CN110175548A - Remote sensing images building extracting method based on attention mechanism and channel information - Google Patents
Remote sensing images building extracting method based on attention mechanism and channel information Download PDFInfo
- Publication number
- CN110175548A CN110175548A CN201910419275.1A CN201910419275A CN110175548A CN 110175548 A CN110175548 A CN 110175548A CN 201910419275 A CN201910419275 A CN 201910419275A CN 110175548 A CN110175548 A CN 110175548A
- Authority
- CN
- China
- Prior art keywords
- building
- channel information
- attention mechanism
- remote sensing
- sensing images
- 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
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
- G06F18/217—Validation; Performance evaluation; Active pattern learning techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/176—Urban or other man-made structures
-
- 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
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The remote sensing images building extracting method based on attention mechanism and channel information that the present invention provides a kind of, it is intended to solve the not high technical problem of existing structure extracting method precision.Method includes the following steps: step 1, acquisition training sample;Step 2 builds deep learning frame and constructs the U-shaped network based on attention mechanism and channel information;Step 3 carries out enhancing processing to training sample;Step 4, allowable loss function optimize it;Step 5, the U-shaped network of training obtain building and extract model;Step 6, building extract;The beneficial technical effect of the present invention lies in: it can quickly and accurately extract building.
Description
Technical field
The present invention relates to technical field of image segmentation, and in particular to a kind of remote sensing based on attention mechanism and channel information
Image building extracting method.
Background technique
Architecture information is in military surveillance, the update of GIS-Geographic Information System (GIS) data, urban planning, diaster prevention and control, resource tune
It looks into and is played an important role in digital urban construction etc. application.With the development of remote sensing technology, a large amount of remotely-sensed data at
It is possible.However, since road and place context are complicated, it is difficult to extract buildings in common RGB remote sensing images.Therefore,
It develops reliable and accurate building extracting method and has become an important and challenging research topic.
In the past few decades, the extraction research of many buildings is all based on traditional image processing method.In spy
It levies in engineering, traditional method is using spectrum, shape and texture as input feature vector, support vector machines (SVM), random forest (RF)
With Adaboost as sorting algorithm.However, these methods are to extract information from simple remote sensing scene mostly, not yet building
The higher complex area of diversity is built to be assessed and used.
Meanwhile with the fast development of urban construction, building is to be easiest to increase in geographical data bank and change,
And the part of update is needed most, and it is often huge to update workload.Develop a kind of high degree of automation and accuracy
High technology is very necessary.
Summary of the invention
The technical problem to be solved in the present invention is to provide a kind of remote sensing images based on attention mechanism and channel information to build
Object extracting method is built, to solve the not high enough disadvantage of existing method extraction accuracy.
In order to solve the above technical problems, the present invention adopts the following technical scheme: a kind of believed based on attention mechanism and channel
The remote sensing images building extracting method of breath, comprising the following steps:
Step 1, acquisition training sample: equipment is acquired by remote sensing images and acquires target image, target image is marked
Remember and to pre-processing, so that image is met call format, building training sample set;
Step 2, the U-shaped network built deep learning frame and construct the extraction of remote sensing images building: U-shaped network is based on note
Power mechanism of anticipating and channel information;
Step 3 carries out enhancing processing to training sample: training sample that step 1 is acquired and marked carry out data enhancing,
Including overturning, cutting, rotation, color jitter, normalization;
Step 4, allowable loss function optimize it: Lovasz loss function are used, with Adam optimizer to network
Carry out study optimization.
Step 5, the U-shaped network of training obtain building and extract model: by the collected training sample of step 1 by step 3
Building is obtained by U-shaped network training after data enhancing and extracts model;
Step 6, building extract: building extraction is carried out using the U-shaped network based on attention mechanism and channel information,
Remote sensing images to be extracted are inputted, output is the building coordinate information divided.
Further, the training sample of step 1 acquires equipment by remote sensing images and acquires target image, and uses and manually will
Building extracts.
Further, step 2 uses U-shaped network integration attention mechanism and channel information, wherein attention mechanism
It is inputted according to the coding stage of U-shaped networkWith the output of a upper layer decoderGain attention force coefficient αi∈
[0,1], and have and be by the output of attention mechanismCalculation method is as follows:
Wherein, σ2Activation primitive, and have
Channel information is then using excitation and squeeze operation and used in the feature extraction coding stage of U-shaped network, first basis
Space C × H × W is compressed to C × 1 × 1 and obtained by following formula
Again by motivating operation to obtain zoom factor s:
S=Fex(z, W)=σ (g (z, W))=σ (W2σ(W1z))
Wherein, σ is Relu activation primitive,
Eventually pass through the output of compression and excitation are as follows:
xc=Fscale(uc,sc)=sc·uc。
Further, step 3 image preprocessing use image enhancement, including (- 180 °, 180 °) of random angles rotation and
Random scaled (0.5 times -2 times), color jitter, normalization.
Further, the Lovasz loss function that step 4 is taken, is improved to symmetrical type function, i.e., by:
It improves are as follows:
Wherein, y*It is label value, FiIt (x) is that network exports, x representative image,It is loss vector.
The invention has the following advantages over the prior art:
The present invention is damaged by the way that attention mechanism and channel information are introduced U-shaped network with improved symmetric form Lovasz
Function training network is lost, the extraction accuracy of building, and convergence rate when training network are improved.
Detailed description of the invention
Fig. 1 is general frame figure of the invention;
Fig. 2 is for improved attention schematic diagram of mechanism;
Fig. 3 is the residual error module of the channel information for encoder stage;
Fig. 4 is the training loss curve of the present invention and Unet.
Specific embodiment
Specific embodiments of the present invention will be described in detail with reference to the accompanying drawing.But following embodiment is used only in detail
Illustrate the present invention, does not limit the scope of the invention in any way.Program that is involved or relying on is in following embodiment
The conventional program or simple program of the art, those skilled in the art can make conventional selection according to concrete application scene
Or it is adaptively adjusted.
As shown in Figure 1-3, a kind of remote sensing images building extraction side based on attention mechanism and channel information of the present invention
Method, comprising the following steps:
Step 1, acquisition training sample: equipment is acquired by remote sensing images and acquires target image, target image is marked
Remember and to pre-processing, so that image is met call format, building training sample set;
Step 2, the U-shaped network built deep learning frame and construct the extraction of remote sensing images building: U-shaped network is based on note
Power mechanism of anticipating and channel information;
Step 3 carries out enhancing processing to training sample: training sample that step 1 is acquired and marked carry out data enhancing,
Including overturning, cutting, rotation, color jitter, normalization etc.;
Step 4, allowable loss function optimize it: Lovasz loss function are used, with Adam optimizer to network
Carry out study optimization;
Step 5, the U-shaped network of training obtain building and extract model: by the collected training sample of step 1 by step 3
Building is obtained by U-shaped network training after data enhancing and extracts model;
Step 6, building extract: building extraction is carried out using the U-shaped network based on attention mechanism and channel information,
Remote sensing images to be extracted are inputted, output is the building coordinate information divided.
In order to verify effectiveness of the invention, selected from Unet as comparative example, the building data set of standard is used
(1000 samples) comes comparison result, the Average Accuracy and recall rate of comparison algorithm.
Table 1: the result of embodiment and comparative example on data set compares
Method | Accuracy rate | Recall rate |
Unet | 0.862 | 0.878 |
The present invention | 0.870 | 0.884 |
As shown in Table 1, algorithm has promotion in accuracy rate and recall rate compared to Unet.
Trained loss curve is as shown in figure 4, the deeper curve of color is the training curve of Unet, the shallower curve of color
It is training curve of the invention, it is seen that training curve convergence rate of the invention is faster.
Claims (5)
1. a kind of remote sensing images building extracting method based on attention mechanism and channel information, which is characterized in that including with
Lower step:
Step 1, acquisition training sample: equipment is acquired by remote sensing images and acquires target image, target image is marked simultaneously
To pre-processing, image is made to meet call format, constructs training sample set;
Step 2, the U-shaped network built deep learning frame and construct the extraction of remote sensing images building: U-shaped network is based on attention
Mechanism and channel information;
Step 3 carries out enhancing processing to training sample: training sample that step 1 is acquired and marked carry out data enhancing including
Overturning, cutting, rotation, color jitter, normalization;
Step 4, allowable loss function optimize it: using Lovasz loss function, carried out with Adam optimizer to network
Study optimization;
Step 5, the U-shaped network of training obtain building and extract model: by the collected training sample of step 1 by step 3 data
Building is obtained by U-shaped network training after enhancing and extracts model;
Step 6, building extract: carrying out building extraction, input using the U-shaped network based on attention mechanism and channel information
Remote sensing images to be extracted, output are the building coordinate information divided.
2. a kind of remote sensing images building extraction side based on attention mechanism and channel information according to claim 1
Method, which is characterized in that the training sample of step 1 acquires equipment by remote sensing images and acquires target image, and using will manually build
Build object extraction.
3. a kind of remote sensing images building extraction side based on attention mechanism and channel information according to claim 1
Method, which is characterized in that step 2 uses U-shaped network integration attention mechanism and channel information, wherein attention mechanism according to
The coding stage of U-shaped network inputsWith the output of a upper layer decoderGain attention force coefficient αi∈[0,
1], and have and be by the output of attention mechanismCalculation method is as follows:
Wherein σ2Activation primitive, and have
Channel information is then using excitation and squeeze operation and used in the feature extraction coding stage of U-shaped network, first according to following
Space C × H × W is compressed to C × 1 × 1 and obtained by formula
Again by motivating operation to obtain zoom factor s:
S=Fex(z, W)=σ (g (z, W))=σ (W2σ(W1z))
Wherein, σ is Relu activation primitive,
Eventually pass through the output of compression and excitation are as follows:
Xc=Fscale(uc,sc)=sc·uc。
4. a kind of remote sensing images building extraction side based on attention mechanism and channel information according to claim 1
Method, which is characterized in that step 3 image preprocessing use image enhancement, including (- 180 °, 180 °) of random angles rotation and with
Machine scaled (0.5 times -2 times), color jitter, normalization.
5. a kind of remote sensing images building extraction side based on attention mechanism and channel information according to claim 1
Method, which is characterized in that the Lovasz loss function that step 4 is taken is improved to symmetrical type function, i.e., by:
It improves are as follows:
Wherein, y*It is label value, FiIt (x) is that network exports, x representative image,It is loss vector.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910419275.1A CN110175548B (en) | 2019-05-20 | 2019-05-20 | Remote sensing image building extraction method based on attention mechanism and channel information |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910419275.1A CN110175548B (en) | 2019-05-20 | 2019-05-20 | Remote sensing image building extraction method based on attention mechanism and channel information |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110175548A true CN110175548A (en) | 2019-08-27 |
CN110175548B CN110175548B (en) | 2022-08-23 |
Family
ID=67691749
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910419275.1A Active CN110175548B (en) | 2019-05-20 | 2019-05-20 | Remote sensing image building extraction method based on attention mechanism and channel information |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110175548B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111144335A (en) * | 2019-12-30 | 2020-05-12 | 自然资源部国土卫星遥感应用中心 | Method and device for building deep learning model |
CN111652852A (en) * | 2020-05-08 | 2020-09-11 | 浙江华睿科技有限公司 | Method, device and equipment for detecting surface defects of product |
CN112434663A (en) * | 2020-12-09 | 2021-03-02 | 国网湖南省电力有限公司 | Power transmission line forest fire detection method, system and medium based on deep learning |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108564109A (en) * | 2018-03-21 | 2018-09-21 | 天津大学 | A kind of Remote Sensing Target detection method based on deep learning |
CN108921173A (en) * | 2018-06-01 | 2018-11-30 | 中南大学 | A kind of deep learning method of combination OSM and remote sensing image extraction overpass |
CN109446992A (en) * | 2018-10-30 | 2019-03-08 | 苏州中科天启遥感科技有限公司 | Remote sensing image building extracting method and system, storage medium, electronic equipment based on deep learning |
CN109544579A (en) * | 2018-11-01 | 2019-03-29 | 上海理工大学 | A method of damage building is assessed after carrying out calamity using unmanned plane |
CN112084869A (en) * | 2020-08-10 | 2020-12-15 | 北京航空航天大学 | Compact quadrilateral representation-based building target detection method |
-
2019
- 2019-05-20 CN CN201910419275.1A patent/CN110175548B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108564109A (en) * | 2018-03-21 | 2018-09-21 | 天津大学 | A kind of Remote Sensing Target detection method based on deep learning |
CN108921173A (en) * | 2018-06-01 | 2018-11-30 | 中南大学 | A kind of deep learning method of combination OSM and remote sensing image extraction overpass |
CN109446992A (en) * | 2018-10-30 | 2019-03-08 | 苏州中科天启遥感科技有限公司 | Remote sensing image building extracting method and system, storage medium, electronic equipment based on deep learning |
CN109544579A (en) * | 2018-11-01 | 2019-03-29 | 上海理工大学 | A method of damage building is assessed after carrying out calamity using unmanned plane |
CN112084869A (en) * | 2020-08-10 | 2020-12-15 | 北京航空航天大学 | Compact quadrilateral representation-based building target detection method |
Non-Patent Citations (3)
Title |
---|
ABDELILAH ADIBA ETC.: "Transfer learning and U-Net for buildings segmentation", 《SMC "19: PROCEEDINGS OF THE NEW CHALLENGES IN DATA SCIENCES: ACTS OF THE SECOND CONFERENCE OF THE MOROCCAN CLASSIFICATION SOCIETY》 * |
伍广明 等: "基于U型卷积神经网络的航空影像建筑物检测", 《测绘学报》 * |
顾炼 等: "基于FlowS-Unet的遥感图像建筑物变化检测", 《自动化学报》 * |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111144335A (en) * | 2019-12-30 | 2020-05-12 | 自然资源部国土卫星遥感应用中心 | Method and device for building deep learning model |
CN111652852A (en) * | 2020-05-08 | 2020-09-11 | 浙江华睿科技有限公司 | Method, device and equipment for detecting surface defects of product |
CN111652852B (en) * | 2020-05-08 | 2024-03-29 | 浙江华睿科技股份有限公司 | Product surface defect detection method, device and equipment |
CN112434663A (en) * | 2020-12-09 | 2021-03-02 | 国网湖南省电力有限公司 | Power transmission line forest fire detection method, system and medium based on deep learning |
CN112434663B (en) * | 2020-12-09 | 2023-04-07 | 国网湖南省电力有限公司 | Power transmission line forest fire detection method, system and medium based on deep learning |
Also Published As
Publication number | Publication date |
---|---|
CN110175548B (en) | 2022-08-23 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108537743B (en) | Face image enhancement method based on generation countermeasure network | |
US20230186056A1 (en) | Grabbing detection method based on rp-resnet | |
CN114022432B (en) | Insulator defect detection method based on improved yolov5 | |
CN111047551A (en) | Remote sensing image change detection method and system based on U-net improved algorithm | |
CN106897714A (en) | A kind of video actions detection method based on convolutional neural networks | |
CN110175548A (en) | Remote sensing images building extracting method based on attention mechanism and channel information | |
CN111160533A (en) | Neural network acceleration method based on cross-resolution knowledge distillation | |
CN110060286B (en) | Monocular depth estimation method | |
CN110827312B (en) | Learning method based on cooperative visual attention neural network | |
CN111612017A (en) | Target detection method based on information enhancement | |
CN110675421B (en) | Depth image collaborative segmentation method based on few labeling frames | |
CN112989995B (en) | Text detection method and device and electronic equipment | |
CN113592822A (en) | Insulator defect positioning method for power inspection image | |
CN112528858A (en) | Training method, device, equipment, medium and product of human body posture estimation model | |
CN109447111A (en) | A kind of remote sensing supervised classification method based on subclass training sample | |
CN114170608A (en) | Super-resolution text image recognition method, device, equipment and storage medium | |
CN110992374A (en) | Hair refined segmentation method and system based on deep learning | |
CN111860683A (en) | Target detection method based on feature fusion | |
CN116699096B (en) | Water quality detection method and system based on deep learning | |
CN111339924A (en) | Polarized SAR image classification method based on superpixel and full convolution network | |
CN112800982A (en) | Target detection method based on remote sensing scene classification | |
CN115620081A (en) | Training method of target detection model, target detection method and device | |
CN115953408A (en) | YOLOv 7-based lightning arrester surface defect detection method | |
CN101916381B (en) | Object contour extraction method based on sparse representation | |
CN114581789A (en) | Hyperspectral image classification method and system |
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 |