CN112819792A - DualNet-based urban area change detection method - Google Patents
DualNet-based urban area change detection method Download PDFInfo
- Publication number
- CN112819792A CN112819792A CN202110153438.3A CN202110153438A CN112819792A CN 112819792 A CN112819792 A CN 112819792A CN 202110153438 A CN202110153438 A CN 202110153438A CN 112819792 A CN112819792 A CN 112819792A
- Authority
- CN
- China
- Prior art keywords
- image
- change detection
- dualnet
- change
- detection method
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 33
- 238000012549 training Methods 0.000 claims abstract description 21
- 230000011218 segmentation Effects 0.000 claims abstract description 3
- 238000000034 method Methods 0.000 claims description 17
- 238000010606 normalization Methods 0.000 claims description 3
- 238000011176 pooling Methods 0.000 claims description 3
- 230000004913 activation Effects 0.000 claims 1
- 230000007547 defect Effects 0.000 abstract description 4
- 238000010586 diagram Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000012937 correction Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 230000002708 enhancing effect Effects 0.000 description 2
- 238000003709 image segmentation Methods 0.000 description 2
- 230000009191 jumping Effects 0.000 description 2
- 230000002123 temporal effect Effects 0.000 description 2
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/60—Rotation of whole images or parts thereof
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20112—Image segmentation details
- G06T2207/20132—Image cropping
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30181—Earth observation
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to the technical field of urban area change detection, in particular to an urban area change detection method based on DualNet, which belongs to the technical field of detection and is characterized by comprising the following steps: (1) reading a data set, reading image pairs in a change detection data set produced by a geographic information base, each image pair comprising a pre-change image and a post-change image of the region; after reading the data set, taking the image pairs before and after the change as input in a mode of a double-temporal image; (2) a data enhancement strategy is adopted after data are read, so that an overfitting phenomenon is avoided during network training; (3) and (3) constructing a semantic segmentation network DualNet, respectively inputting the image pairs into an encoder during input, and performing skip connection and decoding operation after obtaining the characteristic image. The invention has the advantages that the defect that the traditional change detection method cannot automatically acquire image information can be overcome, and a better prediction result can be obtained through a new jump connection mode.
Description
Technical Field
The invention relates to the technical field of urban area change detection, in particular to an urban area change detection method based on DualNet.
Background
Today, more and more satellites are launched into space to monitor the earth, and therefore telemetry data such as satellite images is applied to a number of telemetry tasks, of which change detection is one of the most important. Change detection provides useful information for a number of areas, such as city management, ecosystem monitoring, and national land use map updates. Conventional change detection algorithms can obtain the result of change detection, but the methods cannot automatically obtain the information of the change area in the image, so that the result of change detection is too dependent on human judgment.
Due to the rapid development of deep learning in recent years, the image processing field is greatly improved, including the fields of image classification, image retrieval, semantic segmentation and the like. The deep learning algorithm can spontaneously learn deep features of an image without human participation, so that aiming at overcoming the defect that the traditional change detection method cannot automatically acquire image information, the invention provides a city region change detection method based on DualNet based on Unet and SegNet, and the network uses improved skip-connected context-skip and can effectively learn the image context information.
Disclosure of Invention
The invention provides a city region change detection method based on DualNet aiming at the requirement of city change detection, and the method can automatically acquire the image characteristics of a change region without human participation.
The invention comprises the following steps:
step 1, reading image pairs in change detection data set produced by a geographic information base, wherein each image pair comprises an image before the change of the area and an image after the change of the area, and the change area comprises water body change, ground change, low vegetation change, tree change, building change, sports field change and the like. Data set image pairs can be divided into training sets, test sets, and test sets.
In the present invention, each image pixel is assigned to a changed region and an unchanged region, and change detection is regarded as an image segmentation task, thereby performing training of a neural network. Since the change detection is based on the image data of two different time periods, the input image is a double temporal image.
And 2, enhancing data. In order to avoid the over-fitting phenomenon generated during network training, the data enhancement strategy is adopted after data is read, and the method comprises the following operations:
(1)90 °, 180 ° and 270 ° rotation operations;
(2) performing random rotation operation at intervals of 3-20 degrees;
(3) taking a random factor in [0.75, 1.25] to carry out scaling operation;
(4) a random brightness change operation;
(5) random contrast variation operation;
(6) converting the image into HSV space, and the like.
Meanwhile, because the network setting input is 256x256, the invention randomly performs clipping after enhancement to meet the network training requirement.
And 3, constructing a network based on Context-Skip. The network system structure used by the invention is called DualNet, and the image pair is respectively input and coded when in input, and then jump-connection and decoding operation are carried out after the characteristic image is obtained.
1. An encoder. In the encoder branch, there are four convolutional blocks, each of which is formed by stacking the following operations in sequence: (1) a convolutional layer, 3 layers in total;
(2) a batch normalization layer;
(3) the function Relu is activated.
A maximum pooling layer for downsampling is also stacked at the end of the convolution block. At the encoder stage, the image pairs are input separately to obtain two feature maps.
2. And (4) jumping and connecting. In order to obtain the information of the change area, the invention carries out jump-connection splicing on the characteristic diagram obtained in the encoding stage. At this stage, the invention provides a context-skip connection method. The method firstly uses convolution with the size of 3x3 as a window to aggregate the context information in a neighborhood of the feature map, then splices the feature pairs, and uses convolution with the size of 1x1 to perform size correction.
3. A decoder. In the decoder branch, the jump-connected feature maps are subjected to decoding operation. Similar to the encoder branch, there are 4 convolution blocks, each stacked once by the above operations, except that at the end of each convolution block, a bilinear interpolation layer for upsampling is stacked.
Finally, the network uses SoftMax to get a result graph.
And 4, training the network. Network training is carried out according to the set hyper-parameters, and in the method, a loss function used in training is focal loss;
so far, training of the DualNet-based urban area change detection method model is completed.
Compared with the prior art, the invention has the beneficial effects that: the invention has the advantages that the defect that the traditional change detection method cannot automatically acquire image information can be overcome, and a better prediction result can be obtained through a new jump connection mode.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the principles of the invention and not to limit the invention.
Fig. 1 is a schematic diagram of an overall structure of a method for detecting urban regional variation based on DualNet according to the present invention;
fig. 2 is a schematic diagram of a skip-connection context-skip mode of the method for detecting urban regional variation based on DualNet according to the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments.
In the description of the present invention, it is to be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", and the like, indicate orientations or positional relationships based on the orientations or positional relationships shown in the drawings, are merely for convenience in describing the present invention and simplifying the description, and do not indicate or imply that the device or element being referred to must have a particular orientation, be constructed and operated in a particular orientation, and thus, should not be construed as limiting the present invention.
Referring to fig. 1 and 2, a method for detecting urban regional variation based on DualNet includes the following steps:
step 1, reading image pairs in change detection data set produced by a geographic information base, wherein each image pair comprises an image before the change of the area and an image after the change of the area, and the change area comprises water body change, ground change, low vegetation change, tree change, building change, sports field change and the like. Data set image pairs can be divided into training sets, test sets, and test sets.
In the present invention, each image pixel is assigned to a changed region and an unchanged region, and change detection is regarded as an image segmentation task, thereby performing training of a neural network. Since the change detection is based on the image data of two different time periods, the input image is a double temporal image.
And 2, enhancing data. In order to avoid the over-fitting phenomenon generated during network training, the data enhancement strategy is adopted after data is read, and the method comprises the following operations:
(1)90 °, 180 ° and 270 ° rotation operations;
(2) performing random rotation operation at intervals of 3-20 degrees;
(3) taking a random factor in [0.75, 1.25] to carry out scaling operation;
(4) a random brightness change operation;
(5) random contrast variation operation;
(6) converting the image into HSV space, and the like.
Meanwhile, because the network setting input is 256x256, the invention randomly performs clipping after enhancement to meet the network training requirement.
And 3, constructing a network based on Context-Skip. The network system structure used by the invention is called DualNet, and the image pair is respectively input and coded when in input, and then jump-connection and decoding operation are carried out after the characteristic image is obtained.
1. An encoder. In the encoder branch, there are four convolutional blocks, each of which is formed by stacking the following operations in sequence: (1) a convolutional layer, 3 layers in total;
(2) a batch normalization layer;
(3) the function Relu is activated.
A maximum pooling layer for downsampling is also stacked at the end of the convolution block. At the encoder stage, the image pairs are input separately to obtain two feature maps.
2. And (4) jumping and connecting. In order to obtain the information of the change area, the invention carries out jump-connection splicing on the characteristic diagram obtained in the encoding stage. At this stage, the invention provides a context-skip connection method. The method firstly uses convolution with the size of 3x3 as a window to aggregate the context information in a neighborhood of the feature map, then splices the feature pairs, and uses convolution with the size of 1x1 to perform size correction.
3. A decoder. In the decoder branch, the jump-connected feature maps are subjected to decoding operation. Similar to the encoder branch, there are 4 convolution blocks, each stacked once by the above operations, except that at the end of each convolution block, a bilinear interpolation layer for upsampling is stacked.
Finally, the network uses SoftMax to get a result graph.
And 4, training the network. Network training is carried out according to the set hyper-parameters, and in the method, a loss function used in training is focal loss;
the method has the advantages that the defect that the traditional change detection method cannot automatically acquire image information can be overcome, and a better prediction result can be obtained through a new jump connection mode.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art should be considered to be within the technical scope of the present invention, and the technical solutions and the inventive concepts thereof according to the present invention should be equivalent or changed within the scope of the present invention.
Claims (7)
1. A DualNet-based urban area change detection method is characterized by comprising the following steps:
(1) reading a data set, reading image pairs in a change detection data set produced by a geographic information base, each image pair comprising a pre-change image and a post-change image of the region; after reading the data set, taking the image pairs before and after the change as input in a mode of a double-temporal image;
(2) data enhancement, namely adopting a data enhancement strategy after reading data to avoid an overfitting phenomenon during network training;
(3) constructing a semantic segmentation network DualNet, respectively inputting image pairs into an encoder during input, and performing skip connection and decoding operation after obtaining a characteristic image;
(4) and (5) network training, namely performing network training according to the set hyper-parameters to obtain a model.
2. The DualNet-based urban area change detection method according to claim 1, wherein said data enhancement strategy comprises the steps of:
(1)90 °, 180 ° and 270 ° rotation operations;
(2) performing random rotation operation at intervals of 3-20 degrees;
(3) taking a random factor in [0.75, 1.25] to carry out scaling operation;
(4) a random brightness change operation;
(5) random contrast variation operation;
(6) converting the image into HSV space, and the like.
3. The method as claimed in claim 2, wherein the network setting input is 256 × 256, so that after the enhancement, the network training requirement is satisfied by randomly clipping.
4. The DualNet-based urban area change detection method according to claim 3, wherein said encoder has four convolutional blocks in the encoder branch, and each convolutional block is formed by stacking the following operations in sequence:
(1) a convolutional layer, 3 layers in total;
(2) a batch normalization layer;
(3) an activation function Relu;
at the end of the convolution block, a maximum pooling layer for downsampling is also stacked, and at the encoder stage, the image pairs are input separately, resulting in two feature maps.
5. The DualNet-based urban area change detection method according to claim 4, wherein the jump-connection is to obtain the information of the change area and to perform jump-connection splicing on the feature map obtained in the encoding stage; the jump connection is a context-skip jump connection method.
6. The DualNet-based urban area change detection method according to claim 5, wherein said decoder performs a decoding operation on the jump-connected feature map in a decoder branch; similar to the encoder branch, the decoder has 4 convolutional blocks, each of which is stacked once by the above operations, except that at the end of each convolutional block, a bilinear interpolation layer for upsampling is stacked, and finally the network uses SoftMax to obtain a result graph.
7. The method of claim 1, wherein the loss function used in training the network is focal loss.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110153438.3A CN112819792A (en) | 2021-02-03 | 2021-02-03 | DualNet-based urban area change detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110153438.3A CN112819792A (en) | 2021-02-03 | 2021-02-03 | DualNet-based urban area change detection method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112819792A true CN112819792A (en) | 2021-05-18 |
Family
ID=75861155
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110153438.3A Pending CN112819792A (en) | 2021-02-03 | 2021-02-03 | DualNet-based urban area change detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112819792A (en) |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108776772A (en) * | 2018-05-02 | 2018-11-09 | 北京佳格天地科技有限公司 | Across the time building variation detection modeling method of one kind and detection device, method and storage medium |
| CN109635711A (en) * | 2018-12-07 | 2019-04-16 | 上海衡道医学病理诊断中心有限公司 | A kind of pathological image dividing method based on deep learning network |
| CN109657551A (en) * | 2018-11-15 | 2019-04-19 | 天津大学 | A kind of method for detecting human face based on contextual information enhancing |
| CN110136170A (en) * | 2019-05-13 | 2019-08-16 | 武汉大学 | A kind of remote sensing image building change detecting method based on convolutional neural networks |
| CN110738663A (en) * | 2019-09-06 | 2020-01-31 | 上海衡道医学病理诊断中心有限公司 | Double-domain adaptive module pyramid network and unsupervised domain adaptive image segmentation method |
| CN111612750A (en) * | 2020-05-13 | 2020-09-01 | 中国矿业大学 | Overlapping chromosome segmentation network based on multi-scale feature extraction |
| CN111709290A (en) * | 2020-05-18 | 2020-09-25 | 杭州电子科技大学 | Crowd counting method based on coding and decoding-jumping connection scale pyramid network |
| CN111860204A (en) * | 2020-06-29 | 2020-10-30 | 成都数之联科技有限公司 | Multi-temporal remote sensing image change detection method and medium based on semantic segmentation technology |
| WO2020238123A1 (en) * | 2019-05-31 | 2020-12-03 | Guangdong Oppo Mobile Telecommunications Corp., Ltd. | Method, system, and computer-readable medium for improving color quality of images |
-
2021
- 2021-02-03 CN CN202110153438.3A patent/CN112819792A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108776772A (en) * | 2018-05-02 | 2018-11-09 | 北京佳格天地科技有限公司 | Across the time building variation detection modeling method of one kind and detection device, method and storage medium |
| CN109657551A (en) * | 2018-11-15 | 2019-04-19 | 天津大学 | A kind of method for detecting human face based on contextual information enhancing |
| CN109635711A (en) * | 2018-12-07 | 2019-04-16 | 上海衡道医学病理诊断中心有限公司 | A kind of pathological image dividing method based on deep learning network |
| CN110136170A (en) * | 2019-05-13 | 2019-08-16 | 武汉大学 | A kind of remote sensing image building change detecting method based on convolutional neural networks |
| WO2020238123A1 (en) * | 2019-05-31 | 2020-12-03 | Guangdong Oppo Mobile Telecommunications Corp., Ltd. | Method, system, and computer-readable medium for improving color quality of images |
| CN110738663A (en) * | 2019-09-06 | 2020-01-31 | 上海衡道医学病理诊断中心有限公司 | Double-domain adaptive module pyramid network and unsupervised domain adaptive image segmentation method |
| CN111612750A (en) * | 2020-05-13 | 2020-09-01 | 中国矿业大学 | Overlapping chromosome segmentation network based on multi-scale feature extraction |
| CN111709290A (en) * | 2020-05-18 | 2020-09-25 | 杭州电子科技大学 | Crowd counting method based on coding and decoding-jumping connection scale pyramid network |
| CN111860204A (en) * | 2020-06-29 | 2020-10-30 | 成都数之联科技有限公司 | Multi-temporal remote sensing image change detection method and medium based on semantic segmentation technology |
Non-Patent Citations (3)
| Title |
|---|
| 刘玉良 等: "《深度学习》", 30 November 2019 * |
| 焦李成 等: "减少过拟合", 《人工智能前沿技术丛书 计算智能导论》 * |
| 马晓路 等: "《MATLAB图像处理从入门到精通》", 28 February 2013 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113298818B (en) | Remote sensing image building segmentation method based on attention mechanism and multi-scale features | |
| CN109241972B (en) | Image semantic segmentation method based on deep learning | |
| CN113780296B (en) | Remote sensing image semantic segmentation method and system based on multi-scale information fusion | |
| CN111259906B (en) | Method for generating remote sensing image target segmentation countermeasures under condition containing multilevel channel attention | |
| CN113011427A (en) | Remote sensing image semantic segmentation method based on self-supervision contrast learning | |
| CN110599401A (en) | Remote sensing image super-resolution reconstruction method, processing device and readable storage medium | |
| CN111860233B (en) | SAR image complex building extraction method and system based on attention network selection | |
| CN114187450A (en) | Remote sensing image semantic segmentation method based on deep learning | |
| CN111160351A (en) | Fast high-resolution image segmentation method based on block recommendation network | |
| CN111882620A (en) | Road drivable area segmentation method based on multi-scale information | |
| CN113609889A (en) | High-resolution remote sensing image vegetation extraction method based on sensitive feature focusing perception | |
| CN110751111A (en) | Road extraction method and system based on high-order spatial information global automatic perception | |
| CN112749578A (en) | Remote sensing image automatic road extraction method based on deep convolutional neural network | |
| CN117058367A (en) | Semantic segmentation method and device for high-resolution remote sensing image building | |
| CN113689382A (en) | Tumor postoperative life prediction method and system based on medical images and pathological images | |
| CN116091929A (en) | Remote sensing image semantic segmentation method combining Unet and Transformer | |
| CN116206214A (en) | Automatic landslide recognition method, system, equipment and medium based on lightweight convolutional neural network and double attention | |
| CN114092824A (en) | Remote sensing image road segmentation method combining intensive attention and parallel up-sampling | |
| CN113628180B (en) | Remote sensing building detection method and system based on semantic segmentation network | |
| CN112869704B (en) | Diabetic retinopathy area automatic segmentation method based on circulation self-adaptive multi-target weighting network | |
| CN113920421B (en) | Full convolution neural network model capable of achieving rapid classification | |
| Jiang et al. | AGD-Linknet: a road semantic segmentation model for high resolution remote sensing images integrating attention mechanism, gated decoding block and dilated convolution | |
| CN117726954A (en) | Sea-land segmentation method and system for remote sensing image | |
| CN117611813A (en) | Remote sensing image segmentation method based on optical image-SAR image feature alignment | |
| CN112819792A (en) | DualNet-based urban area change detection method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210518 |