CN109145730B - Automatic semantic segmentation method for mining area in remote sensing image - Google Patents
Automatic semantic segmentation method for mining area in remote sensing image Download PDFInfo
- Publication number
- CN109145730B CN109145730B CN201810770020.5A CN201810770020A CN109145730B CN 109145730 B CN109145730 B CN 109145730B CN 201810770020 A CN201810770020 A CN 201810770020A CN 109145730 B CN109145730 B CN 109145730B
- Authority
- CN
- China
- Prior art keywords
- layer
- mining area
- network
- remote sensing
- convolution
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- 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
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)
- Astronomy & Astrophysics (AREA)
- Remote Sensing (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a method for automatically segmenting the semantics of a mining area in a remote sensing image, which belongs to the technical field of automatic target detection and deep learning and is characterized by comprising the following steps: the method comprises the following specific steps: step one, establishing a training sample set: acquiring remote sensing images of a mining area, manually drawing boundaries of the mining area to form boundary raster files, generating fishing nets of 448 x 448 and 512 x 512 by using ArcGIS, cutting the remote sensing images of the mining area in batches by adopting the two generated fishing nets to generate image blocks with different sizes as input data of a deep learning network, cutting the boundary files of the mining area raster in the images by the fishing nets to generate boundary files corresponding to the image blocks of each mining area as label data of the network; the invention can highly abstract the extracted features while keeping the feature integrity through the hybrid network Den-Res Net, can solve the feature redundancy problem of the Dense Net network, has high working efficiency, can automatically perform semantic separation, and has high accuracy.
Description
Technical Field
The invention relates to the technical field of automatic target detection and deep learning, in particular to a method for automatically segmenting the semantics of a mining area in a remote sensing image.
Background
The mining area target semantic segmentation is carried out on the multisource and multi-temporal remote sensing image of a certain place, the time-space change of the mining area target semantic segmentation can be rapidly and visually detected, the analysis of the mining area time-space change characteristics is facilitated, timely, objective and accurate technical support and detailed scientific data can be provided for the sustainable development of mineral resources, and meanwhile, a scientific basis is provided for the treatment and transformation of mining cities and the planning and construction of 'green cities'. Therefore, the semantic segmentation of the target in the image mining area has important significance.
At present, the method mainly comprises visual interpretation, pixel-based and object-oriented remote sensing image target semantic segmentation methods and the like. The visual interpretation method has low working efficiency, requires an interpreter to have certain professional knowledge and has strong subjectivity, the pixel-based target semantic segmentation and object-oriented target semantic segmentation methods are difficult to adapt to mass data, the dependent feature expression is manually designed and is time-consuming, and the dependence on the professional knowledge and the characteristics of the data is strong, so that an effective classifier is difficult to learn from the mass data so as to fully mine the association between the data. Based on the method, the invention designs an automatic semantic segmentation method for the mining area in the remote sensing image to solve the problems.
Disclosure of Invention
The invention aims to provide an automatic semantic segmentation method for a mining area in a remote sensing image, and the method is used for solving the problem that the existing method provided in the background technology is low in efficiency.
In order to achieve the purpose, the invention provides the following technical scheme: a mining area automatic semantic segmentation method in a remote sensing image is characterized by comprising the following steps: the method comprises the following specific steps:
step one, establishing a training sample set
(1) Acquiring a remote sensing image of a mining area, manually drawing the boundary of the mining area to form a boundary raster file;
(2) generating 448 x 448 and 512 x 512 two-scale fishing nets by utilizing ArcGIS;
(3) adopting the two generated fishing nets to cut the remote sensing images of the mining area in batches, and generating image blocks with different sizes as input data of a deep learning network;
(4) cutting the mine area grid boundary file in the image through the fishing net to generate a boundary file corresponding to each mine area image block as network label data;
step two, mining area segmentation construction model
Constructing a hybrid network Den-Res Net combining a DenseNet network and a Res Net network;
the Den-Res Net comprises an Input layer, a convolution layer, a Dense layer consisting of more than two convolution layers, a mean pooling layer with the window size of 2 multiplied by 2, a jump residual connecting layer, a feature superposition layer, a feature upsampling layer and a classification layer;
step three, training a mining area segmentation model
Randomly extracting 80% of data from the sample set constructed in the first step to serve as a training set; taking the mining area remote sensing image blocks in the training set as input data of the network constructed in the second step, obtaining a characteristic layer with high abstraction of the mining area images after the convolution layer, the Dense layer, the mean pooling layer and the jump residual connecting layer of the network are subjected to down-sampling, obtaining an output result after the up-sampling, the Dense layer, the jump residual connecting layer and the soft-max layer, forming a cross entropy by the result and the label data in the training set, adjusting the weight of the network by combining an error back propagation mechanism, repeatedly training, and performing cross training by combining the remaining 20% of data until the network converges to obtain an optimal deep learning network;
step four, image segmentation to be identified
And inputting the remote sensing image of the mining area to be subjected to semantic segmentation into a trained network to obtain a semantic segmentation result.
Preferably, the convolution kernel size of the convolutional layer is 7 × 7, and the step size is 2.
Preferably, skip connection is added in the convolutional layer convolution process.
Preferably, the density layer is formed of a convolutional layer having a convolutional kernel size of 3 × 3.
Preferably, the feature upsampling layer employs a 3 × 3 convolution kernel, as opposed to a convolution process.
Preferably, the convolution of the sense layer is performed after the obtained image and the pooled feature map in the down-sampling are superimposed after the feature is up-sampled by the layer.
Compared with the prior art, the invention has the beneficial effects that: the invention can highly abstract the extracted features while keeping the feature integrity through the hybrid network Den-Res Net, can solve the feature redundancy problem of the Dense Net network, has high working efficiency, can automatically perform semantic separation, and has high accuracy.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a flow chart of the remote sensing image mining area semantic segmentation of the present invention.
Fig. 2 is an image convolution.
FIG. 3 is a Den-Res Net network diagram of semantic segmentation of a remote sensing image mining area.
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. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Referring to fig. 1-3, the present invention provides a technical solution: a method for automatically segmenting semantics of a mining area in a remote sensing image comprises the following specific steps:
step one, establishing a training sample set
(1) Acquiring a remote sensing image of a mining area, manually drawing the boundary of the mining area to form a boundary raster file;
(2) generating 448 x 448 and 512 x 512 two-scale fishing nets by utilizing ArcGIS;
(3) adopting the two generated fishing nets to cut the remote sensing images of the mining area in batches, and generating image blocks with different sizes as input data of a deep learning network;
(4) cutting the mine area grid boundary file in the image through the fishing net to generate a boundary file corresponding to each mine area image block as network label data;
step two, mining area segmentation construction model
Constructing a hybrid network Den-Res Net combining a DenseNet network and a Res Net network;
the Den-Res Net comprises an Input layer, a convolution layer, a Dense layer consisting of more than two convolution layers, a mean pooling layer with the window size of 2 multiplied by 2, a jump residual connecting layer, a feature superposition layer, a feature upsampling layer and a classification layer;
step three, training a mining area segmentation model
Randomly extracting 80% of data from the sample set constructed in the first step to serve as a training set; taking the mining area remote sensing image blocks in the training set as input data of the network constructed in the second step, obtaining a characteristic layer with high abstraction of the mining area images after the convolution layer, the Dense layer, the mean pooling layer and the jump residual connecting layer of the network are subjected to down-sampling, obtaining an output result after the up-sampling, the Dense layer, the jump residual connecting layer and the soft-max layer, forming a cross entropy by the result and the label data in the training set, adjusting the weight of the network by combining an error back propagation mechanism, repeatedly training, and performing cross training by combining the remaining 20% of data until the network converges to obtain an optimal deep learning network;
step four, image segmentation to be identified
And inputting the remote sensing image of the mining area to be subjected to semantic segmentation into a trained network to obtain a semantic segmentation result.
The convolution kernel size of the convolution layer is 7 x 7, the step length is 2, the features of the convolution layer can be conveniently and better extracted, the size of the obtained feature image is 1/2 of the size of an original image, skip connection is added in the convolution process of the convolution layer, on one hand, the image is highly abstracted, on the other hand, the gradient diffusion phenomenon caused by excessive convolution layers can be avoided, the sense layer is formed by the convolution layer with the convolution kernel size of 3 x 3, the extracted features are well multiplexed, the feature upper sampling layer adopts 3 x 3 convolution kernels, and in contrast to the convolution process, the convolution of the sense layer is carried out after the obtained image and the feature map after pooling in down sampling are overlapped after the feature upper sampling layer, the superposition of the feature map is mainly used for utilizing the features of the lower layer of the image, and the sense layer can multiplex the features of the lower layer while the features of the upper sampling image are better extracted.
One specific application of this embodiment is:
1: image training set production
(1) Collecting a mining area remote sensing image, and manually marking the boundary of the mining area in the image to form a mining area boundary raster file;
(2) generating fishing nets with 448 x 448 and 512 x 512 scales by adopting ArcGIS;
(3) adopting the fishing net files produced in the step (2) to cut the mining area images and the corresponding boundary files in batches, and enabling the mining area images and the corresponding boundary files to correspond to one another;
(4) in order to increase the diversity of the samples, the images and the corresponding mine area boundary image pairs corresponding to the images and the images in the step (3) are simultaneously rotated and adjusted in size, and training images and corresponding labels are obtained. And the training pictures are represented as x1, x2, …, xn, and the corresponding labels are represented as y1, y2, …, yn.
2: network model training
The existing Dense Net network can multiplex multilayer features, has better performance in image semantic segmentation, and has serious feature redundancy along with the increase of a Dense layer. Because residual connection is added into the network, the Res Net can retain the integrity of the features and can perform high abstraction on the extracted features because of the residual connection, and the problem of feature redundancy of the Dense Net network can be solved by adopting the Res Net. Thus a hybrid network Den-Res Net is constructed that combines both networks.
For a better understanding of the network model, the image convolution is first explained:
the purpose of image convolution is to extract the features of an image to obtain an image feature map (as shown in fig. 2), wherein an input image is x, the size of the input image is h _ in × w _ in, the number of channels is c _ in, m _ c h _ k × w _ k convolution kernels are used for carrying out convolution operation to generate m _ c h _ out × w _ out feature maps, and a feature map size calculation formula is shown as a formula I
w _ out ═ w _ in +2 w _ pad-w _ k ÷ w _ str +1 expression one
In this case, the size of the feature map after convolution is identical to that of the original map, h _ str and w _ str are corresponding convolution kernel shift steps, and the pixel value of the generated feature map (i, j) is:
the semantic segmentation network model of the remote sensing image mining area is shown in fig. 3, and the following detailed description is given by combining training data to perform model training.
(1) In the data set generated in 1, 80% of the data was randomly decimated as data for network training, and first, images of two sizes were convolved with a convolution kernel size of 7 × 7 and a step size of 2, and this layer mainly functions to extract features of the images, and since the input image size is large and its features are extracted more favorably, a large field of view is used, and since the convolution step size is 2, the obtained feature image size is 1/2 of the original image size.
(2) Inputting the feature image extracted in the step (1) into two alternative Dense layers (consisting of two 3 x 3 convolutional layers) and an average pooling layer to obtain a feature image, and then inputting the feature image into one Dense layer (consisting of three 3 x 3 convolutional layers). Each layer of the Dense layer is connected with the previous layers, so that the extracted features are well multiplexed, and the mean pooling layer can reduce the image redundant information and the calculation dimensionality well.
(3) And (3) pooling the output characteristic image of the step (2), then performing two different convolution operations, wherein the first convolution operation is performed directly by 3 x 3, the second convolution operation is performed by 3 x 3, and the results obtained by the two convolution operations are subjected to jump connection. And continuing the operation on the feature image obtained by convolution. The main purpose of the convolution layer is to perform high abstraction on the image, so that only performing the convolution operations in (1) and (2) is not enough to extract enough information, and therefore, the convolution operation on the feature map needs to be continued, but an excessive number of sense layers cause serious feature redundancy, so that skip connection similar to Res Net is added when the image is convolved, on one hand, the high abstraction is performed on the image, and on the other hand, the gradient diffusion phenomenon caused by an excessive number of convolution layers can be avoided.
(4) After the steps (1), (2) and (3), highly abstract image features are obtained, the feature image size is 1/32 of the input image, in order to perform semantic segmentation on the image, the image size needs to be gradually changed into the original image size, firstly, the feature image is subjected to first up-sampling, and the feature image obtained by the first up-sampling and the feature image obtained by pooling in the down-sampling are superposed, so that the features of the low-level image can be better utilized.
(5) And (3) performing jump connection on the image features after superposition in the step (4) after two convolution operations similar to the operation in the step (3), mainly for performing feature extraction on the up-sampled image and avoiding gradient dispersion.
(6) And (5) performing up-sampling on the feature image obtained in the step (5) for four times, and performing convolution on a Dense layer after the obtained image is overlapped with the pooled feature image in the down-sampling after each up-sampling. The superposition of the feature maps is mainly used for utilizing the low-level features of the images, and the Dense layer can better extract the features of the images after up-sampling and simultaneously multiplex the low-level features.
(7) And finally, performing convolution operation on the image obtained in the step (6), wherein the size of the obtained image is consistent with that of the original image, the number of channels is 2, the image background and the mine area are respectively represented, the mine area boundary label corresponding to the image is generated in the step 1, the soft-max layer is adopted to establish an error function of the network, and the error backward propagation mechanism is utilized to update the weight of each layer of the network until the error is converged.
3: network testing
The remaining 20% of the data in 1 were used as test data for training the network. And inputting the cut image as input data into the trained network to obtain a mining area boundary result, and comparing the mining area boundary result with the real mining area boundary to obtain the accuracy of the network.
In the description herein, references to the description of "one embodiment," "an example," "a specific example" or the like are intended to mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
The preferred embodiments of the invention disclosed above are intended to be illustrative only. The preferred embodiments are not intended to be exhaustive or to limit the invention to the precise embodiments disclosed. Obviously, many modifications and variations are possible in light of the above teaching. The embodiments were chosen and described in order to best explain the principles of the invention and the practical application, to thereby enable others skilled in the art to best utilize the invention. The invention is limited only by the claims and their full scope and equivalents.
Claims (6)
1. A mining area automatic semantic segmentation method in a remote sensing image is characterized by comprising the following steps: the method comprises the following specific steps:
step one, establishing a training sample set
(1) Acquiring a remote sensing image of a mining area, manually drawing the boundary of the mining area to form a boundary raster file;
(2) generating 448 x 448 and 512 x 512 two-scale fishing nets by utilizing ArcGIS;
(3) adopting the two generated fishing nets to cut the remote sensing images of the mining area in batches, and generating image blocks with different sizes as input data of a deep learning network;
(4) cutting the mine area grid boundary file in the image through the fishing net to generate a boundary file corresponding to each mine area image block as network label data;
step two, mining area segmentation construction model
Constructing a hybrid network Den-Res Net combining a Dense Net network and a Res Net network;
the Den-Res Net comprises an Input layer, a convolution layer, a Dense layer consisting of more than two convolution layers, a mean pooling layer with the window size of 2 multiplied by 2, a jump residual connecting layer, a feature superposition layer, a feature upsampling layer and a classification layer;
step three, training a mining area segmentation model
Randomly extracting 80% of data from the sample set constructed in the first step to serve as a training set; taking the mining area remote sensing image blocks in the training set as input data of the network constructed in the second step, obtaining a characteristic layer with high abstraction of the mining area images after the convolution layer, the Dense layer, the mean pooling layer and the jump residual connecting layer of the network are subjected to down-sampling, obtaining an output result after the up-sampling, the Dense layer, the jump residual connecting layer and the soft-max layer, forming a cross entropy by the result and the label data in the training set, adjusting the weight of the network by combining an error back propagation mechanism, repeatedly training, and performing cross training by combining the remaining 20% of data until the network converges to obtain an optimal deep learning network;
step four, image segmentation to be identified
And inputting the remote sensing image of the mining area to be subjected to semantic segmentation into a trained network to obtain a semantic segmentation result.
2. The method for automatic semantic segmentation of the mining area in the remote sensing image according to claim 1, characterized in that: the convolution kernel size of the convolutional layer is 7 × 7, and the step size is 2.
3. The method for automatic semantic segmentation of the mining area in the remote sensing image according to claim 1, characterized in that: skip connection is added in the convolution layer convolution process.
4. The method for automatic semantic segmentation of the mining area in the remote sensing image according to claim 1, characterized in that: the Dense layer is formed by convolution layers with convolution kernel size of 3 multiplied by 3.
5. The method for automatic semantic segmentation of the mining area in the remote sensing image according to claim 1, characterized in that: the feature upsampling layer uses a 3 x 3 convolution kernel, as opposed to a convolution process.
6. The method for automatic semantic segmentation of the mining area in the remote sensing image according to claim 1, characterized in that: and after the characteristic is subjected to the upsampling layer, the obtained image and the pooled characteristic graph in the downsampling are superposed, and then the convolution of a Dense layer is carried out.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810770020.5A CN109145730B (en) | 2018-07-13 | 2018-07-13 | Automatic semantic segmentation method for mining area in remote sensing image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810770020.5A CN109145730B (en) | 2018-07-13 | 2018-07-13 | Automatic semantic segmentation method for mining area in remote sensing image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109145730A CN109145730A (en) | 2019-01-04 |
| CN109145730B true CN109145730B (en) | 2021-08-13 |
Family
ID=64800612
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810770020.5A Active CN109145730B (en) | 2018-07-13 | 2018-07-13 | Automatic semantic segmentation method for mining area in remote sensing image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109145730B (en) |
Families Citing this family (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109766999A (en) * | 2019-01-16 | 2019-05-17 | 宽凳（北京）科技有限公司 | A method of optimizing more element identification classifications and obscures |
| CN109886238A (en) * | 2019-03-01 | 2019-06-14 | 湖北无垠智探科技发展有限公司 | Unmanned plane Image Change Detection algorithm based on semantic segmentation |
| CN110059758B (en) * | 2019-04-24 | 2020-07-10 | 海南长光卫星信息技术有限公司 | Remote sensing image culture pond detection method based on semantic segmentation |
| CN110119744A (en) * | 2019-05-07 | 2019-08-13 | 上海交通大学 | The remote sensing image semantic segmentation method of multi-scale division constraint |
| CN110598553A (en) * | 2019-08-09 | 2019-12-20 | 中国科学院南京地理与湖泊研究所 | Original true landform mining damaged area detection method based on remote sensing image and topographic data |
| CN110619634B (en) * | 2019-09-20 | 2022-09-09 | 北京达佳互联信息技术有限公司 | Image processing method, image processing apparatus, electronic device, and medium |
| CN111050315B (en) * | 2019-11-27 | 2021-04-13 | 北京邮电大学 | Wireless transmitter identification method based on multi-core two-way network |
| CN111368843B (en) * | 2020-03-06 | 2022-06-10 | 电子科技大学 | Method for extracting lake on ice based on semantic segmentation |
| CN111833328B (en) * | 2020-07-14 | 2023-07-25 | 汪俊 | Aircraft engine blade surface defect detection method based on deep learning |
| CN112633242B (en) * | 2020-12-31 | 2023-04-14 | 长光卫星技术股份有限公司 | Port ore heap segmentation and reserve calculation method based on improved UNet network |
| CN114549534B (en) * | 2022-01-17 | 2022-11-15 | 中国矿业大学(北京) | Mining area land utilization identification method, device, equipment and medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108170751A (en) * | 2017-12-21 | 2018-06-15 | 百度在线网络技术（北京）有限公司 | For handling the method and apparatus of image |
| CN108197606A (en) * | 2018-01-31 | 2018-06-22 | 浙江大学 | The recognition methods of abnormal cell in a kind of pathological section based on multiple dimensioned expansion convolution |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| GB0422447D0 (en) * | 2004-10-08 | 2004-11-10 | Univ Cambridge Tech | Use of ionic liquids |
| US8345976B2 (en) * | 2010-08-06 | 2013-01-01 | Sony Corporation | Systems and methods for segmenting digital images |
-
2018
- 2018-07-13 CN CN201810770020.5A patent/CN109145730B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108170751A (en) * | 2017-12-21 | 2018-06-15 | 百度在线网络技术（北京）有限公司 | For handling the method and apparatus of image |
| CN108197606A (en) * | 2018-01-31 | 2018-06-22 | 浙江大学 | The recognition methods of abnormal cell in a kind of pathological section based on multiple dimensioned expansion convolution |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109145730A (en) | 2019-01-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109145730B (en) | Automatic semantic segmentation method for mining area in remote sensing image | |
| CN107563422B (en) | A kind of polarization SAR classification method based on semi-supervised convolutional neural networks | |
| CN109800736A (en) | A kind of method for extracting roads based on remote sensing image and deep learning | |
| CN106023145A (en) | Remote sensing image segmentation and identification method based on superpixel marking | |
| CN110020676A (en) | Method for text detection, system, equipment and medium based on more receptive field depth characteristics | |
| Abdollahi et al. | Improving road semantic segmentation using generative adversarial network | |
| CN106372648A (en) | Multi-feature-fusion-convolutional-neural-network-based plankton image classification method | |
| CN107564025A (en) | A kind of power equipment infrared image semantic segmentation method based on deep neural network | |
| CN114419449B (en) | Self-attention multi-scale feature fusion remote sensing image semantic segmentation method | |
| CN108961220B (en) | Image collaborative saliency detection method based on multilayer convolution feature fusion | |
| CN103593853B (en) | The multiple dimensioned object-oriented classification method of remote sensing image of expressing based on joint sparse | |
| CN112232328A (en) | Remote sensing image building area extraction method and device based on convolutional neural network | |
| CN113688836A (en) | Real-time road image semantic segmentation method and system based on deep learning | |
| CN111144418B (en) | Railway track area segmentation and extraction method | |
| CN111414954B (en) | Rock image retrieval method and system | |
| CN110059539A (en) | A kind of natural scene text position detection method based on image segmentation | |
| CN110349087A (en) | RGB-D image superior quality grid generation method based on adaptability convolution | |
| CN114359130A (en) | Road crack detection method based on unmanned aerial vehicle image | |
| CN111429510B (en) | Pollen detection method based on self-adaptive feature pyramid | |
| CN111860233A (en) | SAR image complex building extraction method and system based on attention network selection | |
| CN113256649A (en) | Remote sensing image station selection and line selection semantic segmentation method based on deep learning | |
| CN109034198A (en) | The Scene Segmentation and system restored based on characteristic pattern | |
| CN112200720A (en) | Super-resolution image reconstruction method and system based on filter fusion | |
| CN114943902A (en) | Urban vegetation unmanned aerial vehicle remote sensing classification method based on multi-scale feature perception network | |
| CN115631162A (en) | Landslide hidden danger identification method, system, medium and equipment |
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 | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20220831 Address after: Room C311, Hefei National University Science and Technology Park, No. 602 Huangshan Road, High-tech Zone, Hefei City, Anhui Province, 230000 Patentee after: Hefei deep blue space intelligent technology Co.,Ltd. Address before: 246001 No. 111, Jiulong Road, economic and Technological Development Zone, Hefei, Anhui Patentee before: ANHUI University |
|
| TR01 | Transfer of patent right |