CN114519819A - Remote sensing image target detection method based on global context awareness - Google Patents
Remote sensing image target detection method based on global context awareness Download PDFInfo
- Publication number
- CN114519819A CN114519819A CN202210126106.0A CN202210126106A CN114519819A CN 114519819 A CN114519819 A CN 114519819A CN 202210126106 A CN202210126106 A CN 202210126106A CN 114519819 A CN114519819 A CN 114519819A
- Authority
- CN
- China
- Prior art keywords
- feature
- target
- features
- feature map
- candidate region
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 33
- 238000011176 pooling Methods 0.000 claims abstract description 23
- 238000000034 method Methods 0.000 claims abstract description 9
- 238000012549 training Methods 0.000 claims description 17
- 238000013528 artificial neural network Methods 0.000 claims description 15
- 238000012360 testing method Methods 0.000 claims description 7
- 238000004422 calculation algorithm Methods 0.000 claims description 6
- 238000012795 verification Methods 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 3
- 230000008447 perception Effects 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 2
- 230000006870 function Effects 0.000 claims 1
- 238000000605 extraction Methods 0.000 abstract description 5
- 239000000284 extract Substances 0.000 abstract description 3
- WDLRUFUQRNWCPK-UHFFFAOYSA-N Tetraxetan Chemical compound OC(=O)CN1CCN(CC(O)=O)CCN(CC(O)=O)CCN(CC(O)=O)CC1 WDLRUFUQRNWCPK-UHFFFAOYSA-N 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
Images
Classifications
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a remote sensing image target detection method based on global context awareness, which comprises the steps of extracting the characteristics of an image by using a depth residual error Network (ResNet 101), further extracting the characteristics by using a characteristic Pyramid Network (FPN, Feature Pyramid Network) and generating a candidate region; after the candidate region is generated, using the feature pooling alignment feature; adding a global context extraction module at the highest layer of the feature extraction network, and fusing extracted features and original features in an addition mode to obtain new features; and finally, classifying the new features by using the full connection layer to generate a target class and a frame. The method fully extracts scene information of the image by utilizing the characteristic of rich high-level feature semantic information, further enhances feature representation, increases the recognition accuracy of dense targets, and also improves the recognition accuracy of other targets to a certain extent, thereby integrally improving the target detection performance in the remote sensing image.
Description
Technical Field
The invention belongs to the technical field of pattern recognition, and particularly relates to a remote sensing image target detection method.
Background
Remote sensing image analysis is always a hot point of research in the field of computer vision and is widely applied to the fields of urban planning, land utilization management, environmental monitoring and the like. The target detection is a basic task in the field of computer vision, and can provide support for subsequent tasks such as event detection, target tracking, human-computer interaction, scene segmentation and the like. The remote sensing image is usually shot from high altitude, and the shooting angle and the shooting height are not fixed due to the difference of an onboard or satellite-borne sensor. Compared with natural images, the remote sensing image has richer scene information, more target types and dense arrangement, so that the remote sensing image target detection faces a great challenge. Due to the problems, although some algorithms have been proposed for remote sensing image target detection, the performance still has room for improvement, so that remote sensing image target detection is still one of the hot problems of current research.
Stanku (the feature-enhanced SSD algorithm and the application thereof in remote sensing target detection), photonics newspaper 2020,49(01): 154-. The method improves the extraction capability of the network on the small target features by designing a shallow feature enhancement module; and designing a deep feature enhancement module to replace a deep network in the SSD pyramid feature layer. However, the method does not fully utilize rich scene information in the remote sensing image, and the improvement effect is limited.
The original FPN detection algorithm has a poor detection effect on dense targets in a remote sensing image because of the lack of sufficient scene information in a feature pyramid network. The detection of dense objects needs to rely on scene information, for example, cars are only present in parking lots or roads, and cars are generally present around cars. Thus, lack of awareness of context information, i.e., global context, makes it difficult for the network to identify dense targets.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides a remote sensing image target detection method based on global context sensing, which comprises the steps of extracting the characteristics of an image by using a depth residual error Network (ResNet 101), further extracting the characteristics by using a characteristic Pyramid Network (FPN) and generating a candidate region; after the candidate region is generated, using the feature pooling alignment feature; adding a global context extraction module at the highest layer of the feature extraction network, and fusing extracted features and original features in an addition mode to obtain new features; and finally, classifying the new features by using the full connection layer to generate a target class and a frame. The method fully extracts scene information of the image by utilizing the characteristic of rich high-level feature semantic information, further enhances feature representation, increases the recognition accuracy of dense targets, and also improves the recognition accuracy of other targets to a certain extent, thereby integrally improving the target detection performance in the remote sensing image.
The technical scheme adopted by the invention for solving the technical problem comprises the following steps:
step 1: preprocessing and dividing the data set;
uniformly cutting the marked images in the standard data set into a plurality of 1024-by-1024 images, wherein the width and the height of the images are respectively reserved with the coincidence rate of 10% of pixels during cutting, and then randomly dividing the images into a training set, a verification set and a test set, wherein the training set, the verification set and the test set are not intersected;
step 2: constructing a target detection deep neural network and training the target detection deep neural network by adopting a gradient descent and back propagation algorithm; the target detection deep neural network firstly adopts Res101 residual error network to extract features, then uses feature pyramid network FPN to generate candidate regions, then carries out local context perception on the candidate regions, and finally obtains target categories and bounding boxes through feature pooling and full connection layers, wherein the method specifically comprises the following steps:
step 2-1: initializing Res101 model parameters by using a pre-training model;
step 2-2: inputting 1024 × 1024 images into a Res101 residual network to extract features, and generating 5 feature maps with different sizes, which are marked as C1-C6, and the scales are 512 × 512,256 × 256,128 × 128,64 × 64,32 × 32 and 16 × 16 respectively;
step 2-3: performing global maximum pooling on the feature map C6 to obtain scene features containing scene information; the scene features are subjected to 10 × 10 convolution and 1 × 1 convolution to obtain global features;
step 2-4: taking the feature map C5 as a feature map P5 of a feature pyramid;
upsampling the feature map C5, and adding the upsampled feature map C4 subjected to 1-by-1 convolution to generate a feature map P4 of a feature pyramid;
upsampling the feature map C4, and adding the upsampled feature map C3 subjected to 1-by-1 convolution to generate a feature map P3 of a feature pyramid;
upsampling the feature map C3, and adding the upsampled feature map C2 subjected to 1-by-1 convolution to generate a feature map P2 of a feature pyramid;
step 2-5: feature maps P2, P3, P4, and P5 of the feature pyramid are 256 in size, respectively2、1282、642、322(ii) a Generating anchor points anchorages for each feature image in the feature pyramid by using an area generation network, wherein the aspect ratio corresponding to each anchorage comprises three types, namely 1:2, 1:1 and 2: 1; thus, the feature pyramid generates 15 different anchors;
and generating a target candidate region by using the anchor, wherein the calculation formula is as follows:
wherein (x)c,yc) As the anchor point coordinates, (w, h) are the width and height of the target candidate region, respectively, (x)1,y1) And (x)2,y2) Coordinates of the upper left corner and the lower right corner of the target candidate area;
calculating the intersection ratio IoU of the target candidate region and the real label: if IoU is greater than or equal to 0.7, the target candidate area is set as a positive sample; if IoU <0.3, set the target candidate region to a negative sample; taking the obtained positive sample and negative sample as labels of the training target candidate area;
step 2-6: performing feature pooling on the target candidate region, and calculating a feature layer corresponding to the target candidate region after the feature pooling by adopting an equation (2):
wherein 1024 refers to the input image size, k0Is a reference value;
since the target candidate region is generated from four different feature maps P2, P3, P4 and P5 through anchors, the feature pooling corresponds to 4 different feature layers;
the value rules of 4 different feature layers after the feature pooling are as follows:
after the target candidate regions in the feature maps P2, P3, P4 and P5 are subjected to feature pooling, each target candidate region respectively outputs 7 × 7 results, that is, 49 features are extracted;
step 2-7: adding the 49 features obtained in the step 2-6 and the global features obtained in the step 2-3, and sequentially inputting the 49 features and the global features into two full-connection layers, wherein the output results of the two full-connection layers are a target category and a target boundary box;
and step 3: and inputting the remote sensing image to be detected into the trained target detection depth neural network, and outputting to obtain a class boundary box of the target.
Preferably, k is0＝4。
The invention has the following beneficial effects:
the method fully extracts scene information of the image by utilizing the characteristic of rich high-level feature semantic information, further enhances feature representation, increases the recognition accuracy of dense targets, and also improves the recognition accuracy of other targets to a certain extent, thereby integrally improving the target detection performance in the remote sensing image.
Drawings
FIG. 1 is a flow chart of the method of the present invention.
FIG. 2 is a diagram of a target detection deep neural network according to the present invention.
Detailed Description
The invention is further illustrated with reference to the following figures and examples.
The invention discloses a remote sensing image target detection method based on global context sensing, which is designed to improve the accuracy of target identification in a remote sensing image by extracting global context characteristic enhancement feature representation.
As shown in fig. 1, a remote sensing image target detection method based on global context awareness includes the following steps:
step 1: the DOTA dataset is processed. Because the size of the original data image of the DOTA data set is not fixed and the labeled data of the test set is not disclosed, 1869 images with labels are uniformly cut into 1024 x 1024 images for the convenience of neural network training, and the width and the height of each image are respectively reserved with the overlapping rate of 10% of pixels in order to prevent the target from being lost due to image cutting during cutting. The 19219 images and the labeling information thereof are obtained after processing, and are randomly divided into 11531 images of the training set, 3844 images of the verification set and 3844 images of the test set, so that no intersection exists among the training set, the verification set and the test set in the image sample space.
Step 2: as shown in fig. 2, a target detection deep neural network is constructed and trained by using gradient descent and back propagation algorithm; the target detection deep neural network firstly adopts Res101 residual error network to extract features, then uses feature pyramid network FPN to generate candidate regions, then carries out local context perception on the candidate regions, and finally obtains target categories and bounding boxes through feature pooling and full connection layers, wherein the method specifically comprises the following steps:
step 2-1: because the neural network parameters are various and are difficult to train, before the model is trained, a pre-training model is used for initializing Res101 model parameters;
step 2-2: training a neural network on a training data set, inputting 1024 × 1024 images into a Res101 residual network to extract features, and generating 5 feature maps with different sizes, which are marked as C1-C6, and the scales are 512 × 512,256 × 256,128 × 128,64 × 64,32 × 32 and 16 × 16 respectively; selecting C2, C3, C4, and C5 establishes a pyramid. If C1 is used, too much memory is occupied, so the pyramid is not built by C1;
step 2-3: performing global maximum pooling on the feature map C6 to obtain scene features containing scene information; the scene features are subjected to 10 × 10 convolution and 1 × 1 convolution to obtain global features;
step 2-4: taking the feature map C5 as a feature map P5 of a feature pyramid;
upsampling the feature map C5, and adding the upsampled feature map C4 subjected to 1-by-1 convolution to generate a feature map P4 of a feature pyramid;
upsampling the feature map C4, and adding the upsampled feature map C3 subjected to 1-by-1 convolution to generate a feature map P3 of a feature pyramid;
upsampling the feature map C3, and adding the upsampled feature map C2 subjected to 1-by-1 convolution to generate a feature map P2 of a feature pyramid;
the 1 × 1 convolution is to ensure that the number of added feature map channels is the same;
step 2-5: the feature maps P2, P3, P4 and P5 of the feature pyramid are 256 in size respectively2、1282、642、322(ii) a Generating anchor points anchorages for each feature image in the feature pyramid by using an area generation network, wherein the aspect ratio corresponding to each anchorage comprises three types, namely 1:2, 1:1 and 2: 1; thus, the feature pyramid generates 15 different anchors;
and generating a target candidate region by using an anchor, wherein the calculation formula is as follows:
wherein (x)c,yc) As the anchor point coordinates, (w, h) are the width and height of the target candidate region, respectively, (x)1,y1) And (x)2,y2) Coordinates of the upper left corner and the lower right corner of the target candidate area;
calculating the intersection ratio IoU (intersection over Union) of the target candidate region and the real label: if IoU is more than or equal to 0.7, the target candidate area is set as a positive sample; if IoU <0.3, set the target candidate region to a negative sample; using the obtained positive sample and negative sample as labels of training target candidate regions (corresponding to each anchor);
step 2-6: performing feature pooling on the target candidate region, and calculating a feature layer corresponding to the target candidate region after the feature pooling by adopting an equation (2):
wherein 1024 refers to the input image size, k0Taking 4 as a reference value generally;
since the target candidate region is generated from four different feature maps P2, P3, P4 and P5 through anchors, the feature pooling corresponds to 4 different feature layers;
the value rules of 4 different feature layers after the feature pooling are as follows:
after the target candidate regions in the feature maps P2, P3, P4 and P5 are subjected to feature pooling, each target candidate region respectively outputs 7 × 7 results, that is, 49 features are extracted;
step 2-7: adding the 49 features obtained in the step 2-6 and the global features obtained in the step 2-3, and sequentially inputting the 49 features and the global features into two full connection layers, wherein the output results of the two full connection layers are a target category and a target bounding box;
and step 3: and inputting the remote sensing image to be detected into the trained target detection depth neural network, and outputting to obtain a class boundary box of the target.
Claims (2)
1. A remote sensing image target detection method based on global context sensing is characterized by comprising the following steps:
step 1: preprocessing and dividing the data set;
uniformly cutting the marked images in the standard data set into a plurality of 1024 x 1024 images, respectively reserving the coincidence rate of 10% of pixels in width and height during cutting, and randomly dividing the images into a training set, a verification set and a test set, wherein the training set, the verification set and the test set have no intersection;
and 2, step: constructing a target detection deep neural network and training the target detection deep neural network by adopting a gradient descent and back propagation algorithm; the target detection deep neural network firstly adopts Res101 residual error network to extract features, then uses feature pyramid network FPN to generate candidate regions, then carries out local context perception on the candidate regions, and finally obtains target categories and bounding boxes through feature pooling and full connection layers, wherein the method specifically comprises the following steps:
step 2-1: initializing Res101 model parameters by using a pre-training model;
step 2-2: inputting 1024 × 1024 images into a Res101 residual network to extract features, and generating 5 feature maps with different sizes, which are marked as C1-C6, and the scales are 512 × 512,256 × 256,128 × 128,64 × 64,32 × 32 and 16 × 16 respectively;
step 2-3: performing global maximum pooling on the feature map C6 to obtain scene features containing scene information; the scene features are subjected to 10 × 10 convolution and 1 × 1 convolution to obtain global features;
step 2-4: taking the feature map C5 as a feature map P5 of a feature pyramid;
upsampling the feature map C5, and adding the upsampled feature map C4 with the feature map C4 subjected to 1-by-1 convolution to generate a feature map P4 of a feature pyramid;
upsampling the feature map C4, and adding the upsampled feature map C3 subjected to 1-by-1 convolution to generate a feature map P3 of a feature pyramid;
upsampling the feature map C3, and adding the upsampled feature map C2 subjected to 1-by-1 convolution to generate a feature map P2 of a feature pyramid;
step 2-5: the feature maps P2, P3, P4 and P5 of the feature pyramid are 256 in size respectively2、1282、642、322(ii) a Generating anchor points anchorages for each feature image in the feature pyramid by using an area generation network, wherein the aspect ratio corresponding to each anchorage comprises three types, namely 1:2, 1:1 and 2: 1; thus, the feature pyramid generates 15 different anchors;
and generating a target candidate region by using the anchor, wherein the calculation formula is as follows:
wherein (x)c,yc) As the anchor point coordinates, (w, h) are the width and height of the target candidate region, respectively, (x)1,y1) And (x)2,y2) Coordinates of the upper left corner and the lower right corner of the target candidate area;
calculating the intersection ratio IoU of the target candidate region and the real label: if IoU is greater than or equal to 0.7, the target candidate area is set as a positive sample; if IoU <0.3, set the target candidate region to a negative sample; taking the obtained positive sample and negative sample as labels of the training target candidate area;
step 2-6: performing feature pooling on the target candidate region, and calculating a feature layer corresponding to the target candidate region after the feature pooling by adopting an equation (2):
wherein 1024 refers to the input image size, k0Is a reference value;
since the target candidate region is generated from four different feature maps P2, P3, P4 and P5 through anchors, the feature pooling corresponds to 4 different feature layers;
the value rules of 4 different feature layers after the feature pooling are as follows:
after the target candidate regions in the feature maps P2, P3, P4 and P5 are subjected to feature pooling, each target candidate region respectively outputs 7 × 7 results, that is, 49 features are extracted;
step 2-7: adding the 49 features obtained in the step 2-6 and the global features obtained in the step 2-3, and sequentially inputting the 49 features and the global features into two full-connection layers, wherein the output results of the two full-connection layers are a target category and a target boundary box;
and step 3: and inputting the remote sensing image to be detected into the trained target detection depth neural network, and outputting to obtain a class boundary box of the target.
2. The remote sensing image target detection method based on global context awareness, as recited in claim 1, wherein k is a function of the global context awareness0＝4。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210126106.0A CN114519819B (en) | 2022-02-10 | 2022-02-10 | Remote sensing image target detection method based on global context awareness |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210126106.0A CN114519819B (en) | 2022-02-10 | 2022-02-10 | Remote sensing image target detection method based on global context awareness |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114519819A true CN114519819A (en) | 2022-05-20 |
| CN114519819B CN114519819B (en) | 2024-04-02 |
Family
ID=81596492
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210126106.0A Active CN114519819B (en) | 2022-02-10 | 2022-02-10 | Remote sensing image target detection method based on global context awareness |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114519819B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115937672A (en) * | 2022-11-22 | 2023-04-07 | 南京林业大学 | Remote sensing rotating target detection method based on deep neural network |
| CN116486077A (en) * | 2023-04-04 | 2023-07-25 | 中国科学院地理科学与资源研究所 | Remote sensing image semantic segmentation model sample set generation method and device |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018214195A1 (en) * | 2017-05-25 | 2018-11-29 | 中国矿业大学 | Remote sensing imaging bridge detection method based on convolutional neural network |
| CN111368775A (en) * | 2020-03-13 | 2020-07-03 | 西北工业大学 | Complex scene dense target detection method based on local context sensing |
| WO2020181685A1 (en) * | 2019-03-12 | 2020-09-17 | 南京邮电大学 | Vehicle-mounted video target detection method based on deep learning |
| CN112070729A (en) * | 2020-08-26 | 2020-12-11 | 西安交通大学 | Anchor-free remote sensing image target detection method and system based on scene enhancement |
| CN112766409A (en) * | 2021-02-01 | 2021-05-07 | 西北工业大学 | Feature fusion method for remote sensing image target detection |
| CN113111740A (en) * | 2021-03-27 | 2021-07-13 | 西北工业大学 | Characteristic weaving method for remote sensing image target detection |
-
2022
- 2022-02-10 CN CN202210126106.0A patent/CN114519819B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018214195A1 (en) * | 2017-05-25 | 2018-11-29 | 中国矿业大学 | Remote sensing imaging bridge detection method based on convolutional neural network |
| WO2020181685A1 (en) * | 2019-03-12 | 2020-09-17 | 南京邮电大学 | Vehicle-mounted video target detection method based on deep learning |
| CN111368775A (en) * | 2020-03-13 | 2020-07-03 | 西北工业大学 | Complex scene dense target detection method based on local context sensing |
| CN112070729A (en) * | 2020-08-26 | 2020-12-11 | 西安交通大学 | Anchor-free remote sensing image target detection method and system based on scene enhancement |
| CN112766409A (en) * | 2021-02-01 | 2021-05-07 | 西北工业大学 | Feature fusion method for remote sensing image target detection |
| CN113111740A (en) * | 2021-03-27 | 2021-07-13 | 西北工业大学 | Characteristic weaving method for remote sensing image target detection |
Non-Patent Citations (1)
| Title |
|---|
| 乔文凡;慎利;戴延帅;曹云刚;: "联合膨胀卷积残差网络和金字塔池化表达的高分影像建筑物自动识别", 地理与地理信息科学, no. 05, 27 August 2018 (2018-08-27) * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115937672A (en) * | 2022-11-22 | 2023-04-07 | 南京林业大学 | Remote sensing rotating target detection method based on deep neural network |
| CN116486077A (en) * | 2023-04-04 | 2023-07-25 | 中国科学院地理科学与资源研究所 | Remote sensing image semantic segmentation model sample set generation method and device |
| CN116486077B (en) * | 2023-04-04 | 2024-04-30 | 中国科学院地理科学与资源研究所 | Remote sensing image semantic segmentation model sample set generation method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114519819B (en) | 2024-04-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112200161B (en) | Face recognition detection method based on mixed attention mechanism | |
| CN110622213B (en) | System and method for depth localization and segmentation using 3D semantic maps | |
| CN110188705B (en) | Remote traffic sign detection and identification method suitable for vehicle-mounted system | |
| Matzen et al. | Nyc3dcars: A dataset of 3d vehicles in geographic context | |
| CN112183203B (en) | Real-time traffic sign detection method based on multi-scale pixel feature fusion | |
| CN112084869B (en) | Compact quadrilateral representation-based building target detection method | |
| CN111461039B (en) | Landmark identification method based on multi-scale feature fusion | |
| CN114519819B (en) | Remote sensing image target detection method based on global context awareness | |
| CN113688836A (en) | Real-time road image semantic segmentation method and system based on deep learning | |
| CN112287983B (en) | Remote sensing image target extraction system and method based on deep learning | |
| CN112560675A (en) | Bird visual target detection method combining YOLO and rotation-fusion strategy | |
| CN112766409A (en) | Feature fusion method for remote sensing image target detection | |
| Zang et al. | Traffic lane detection using fully convolutional neural network | |
| CN111368775A (en) | Complex scene dense target detection method based on local context sensing | |
| Li et al. | An aerial image segmentation approach based on enhanced multi-scale convolutional neural network | |
| CN117152414A (en) | Target detection method and system based on scale attention auxiliary learning method | |
| Meng et al. | A block object detection method based on feature fusion networks for autonomous vehicles | |
| CN113111740A (en) | Characteristic weaving method for remote sensing image target detection | |
| CN112597996A (en) | Task-driven natural scene-based traffic sign significance detection method | |
| CN114494893B (en) | Remote sensing image feature extraction method based on semantic reuse context feature pyramid | |
| CN112446292B (en) | 2D image salient object detection method and system | |
| Li et al. | Learning to holistically detect bridges from large-size vhr remote sensing imagery | |
| CN114550016A (en) | Unmanned aerial vehicle positioning method and system based on context information perception | |
| Wang et al. | Extraction of main urban roads from high resolution satellite images by machine learning | |
| WO2024000728A1 (en) | Monocular three-dimensional plane recovery method, device, and storage medium |
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 |