFog _ Test according to a ratio of 8: 2; training a semantic segmentation network through a SeaFog _ Train, and verifying the detection precision of the semantic segmentation network through the SeaFog _ Test; the semantic segmentation network is a semantic segmentation model for three classifications of a clear air sea surface, sea fog and cloud regions, the semantic segmentation model is input as a sea fog training set SeaFog _ Train, and the semantic segmentation model is output as a sea ground object classification result;
the input batch size of the training set is 8, a loss function and an initial learning rate are set, ResNet18 model parameters are used as pre-training parameters, cross entropy loss between a marine ground feature classification result predicted by a semantic segmentation network and a sea fog Label set SeaFog _ Label (ground feature real Label) is loss, and a calculation formula is as follows:
FL(pt)＝-αt(1-pt)γlog(pt)
wherein alpha istFor the weighting factor, γ is used to adjust the sample weights, and α is sett＝0.25，γ＝2，ptReflecting the probability that the predicted value is the true value, ptThe calculation formula is as follows:
wherein p represents the probability that the prediction sample class is 1, and y is a real label;
and performing error back propagation by using an Adam optimization algorithm according to the loss error loss, updating the weight parameters of the network layer, verifying by using a cross verification method after each batch of training sets are trained, wherein the cross verification rate is 0.1, and training until the loss value of the loss function is lower than 0.1 to obtain the semantic segmentation network model finally used for sea fog detection.
The method comprises the steps of downloading an MODIS satellite remote sensing image, labeling different marine ground objects through ArcGIS, and taking a labeling result as a training label of a semantic segmentation model; a new semantic segmentation network model is constructed for sea fog detection, a semantic segmentation model is trained on a GPU by using a back propagation algorithm, a deep semantic segmentation network is used for MODIS sea fog detection, and ground object type discrimination is carried out pixel by pixel, so that the sea fog detection precision is improved, and the method has the advantages of being scientific, reasonable, high in automation degree and the like.
Claims (10)
1. A semantic segmentation network model for MODIS sea fog detection is characterized in that the semantic segmentation network model comprises the following steps:
(1) constructing an ocean fog sample set SeaFog _ Dataset, wherein the ocean fog sample set comprises an ocean fog data set SeaFog _ Image and an ocean fog Label set SeaFog _ Label;
(2) constructing a semantic segmentation network for MODIS sea fog detection, adding feature maps obtained by completing encoding and decoding operations in each layer of network, and obtaining a final classification result through a softmax function layer;
(2.1) performing an encoding operation:
completing coding operation by using a coding module, and performing convolution operation and downsampling operation on an input SeaFog _ Dataset data set by using the coding module to obtain a Feature map E _ Feature _ 2;
(2.2) performing a decoding operation:
carrying out convolution operation and up-sampling operation on the obtained Feature image E _ Feature _2 by using a decoding module, and gradually recovering the size of an original image;
firstly, obtaining a Feature map D _ Feature _ front through 1 convolution operation with the convolution kernel size of 1 multiplied by 1, 1 deconvolution operation and 1 convolution operation with the convolution kernel size of 1 multiplied by 1;
then the Feature map D _ Feature _ front obtains a Feature map D _ Feature _ cSE through 1 global pooling layer, 2 convolution operations with convolution kernel size of 1 × 1 × 1 and 1 nonlinear activation function layer;
meanwhile, D _ Feature _ front obtains a Feature map D _ Feature _ sSE through 1 convolution operation with the convolution kernel size of 1 multiplied by 1 and 1 nonlinear activation function layer;
the decoding module performs an adding operation on the Feature map D _ Feature _ cSE and the Feature map D _ Feature _ sSE to obtain a Feature map D _ Feature _ scSE;
finally, adding the coding Feature diagram and the decoding Feature diagram of each layer to obtain a Feature diagram Feature _ 1;
(2.3) carrying out post-processing operation on the finally obtained Feature map Feature _1 by the semantic segmentation network, and obtaining a final classification result through a softmax function layer;
(3) and training and verifying the constructed semantic segmentation network to obtain a semantic segmentation network model finally used for sea fog detection.
2. The semantic segmentation network model for MODIS sea fog detection as claimed in claim 1, wherein the semantic segmentation network in step (2) is 4 layers.
3. The semantically-segmented network model for MODIS sea fog detection as claimed in claim 1 or 2, wherein the encoding operation in step (2.1) comprises the steps of:
firstly, carrying out 1 convolution operation on an input SeaFog _ Dataset data set to obtain a Feature map E _ Feature _ 1;
and then 2 times of coding is carried out on the E _ Feature _1 through 2 residual modules to obtain a Feature map E _ Feature _2, each coding comprises 2 convolutional layers and 1 downsampling layer, a nonlinear activation function layer is accompanied behind each convolutional layer, and the calculation formula of the activation function is as follows:
wherein, x is the input characteristic, α is the adjustable factor, and the value α is 1.0.
4. The semantic segmentation network model for MODIS sea fog detection as claimed in claim 3, wherein the convolution kernel size of the 1 convolution operation in step (2.1) is 7 x 7; the convolution kernel size in the 2 convolutional layers is 3 × 3.
5. The semantic segmentation network model for MODIS sea fog detection as claimed in claim 1, wherein in step (2.3), specifically:
firstly, obtaining a Feature map Feature _2 through 1 deconvolution operation and 1 nonlinear activation function layer;
secondly, obtaining a Feature map Feature _3 through 1 convolution operation and 1 deconvolution operation;
and finally, obtaining a final classification result through a softmax function layer.
6. The semantic segmentation network model for MODIS sea fog detection as claimed in claim 5, wherein the convolution kernel size of the 1 deconvolution operation in step (2.3) is 3 x 3; in the 1 convolution operation and the 1 deconvolution operation, the convolution kernel size of the 1 convolution operation is 3 × 3; the convolution kernel size for the 1 deconvolution operation is 2 × 2.
7. The semantic segmentation network model for MODIS sea fog detection as claimed in claim 1, wherein in step (3), the training and verification of the semantic segmentation network comprises the following steps:
randomly dividing the sea fog sample set SeaFog _ Dataset obtained in the step (1) into a training set SeaFog _ Train and a testing set SeaFog _ Test according to a proportion; training a semantic segmentation network through a SeaFog _ Train, and verifying the detection precision of the semantic segmentation network through the SeaFog _ Test; inputting a sea fog training set SeaFog _ Train and outputting a sea ground feature classification result predicted by a semantic segmentation network;
inputting a training set in batches, setting a loss function and an initial learning rate, taking ResNet18 model parameters as pre-training parameters, and calculating the cross entropy loss between the marine terrain classification result predicted by the semantic segmentation network and the sea fog Label set SeaFog _ Label as loss according to the formula:
FL(pt)＝-αt(1-pt)γlog(pt)
wherein alpha istFor the weighting factor, γ is used to adjust the sample weights, and α is sett＝0.25，γ＝2，ptReflecting the probability that the predicted value is the true value, ptThe calculation formula is as follows:
wherein p represents the probability that the prediction sample class is 1, and y is a real label;
and performing error back propagation by using an Adam optimization algorithm according to the loss error loss, updating the weight parameters of the network layer, verifying by using a cross verification method after each batch of training sets are trained, wherein the cross verification rate is 0.1, and training until the loss value of the loss function is lower than 0.1 to obtain the semantic segmentation network model finally used for sea fog detection.
8. The semantic segmentation network model for MODIS sea fog detection according to claim 7, wherein the training set SeaFog _ Train is larger than the Test set SeaFog _ Test.
9. The semantic segmentation network model for MODIS sea fog detection according to claim 8, wherein the ratio of the training set SeaFog _ Train to the Test set SeaFog _ Test is 8: 2.
10. The model of semantic segmentation network for MODIS sea fog detection as claimed in claim 7, wherein the classification of semantic segmentation network includes clear sky sea surface, sea fog and cloud region.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110676781.6A CN113393466A (en) | 2021-06-18 | 2021-06-18 | Semantic segmentation network model for MODIS sea fog detection |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110676781.6A CN113393466A (en) | 2021-06-18 | 2021-06-18 | Semantic segmentation network model for MODIS sea fog detection |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113393466A true CN113393466A (en) | 2021-09-14 |
Family
ID=77621921
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110676781.6A Pending CN113393466A (en) | 2021-06-18 | 2021-06-18 | Semantic segmentation network model for MODIS sea fog detection |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113393466A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114022782A (en) * | 2021-11-01 | 2022-02-08 | 中国石油大学(华东) | Sea fog detection method based on MODIS satellite data |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110208880A (en) * | 2019-06-05 | 2019-09-06 | 北京邮电大学 | A kind of sea fog detection method based on deep learning and satellite remote sensing technology |
| CN111553391A (en) * | 2020-04-09 | 2020-08-18 | 东南大学 | Feature fusion method in semantic segmentation technology |
| CN112183452A (en) * | 2020-10-15 | 2021-01-05 | 国家气象中心 | Sea fog identification method based on meteorological satellite image and depth migration learning |
| CN112464745A (en) * | 2020-11-09 | 2021-03-09 | 中国科学院计算机网络信息中心 | Ground feature identification and classification method and device based on semantic segmentation |
| CN112560624A (en) * | 2020-12-08 | 2021-03-26 | 中南大学 | High-resolution remote sensing image semantic segmentation method based on model depth integration |
-
2021
- 2021-06-18 CN CN202110676781.6A patent/CN113393466A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110208880A (en) * | 2019-06-05 | 2019-09-06 | 北京邮电大学 | A kind of sea fog detection method based on deep learning and satellite remote sensing technology |
| CN111553391A (en) * | 2020-04-09 | 2020-08-18 | 东南大学 | Feature fusion method in semantic segmentation technology |
| CN112183452A (en) * | 2020-10-15 | 2021-01-05 | 国家气象中心 | Sea fog identification method based on meteorological satellite image and depth migration learning |
| CN112464745A (en) * | 2020-11-09 | 2021-03-09 | 中国科学院计算机网络信息中心 | Ground feature identification and classification method and device based on semantic segmentation |
| CN112560624A (en) * | 2020-12-08 | 2021-03-26 | 中南大学 | High-resolution remote sensing image semantic segmentation method based on model depth integration |
Non-Patent Citations (2)
| Title |
|---|
| 张家强等: "基于深度残差全卷积网络的Landsat8遥感影像云检测方法", 《激光与光电子学进展》 * |
| 李楠: "基于深度学习的卫星图像中海雾识别的研究", 《中国优秀博硕士学位论文全文数据库(硕士) 基础科学辑》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114022782A (en) * | 2021-11-01 | 2022-02-08 | 中国石油大学(华东) | Sea fog detection method based on MODIS satellite data |
| CN114022782B (en) * | 2021-11-01 | 2024-05-10 | 中国石油大学(华东) | Sea fog detection method based on MODIS satellite data |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110909666B (en) | Night vehicle detection method based on improved YOLOv3 convolutional neural network | |
| CN111079683B (en) | Remote sensing image cloud and snow detection method based on convolutional neural network | |
| CN113780296B (en) | Remote sensing image semantic segmentation method and system based on multi-scale information fusion | |
| CN113780211A (en) | Lightweight aircraft detection method based on improved yolk 4-tiny | |
| CN106910202B (en) | Image segmentation method and system for ground object of remote sensing image | |
| CN112434672A (en) | Offshore human body target detection method based on improved YOLOv3 | |
| CN111753677B (en) | Multi-angle remote sensing ship image target detection method based on characteristic pyramid structure | |
| CN112183203A (en) | Real-time traffic sign detection method based on multi-scale pixel feature fusion | |
| CN112488025B (en) | Double-temporal remote sensing image semantic change detection method based on multi-modal feature fusion | |
| CN112949633B (en) | Improved YOLOv 3-based infrared target detection method | |
| CN112508079B (en) | Fine identification method, system, equipment, terminal and application of ocean frontal surface | |
| CN113256649B (en) | Remote sensing image station selection and line selection semantic segmentation method based on deep learning | |
| CN111241970A (en) | SAR image sea surface ship detection method based on yolov3 algorithm and sliding window strategy | |
| CN115471754A (en) | Remote sensing image road extraction method based on multi-dimensional and multi-scale U-net network | |
| CN115830466A (en) | Glacier change remote sensing detection method based on deep twin neural network | |
| CN116309348A (en) | Lunar south pole impact pit detection method based on improved TransUnet network | |
| CN113393466A (en) | Semantic segmentation network model for MODIS sea fog detection | |
| CN112418229A (en) | Unmanned ship marine scene image real-time segmentation method based on deep learning | |
| CN116363610A (en) | Improved YOLOv 5-based aerial vehicle rotating target detection method | |
| CN114898204B (en) | Rail transit peripheral dangerous source detection method based on deep learning | |
| CN117011219A (en) | Method, apparatus, device, storage medium and program product for detecting quality of article | |
| CN115457509A (en) | Traffic sign image segmentation algorithm based on improved space-time image convolution | |
| CN112861869A (en) | Sorghum lodging image segmentation method based on lightweight convolutional neural network | |
| CN113971755B (en) | All-weather sea surface target detection method based on improved YOLOV model | |
| CN113837080A (en) | Small target detection method based on information enhancement and receptive field enhancement |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20210914 |