CN109829507B - Aerial high-voltage transmission line environment detection method - Google Patents
Aerial high-voltage transmission line environment detection method Download PDFInfo
- Publication number
- CN109829507B CN109829507B CN201910129588.3A CN201910129588A CN109829507B CN 109829507 B CN109829507 B CN 109829507B CN 201910129588 A CN201910129588 A CN 201910129588A CN 109829507 B CN109829507 B CN 109829507B
- Authority
- CN
- China
- Prior art keywords
- super
- pixel block
- image
- training
- transmission line
- 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
Abstract
The invention relates to an aerial high-voltage transmission line environment detection method, which is based on super pixels and a deep neural network, and comprises the following steps: step a: making an image training set; step b: preprocessing an image training set; step c: pre-segmenting through a super-pixel segmentation algorithm, and extracting a feature vector of each super-pixel block; step d: training a DNN model according to the feature vector of the super pixel block and the true value corresponding to the super pixel block; step e: repeating the steps b-c for the image to be detected, inputting the extracted super-pixel block feature vector into a trained DNN model to obtain a classification result of the super-pixel block, and applying the classification result to the original image to obtain a final prediction result. Compared with the prior art, the invention solves the problems of large manpower and time required by the traditional detection of the environment around the high-voltage transmission line.
Description
Technical Field
The invention relates to the technical field of image segmentation, in particular to an aerial high-voltage transmission line environment detection method based on super pixels and a deep neural network.
Background
The environment around the high-voltage transmission line has an important influence on the safety of the high-voltage transmission line, and the detection of the environment based on manual operation is a time-consuming and labor-consuming project, and is easy to cause careless mistakes. Therefore, aerial photographing is performed around the high-voltage transmission line through the unmanned aerial vehicle, and the workload of the part can be reduced by detecting the aerial photographed image. Detecting image content by image segmentation is a mainstream detection method, and conventional image segmentation methods include a threshold-based segmentation method, a region-based segmentation method, an edge-based segmentation method, and a cluster-based segmentation method. However, these conventional methods do not produce satisfactory detection results when facing images of the surroundings of the high-voltage power transmission line because of the complex environment. Recently, with the development of deep learning, image segmentation based on neural networks has been greatly successful, and this technique makes it possible to segment images having a complex environment.
Generally, when image segmentation is performed through a neural network, a huge training set is needed to enable the neural network model to fully learn the characteristics of objects in the image. However, creating training sets is a very expensive project. In the absence of an image training set, full convolutional neural networks (Fully Convolutional Networks, FCN), U-networks (U-Net), etc. are all very difficult to train. And the image is pre-segmented by super pixels, one image can generate thousands of super pixel blocks, thousands of feature vectors used for training a network can be obtained from the thousands of super pixel blocks, namely thousands of feature vectors can be generated on one image, and the problem of insufficient data sets is solved.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provide an aerial high-voltage transmission line environment detection method based on a super-pixel and a deep neural network, which uses a super-pixel segmentation algorithm to pre-segment an image and trains a DNN model by extracting characteristics in a super-pixel block, so that the trained DNN is used for classifying the super-pixel block to be detected, and the purpose of image detection is achieved.
The aim of the invention can be achieved by the following technical scheme:
the aerial high-voltage transmission line environment detection method is based on super pixels and a deep neural network, and comprises the following steps of:
step a: making an image training set;
step b: preprocessing an image training set;
step c: pre-segmenting through a super-pixel segmentation algorithm, and extracting a feature vector of each super-pixel block;
step d: training a DNN model according to the feature vector of the super pixel block and the true value corresponding to the super pixel block;
step e: repeating the steps b-c for the image to be detected, inputting the extracted super-pixel block feature vector into a trained DNN model to obtain a classification result of the super-pixel block, and applying the classification result to the original image to obtain a final prediction result.
Preferably, only N images of the training set are made in the step a, wherein N is more than or equal to 5 and less than or equal to 10.
Preferably, the preprocessing in the step b is to perform an image resizing transformation on all training images.
Preferably, the extracting the super pixel block feature in the step c includes the following steps:
step c1: pre-segmenting all the preprocessed training set images through a super-pixel segmentation algorithm to obtain super-pixel pre-segmented images;
step c2: and extracting various characteristics from each super-pixel block of the pre-segmented image to form a characteristic vector.
Preferably, the super-pixel segmentation algorithm adopts a SLIC algorithm.
Preferably, the training the DNN model according to the feature vector and the corresponding true value of the super pixel block in the step d includes the following steps:
step d1: the feature vector of each super pixel block corresponds to the true value one by one to obtain a feature vector training set and a corresponding true value set;
step d2: initializing a DNN network, setting the number of DNN hidden units hidden_units= [25], detecting the number of categories n_class=3, iterating the number of rounds steps=15000, discarding the ratio dropout=0.2, and selecting cross entropy loss as a loss function by using a random gradient descent SGD as an optimization algorithm;
step d3: taking the super pixel block feature vector and the corresponding true value obtained in the step d1 as DNN input, training a DNN model, wherein the training process is an iteration process, minimizing cross entropy loss in each iteration and reversely updating DNN parameters.
Preferably, the step e of obtaining the detection result of the image to be detected includes the following steps:
step e1: the detection stage, repeating the steps b-c for a test image to be segmented to obtain the characteristic vector of each super-pixel block of the whole image, and taking the characteristic vector of each super-pixel block as the input of DNN obtained by training in the step d to obtain the classification result of the super-pixel block;
step e2: and (3) corresponding the classification result of each super pixel block to the region where the super pixel block is located in the original image to obtain a final detection result.
Compared with the prior art, the invention has the beneficial effects that: the invention uses the characteristic of the super pixel block to train, which greatly relieves the requirement of making huge image training set; meanwhile, the advantage of DNN on feature classification is utilized, and a good segmentation result is maintained.
Drawings
Fig. 1 is a flowchart of a training process of the aerial high-voltage transmission line environment detection method of the invention.
Fig. 2 is a flow chart of a detection process of the environment detection method for the aerial high-voltage transmission line.
Fig. 3 is an image of the environment surrounding the original aerial high voltage transmission line in the embodiment.
Fig. 4 is a truth image corresponding to fig. 3 in an embodiment.
FIG. 5 is a pre-segmented image corresponding to FIG. 3 after super-pixel segmentation in an embodiment.
Fig. 6 is a schematic diagram of construction of feature vectors from a super-pixel block.
Fig. 7 is a diagram showing the actual effect of division using the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are some, but not all embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the present invention without making any inventive effort, shall fall within the scope of the present invention.
According to the invention, the unmanned aerial vehicle aerial photographing is used for acquiring the surrounding environment image of the high-voltage transmission line, the super-pixel is utilized for dividing the image and extracting the characteristic vector of each super-pixel block after division, the number requirement on an image training set is obviously reduced, meanwhile, the super-pixel blocks are classified through the characteristic vector training DNN to obtain a prediction result, good precision and detection result are obtained, and the problems of a large amount of manpower and time required by the traditional detection of the surrounding environment of the high-voltage transmission line are solved.
Some mathematical symbols mainly involved in the present invention are explained as follows:
training the number of samples N;
the number M of categories to be detected;
training sample set S trian ＝{I i ,i＝1,2,…,N},I i To the original image of the ith sample in the training set, H i And W is i The height and width of the ith training sample are respectively;
the image of the ith sample in the training set after being resizedH′ i And W' i The height and width of the ith training sample after the size adjustment are respectively;
truth image set gt= { GT i ,i＝1,2,…,N},GT i Corresponds to the image I' i Is a true value image of (2);
label matrix L obtained by SLIC super-pixel pre-segmentation of ith sample in training set i ；
Pixel set of the jth super pixel block constituting the ith sample in the training set
Number of ith sample superpixels in training set
Feature vector of jth super pixel block of ith sample in training set
Truth value of jth superpixel block of ith sample in training set
Feature vector training set X train ；
True value training set Y train ；
Class c= { C to be detected i ,i＝1,2,…,M}，c i Is the i-th category.
Referring to fig. 1, the main flow of the present invention is divided into a training phase and a detection phase. In the embodiment, taking the high-voltage transmission line environment image obtained by aerial photography of the unmanned aerial vehicle as an example, the image is required to be divided into green land, greenhouse and other 3 types altogether, namely M=3, C= { green land, greenhouse and other }. Each class uses a different color representation in the image to visualize the results of the classification. The aerial high-voltage transmission line environment detection method based on the super-pixel and the deep neural network comprises the following specific steps:
step a: and (5) manufacturing a training set. In the present embodimentIn the example, N images respectively provided with greenhouses, greenhouses and other (such as buildings, rivers and the like) are selected from all collected surrounding environment images of the aerial high-voltage transmission line to serve as a training sample set S trian Coloring the green land, greenhouse and other three areas of the images in the training sample set into different colors (such as green, white and gray) respectively to obtain a true value image GT corresponding to the images in the training sample set i . In this embodiment, n=5 is taken, fig. 3 is one of the original aerial images, and fig. 4 is a true image corresponding to fig. 3.
Step b: the training set is preprocessed. Because the resolution ratio of the aerial image is high, the direct processing of the original image is very time-consuming, so the size adjustment transformation is carried out on the images in the training sample set in the step a:
I′ i ＝RESIZE(I i ,(H′ i ,W′ i )) (1)
H′ i ＝H i /k,W′ i ＝W i /k
wherein RESIZE (·) is the transform function of the image size adjustment, k is the scaling factor, and the transformed image is recorded as I' i And =1, 2, …, N. In this embodiment, k=8 is taken.
Step c: pre-segmentation is performed by a super-pixel segmentation algorithm and feature vectors of each super-pixel block are extracted. Super-pixels are small areas composed of pixel points that are adjacent in position and similar in color, brightness, texture, etc. These small areas retain the effective information and boundary information for further image segmentation. The invention uses the super-pixel segmentation algorithm of simple linear iterative clustering (Simple linear iterative clustering, SLIC) to segment the image I' i Pre-segmentation is performed. The SLIC algorithm searches for a number of pixels nearest to each seed point by generating K seed points in the image, classifying them as one type with the center seed point, then recalculates the seed center, iterating the above steps until convergence. The SLIC super-pixel segmentation is described as shown in equation (2):
wherein, the liquid crystal display device comprises a liquid crystal display device,is a label matrix pre-segmented by SLIC algorithm,
parameter n_segment i For the initialized number of superpixel blocks, n_segments are taken in this embodiment i ＝H′ i ×W′ i N_slec_pixels, n_slec_pixels=400. That is, each super-pixel block is guaranteed to have a similar size initially, so that the problem of the dimension of the super-pixel block under images of different sizes is solved. The parameter compatibility controls the compactness of each super-pixel block, and the larger the value, the more regular the shape of the super-pixel, i.e. the closer the shape of the super-pixel is to a rectangle, and conversely, the more fine the edge. In this example, compare=10 is taken. Fig. 5 shows the result after SLIC pre-segmentation.
And extracting the characteristic combination in each super pixel block to form a characteristic vector after the super pixel pre-segmentation result is obtained. In this embodiment, color, texture, and gradient features within the super pixel block are extracted.
Color is the most prominent feature in a color image, in this example, the color feature of each super pixel blockAs shown in formula (3):
wherein the method comprises the steps ofIs I' i Color characteristics of j-th super pixel block pre-segmented by super pixel algorithm, I' i (x, y) is image I' i The color values at the (x, y) coordinates, MEAN (·) are averaging functions. The average color corresponding to each super pixel block can be obtained by the formula (3), and the value is taken as the color characteristic.
Texture is a visual feature reflecting a homogeneous phenomenon in an image, and represents a surface structure organization arrangement attribute of the surface of an object, wherein the surface structure organization arrangement attribute has slow change or periodical change. The texture is formed by repeatedly appearing gray scale distribution at space positions, so that a certain gray scale relation exists between two pixels which are separated by a certain distance in the image space, namely the space correlation characteristic of gray scales in the image. Gray-Level Co-occurrence Matrix, GLCM is a common method for describing textures by studying the spatial correlation properties of Gray. In this embodiment, the gray level co-occurrence matrix of each super pixel block is obtained by calculation, and the feature value describing the texture of the super pixel block is obtained by using the gray level co-occurrence matrix. Record I' i Gray level co-occurrence matrix of jth super pixel block pre-segmented by super pixel algorithmThe generation is as shown in formula (4):
Gray i ＝GRAY(I′ i )
wherein GRAY (·) is a function of converting color image into GRAY image, gray i Is a color image I' i Corresponding gray images, angles and distances are parameters controlling the direction and distance of the gray co-occurrence matrix, and levels control the number of gray levels of the generated gray co-occurrence matrix. In this case the number of the elements to be processed is,distances＝[3]level=64, that is, for each super-pixel block, the correlation characteristics of the gray scale are calculated every 3 pixels in four directions, and the resulting gray scale co-occurrence matrix corresponding to each super-pixel block is a matrix of 4 forms of 64×64.
After the gray level co-occurrence matrix is obtained, the descriptive gray level co-occurrence matrix is obtained according to the formula (5)Numerical value of characteristic
Described are the contrast, difference and angular second moment of the gray co-occurrence matrix of the super pixel block, respectively, will +.>As a texture feature for each super pixel block.
For the greenhouse and the green land, the gradients are obviously different, so that the gradient characteristics added with the super pixel blocks can be better classifiedGenerated by formula (6):
wherein Gray i (x, y) is a Gray image Gray i Gray value at coordinates (x, y), gradient i Is I' i Gradient image of (c) i (x, y) is the gradient value of the gradient image at coordinates (x, y).
Step d: integrating the feature values obtained in the step c into feature vectorsBy pre-segmentation of the label image L i And corresponding truth image GT i Performing the calculation of formula (7) to obtain the corresponding feature vector +.>True value of->
Where mode (·) represents the mode of computing the element within the current super pixel block region. In the present embodiment of the present invention, in the present embodiment,the values are {0,1,2}, which respectively represent greenhouses, greenhouses and other three categories. Combine all +.>And->Obtaining a feature vector training set X train And true value training set Y train As shown in formula (8):
x is to be train ，Y train The DNN is trained as input. The basic DNN network structure used in the present invention is defined as follows: DNN accepts current feature vectorAs input, a signal with hidden_units= [25]The result vector with length n_class=3 is output after the hidden layer of (c). Other parameters set are iterative rounds steps=15000, discard ratio dropout=0.2, and select cross entropy loss as loss function using random gradient descent (SGD) as optimization algorithm. The training process of DNN is an iterative process: each iteration passing forwardThe cross entropy loss is calculated, the loss is minimized, the updated model parameters are back propagated until the iteration round number steps is reached, and finally the DNN model for the detection task is obtained.
Step e: repeating the steps b-c for the image to be detected, inputting the extracted super-pixel block feature vector into the DNN model obtained in the step d to obtain a classification result of the super-pixel block, and corresponding the classification result to the original image to obtain a final prediction result. The method comprises the following steps:
(e1) Repeating the steps b-c to obtain a feature vector set of the image to be detected, and taking the feature vector set as input of the DNN model obtained by training in the step d to obtain a classification result of the super pixel block;
(e2) And (3) corresponding the classification result of each super pixel block to the region where the super pixel block is located in the original image to obtain a detection result.
The detected image after the processing of the present invention is shown in fig. 7. The advantage of the present invention is that a better detection result can be obtained with a small number of data sets, and in this embodiment, the detection is best for greenbelts, which is mainly due to the obvious color characteristics of greenbelts, as shown in fig. 7 (a). The detection of the greenhouse is slightly inferior to that of the green land, as shown in (c) of fig. 7, because the characteristics of the greenhouse are similar to those of a building, a cement road, etc., and are easily confused, and although the characteristics of the gray level co-occurrence matrix distinguish the greenhouse to some extent, the misclassification situation exists. This problem can be solved by introducing more distinguishing features.
While the invention has been described with reference to certain preferred embodiments, it will be understood by those skilled in the art that various changes and substitutions of equivalents may be made and equivalents will be apparent to those skilled in the art without departing from the scope of the invention. Therefore, the protection scope of the invention is subject to the protection scope of the claims.
Claims (6)
1. The aerial high-voltage transmission line environment detection method is based on super pixels and a deep neural network and is characterized by comprising the following steps of:
step a: making an image training set;
step b: preprocessing an image training set;
step c: pre-segmenting through a super-pixel segmentation algorithm, and extracting a feature vector of each super-pixel block;
step d: training a DNN model according to the feature vector of the super pixel block and the true value corresponding to the super pixel block;
step e: repeating the steps b-c for the image to be detected, inputting the extracted super-pixel block feature vector into a trained DNN model to obtain a classification result of the super-pixel block, and applying the classification result to the original image to obtain a final prediction result;
in the step d, training the DNN model according to the feature vector and the corresponding true value of the super pixel block comprises the following steps:
step d1: the feature vector of each super pixel block corresponds to the true value one by one to obtain a feature vector training set and a corresponding true value set;
step d2: initializing a DNN network, setting the number of DNN hidden units hidden_units= [25], detecting the number of categories n_class=3, iterating the number of rounds steps=15000, discarding the ratio dropout=0.2, and selecting cross entropy loss as a loss function by using a random gradient descent SGD as an optimization algorithm;
step d3: taking the super pixel block feature vector and the corresponding true value obtained in the step d1 as DNN input, training a DNN model, wherein the training process is an iteration process, minimizing cross entropy loss in each iteration and reversely updating DNN parameters.
2. The method for detecting the environment of the aerial high-voltage transmission line according to claim 1, wherein in the step a, only N training set images are needed to be manufactured, wherein N is more than or equal to 5 and less than or equal to 10.
3. The method for detecting the environment of the aerial high-voltage transmission line according to claim 1, wherein the preprocessing in the step b is to perform image resizing transformation on all training images.
4. The method for detecting the environment of the aerial high-voltage transmission line according to claim 1, wherein the step c of extracting the super-pixel block features comprises the following steps:
step c1: pre-segmenting all the preprocessed training set images through a super-pixel segmentation algorithm to obtain super-pixel pre-segmented images;
step c2: and extracting various characteristics from each super-pixel block of the pre-segmented image to form a characteristic vector.
5. The aerial high-voltage transmission line environment detection method according to claim 1 or 4, wherein the super-pixel segmentation algorithm adopts a SLIC algorithm.
6. The method for detecting the environment of the aerial high-voltage transmission line according to claim 1, wherein the step e of obtaining the detection result of the image to be detected comprises the following steps:
step e1: the detection stage, repeating the steps b-c for a test image to be segmented to obtain the characteristic vector of each super-pixel block of the whole image, and taking the characteristic vector of each super-pixel block as the input of DNN obtained by training in the step d to obtain the classification result of the super-pixel block;
step e2: and (3) corresponding the classification result of each super pixel block to the region where the super pixel block is located in the original image to obtain a final detection result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910129588.3A CN109829507B (en) | 2019-02-21 | 2019-02-21 | Aerial high-voltage transmission line environment detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910129588.3A CN109829507B (en) | 2019-02-21 | 2019-02-21 | Aerial high-voltage transmission line environment detection method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109829507A CN109829507A (en) | 2019-05-31 |
| CN109829507B true CN109829507B (en) | 2023-09-19 |
Family
ID=66864077
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910129588.3A Active CN109829507B (en) | 2019-02-21 | 2019-02-21 | Aerial high-voltage transmission line environment detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109829507B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112179922A (en) * | 2020-09-24 | 2021-01-05 | 安徽德尔电气集团有限公司 | Wire and cable defect detection system |
| CN112241692B (en) * | 2020-09-25 | 2022-09-13 | 天津大学 | Channel foreign matter intelligent detection and classification method based on aerial image super-pixel texture |
| CN112200246A (en) * | 2020-10-09 | 2021-01-08 | 河北工业大学 | Training method of SVM classifier and petrochemical storage tank corrosion defect segmentation method |
| CN113780259B (en) * | 2021-11-15 | 2022-03-15 | 中移(上海)信息通信科技有限公司 | Road surface defect detection method and device, electronic equipment and readable storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106875395A (en) * | 2017-01-12 | 2017-06-20 | 西安电子科技大学 | Super-pixel level SAR image change detection based on deep neural network |
| CN107545577A (en) * | 2017-08-23 | 2018-01-05 | 电子科技大学 | Sedimentary facies image partition method based on neutral net |
| WO2019001208A1 (en) * | 2017-06-28 | 2019-01-03 | 苏州比格威医疗科技有限公司 | Segmentation algorithm for choroidal neovascularization in oct image |
| CN109325484A (en) * | 2018-07-30 | 2019-02-12 | 北京信息科技大学 | Flowers image classification method based on background priori conspicuousness |
-
2019
- 2019-02-21 CN CN201910129588.3A patent/CN109829507B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106875395A (en) * | 2017-01-12 | 2017-06-20 | 西安电子科技大学 | Super-pixel level SAR image change detection based on deep neural network |
| WO2019001208A1 (en) * | 2017-06-28 | 2019-01-03 | 苏州比格威医疗科技有限公司 | Segmentation algorithm for choroidal neovascularization in oct image |
| CN107545577A (en) * | 2017-08-23 | 2018-01-05 | 电子科技大学 | Sedimentary facies image partition method based on neutral net |
| CN109325484A (en) * | 2018-07-30 | 2019-02-12 | 北京信息科技大学 | Flowers image classification method based on background priori conspicuousness |
Non-Patent Citations (2)
| Title |
|---|
| 基于深度卷积神经网络的PolSAR图像变化检测方法;王剑等;《系统工程与电子技术》;20180515(第07期);第46-53页 * |
| 无人装备野外场景自适应道路识别技术;华夏等;《兵器装备工程学报》;20180625(第06期);第171-第176页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109829507A (en) | 2019-05-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109829507B (en) | Aerial high-voltage transmission line environment detection method | |
| CN108596055B (en) | Airport target detection method of high-resolution remote sensing image under complex background | |
| CN109033998B (en) | Remote sensing image ground object labeling method based on attention mechanism convolutional neural network | |
| CN105139395B (en) | SAR image segmentation method based on small echo pond convolutional neural networks | |
| CN106909902B (en) | Remote sensing target detection method based on improved hierarchical significant model | |
| CN107194336B (en) | Polarized SAR image classification method based on semi-supervised depth distance measurement network | |
| CN110263717B (en) | Method for determining land utilization category of street view image | |
| CN111161218A (en) | High-resolution remote sensing image change detection method based on twin convolutional neural network | |
| CN106920243A (en) | The ceramic material part method for sequence image segmentation of improved full convolutional neural networks | |
| CN109447160B (en) | Method for automatically matching image and vector road intersection | |
| CN107067405B (en) | Remote sensing image segmentation method based on scale optimization | |
| CN110598564B (en) | OpenStreetMap-based high-spatial-resolution remote sensing image transfer learning classification method | |
| CN111414954B (en) | Rock image retrieval method and system | |
| CN110751644B (en) | Road surface crack detection method | |
| CN111160127A (en) | Remote sensing image processing and detecting method based on deep convolutional neural network model | |
| CN108427919B (en) | Unsupervised oil tank target detection method based on shape-guided saliency model | |
| CN107403434A (en) | SAR image semantic segmentation method based on two-phase analyzing method | |
| CN111241994A (en) | Method for extracting remote sensing image rural highway desertification road section for deep learning | |
| CN111652240A (en) | Image local feature detection and description method based on CNN | |
| CN111563408B (en) | High-resolution image landslide automatic detection method with multi-level perception characteristics and progressive self-learning | |
| CN111582004A (en) | Target area segmentation method and device in ground image | |
| CN114241326A (en) | Progressive intelligent production method and system for ground feature elements of remote sensing images | |
| CN111104850A (en) | Remote sensing image building automatic extraction method and system based on residual error network | |
| CN110909623A (en) | Three-dimensional target detection method and three-dimensional target detector | |
| CN111079807B (en) | Ground object classification method and device |
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 |