CN112967296A - Point cloud dynamic region graph convolution method, classification method and segmentation method - Google Patents
Point cloud dynamic region graph convolution method, classification method and segmentation method Download PDFInfo
- Publication number
- CN112967296A CN112967296A CN202110261653.5A CN202110261653A CN112967296A CN 112967296 A CN112967296 A CN 112967296A CN 202110261653 A CN202110261653 A CN 202110261653A CN 112967296 A CN112967296 A CN 112967296A
- Authority
- CN
- China
- Prior art keywords
- point cloud
- information
- convolution
- map
- convolution operation
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10028—Range image; Depth image; 3D point clouds
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20212—Image combination
- G06T2207/20221—Image fusion; Image merging
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a point cloud dynamic area map convolution method, a point cloud dynamic area map classification method and a point cloud dynamic area map segmentation method using the point cloud dynamic area map convolution method. The invention adopts a new convolution operation form aiming at point cloud, and aggregates point characteristic information of a plurality of different neighborhoods by a nonlinear method according to a constructed point cloud picture structure, so that the neuron can select the area size in a self-adaptive manner. Compared with the prior technical scheme of analyzing on a single point, such as PointNet, the method constructs a plurality of different local neighborhood map structures, enables each neuron to adaptively select the proper neighborhood receptive field size, then performs similar convolution operation by utilizing the connection between each point and the neighborhood points to obtain local characteristics, can better combine surrounding neighborhood information, more effectively extract local geometric information, and finally improves the accuracy of classifying or dividing the point cloud data.
Description
Technical Field
The invention belongs to the technical field of computer vision, and particularly relates to a point cloud dynamic area graph convolution method, a classification method and a segmentation method.
Background
The point cloud data contains abundant semantic information and has the characteristics of high density, high precision and the like, but due to the irregularity and disorder of the point cloud data, semantic analysis based on the point cloud data is still a difficult challenge. Some of the earlier methods use features with complex rules for manual extraction to solve such problems. With the recent increase of fire heat of deep learning and machine learning technologies, methods of deep learning are also introduced for analysis and processing of point cloud data. The data to be processed by the deep network is in a regular shape, the point cloud data is basically irregular, and the spatial distribution of the point cloud data is not influenced by the arrangement mode of the point clouds, so that the common method for processing the point cloud data by using the deep learning model is to convert the original point cloud data into data structure forms such as grids, voxels, trees and the like. Some advanced deep learning networks such as PointNet and PointNet + + are designed specifically to deal with irregularities in the point cloud, and can directly process the original point cloud data without converting the point cloud data into a regular shape and then processing the point cloud data. However, neither PointNet nor PointNet + + support convolution operation and cannot effectively extract local geometric information.
Much work is currently focused on processing point cloud data using convolution operations. The 2DCNN is directly expanded to the 3D field, the 3D space is regarded as a volume grid, and the operation is carried out by using 3D convolution. Although 3D convolution works well on the task of point cloud classification and segmentation, their high requirements for storage performance and the high computational cost required make them still suffer from insufficient accuracy on large scale data sets and large scenes.
In summary, how to improve the accuracy of classifying or segmenting point cloud data becomes a problem that needs to be solved urgently by those skilled in the art.
Disclosure of Invention
Aiming at the defects in the prior art, the invention actually solves the problems that: the accuracy of classification or segmentation of point cloud data is improved.
In order to solve the technical problems, the invention adopts the following technical scheme:
a point cloud dynamic region graph convolution method comprises the following steps:
s1, acquiring three-dimensional point cloud data X, X ═ alpha1,α2,α3,…,αi,…,αn}，αiData representing the ith point, n representing the number of points in the three-dimensional point cloud data, alphai＝{xi,yi,zi}，xi、yiAnd ziDenotes alphaiThree-dimensional coordinates of (a);
s2, performing two independent k nearest neighbor operations on the three-dimensional point cloud data X to obtain two local feature maps y and z, wherein k values of the two independent k nearest neighbor operations are different;
s3, fusing the two local feature maps y and z to obtain fused information T, where T is Sum (y, z);
s4, pooling the fusion information T to obtain the characteristic communication information S1，s1＝MAX(T)；
S5, using full connection layer to feature communication information S1Performing compact dimensionality reduction to obtain compact features s2，s2＝FC(s1)；
S6, adaptively selecting branch dimension information of different areas from the compact features by using an attention mechanism, and normalizing the weights by using softmax to obtain normalized information a1And a2，
FC1() And FC2() Represent fully connected layer operations corresponding to y and z, respectively;
s7, multiplying the normalized information by the local feature map, and summing to obtain a feature map U, wherein U is Sum (a)1*y,a2*z)。
Preferably, the k values of the two independent k-nearest neighbor operations in step S2 are 15 and 25, respectively.
A point cloud dynamic regional image classification method adopts the point cloud dynamic regional image convolution method to carry out convolution operation, and the feature spectrum U is used as a feature obtained by each convolution operation.
A point cloud dynamic regional image segmentation method adopts the point cloud dynamic regional image convolution method to carry out convolution operation, and the feature spectrum U is used as a feature obtained by each convolution operation.
In summary, compared with the prior art, the invention has the following technical effects:
the invention adopts a new convolution operation form aiming at point cloud, and aggregates point characteristic information of a plurality of different neighborhoods by a nonlinear method according to a constructed point cloud picture structure, so that the neuron can select the area size in a self-adaptive manner. Compared with the prior technical scheme of analyzing on a single point, such as PointNet, the method constructs a plurality of different local neighborhood map structures, enables each neuron to adaptively select the proper neighborhood receptive field size, then performs similar convolution operation by utilizing the connection between each point and the neighborhood points to obtain local characteristics, can better combine surrounding neighborhood information, more effectively extract local geometric information, and finally improves the accuracy of classifying or dividing the point cloud data.
Drawings
FIG. 1 is a flow chart of a method for convolving a point cloud dynamic region map according to the present invention;
fig. 2 is a k-neighbor map of a local point cloud space.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings.
As shown in fig. 1, the invention discloses a method for convolving a point cloud dynamic region map, comprising the following steps:
s1, acquiring three-dimensional point cloud data X, X ═ alpha1,α2,α3,…,αi,…,αn}，αiData representing the ith point, n representing the number of points in the three-dimensional point cloud data, alphai＝{xi,yi,zi}，xi、yiAnd ziDenotes alphaiThree-dimensional coordinates of (a);
s2, performing two independent k nearest neighbor operations on the three-dimensional point cloud data X to obtain two local feature maps y and z, wherein k values of the two independent k nearest neighbor operations are different;
as shown in fig. 2, alpha is defined for k-neighbor map of local point cloud spacej1,αj2,…,αjkIs alphaiK neighbor points of eijFor edge features, defined as eij＝hθ(αi,αj) Where θ is a training parameter, a non-linear function hθ(αi,αj)：RC×RC→RC，RCIs a feature after polymerization. The output of the ith point in the graph convolution can be expressed as:
similar to the convolution operation in 2D images, will be alphaiCentral pixel, α, seen as a convolution regionjIs then alphaiThe surrounding blocks of pixels. Alpha in the graph structureiAnd alphajFormed directed edge eijThe designed edge function is defined as: h isθ(αi,αj)＝hθ(αi,αi-αj). Such a structure combines both global shape information and local neighborhood information and is implemented by MLP. And the aggregation function selects a max function to perform aggregation operation.
S3, fusing the two local feature maps y and z to obtain fused information T, where T is Sum (y, z);
s4, pooling the fusion information T to obtain the characteristic communication information S1，s1＝MAX(T)；
S5, using full connection layer to feature communication information S1Performing compact dimensionality reduction to obtain compact features s2，s2＝FC(s1)；
Steps S3 to S5 are to carry out the integrated encoding of the information from multiple branches and to transmit the information to the next step, so as to realize the adaptive adjustment of the size of k neighborhood by the neuron. Finally, the full-connection network is used for carrying out compact dimension reduction on the features, so that not only can the region be accurately and adaptively selected, but also the size can be reduced, and the operation efficiency can be improved.
S6, adaptively selecting branch dimension information of different areas from the compact features by using an attention mechanism, and normalizing the weights by using softmax to obtain normalized information a1And a2，
FC1() And FC2() Represent fully connected layer operations corresponding to y and z, respectively;
s7, multiplying the normalized information by the local feature map, and summing to obtain a feature map U, wherein U is Sum (a)1*y,a2*z)。
a1∈1×C′，a2Belongs to 1 multiplied by C ', y belongs to n multiplied by C ', z belongs to n multiplied by C '; c' represents the number of characteristic channels.
In specific implementation, the k values of two independent k-nearest neighbor operations in step S2 are 15 and 25, respectively.
The average class accuracy and the total accuracy when k-neighbor operations of different numbers and k-values are used are shown in table 2, and therefore, in the present invention, the number of k-neighbor operations is preferably 2, and the k-values are 15 and 25, respectively.
TABLE 2
The invention adopts a new convolution operation form aiming at point cloud, and aggregates point characteristic information of a plurality of different neighborhoods by a nonlinear method according to a constructed point cloud picture structure, so that the neuron can select the area size in a self-adaptive manner. Compared with the prior technical scheme of analyzing on a single point, such as PointNet, the method constructs a plurality of different local neighborhood map structures, enables each neuron to adaptively select the proper neighborhood receptive field size, then performs similar convolution operation by utilizing the connection between each point and the neighborhood points to obtain local characteristics, can better combine surrounding neighborhood information, more effectively extract local geometric information, and finally improves the accuracy of classifying or dividing the point cloud data.
The invention also discloses a point cloud dynamic regional image classification method, which adopts the point cloud dynamic regional image convolution method to carry out convolution operation, and takes the characteristic map U as the characteristic obtained by each convolution operation.
In order to verify the effect of the point cloud dynamic region map classification method disclosed by the invention, the classification task is evaluated on a ModelNet40 data set. The dataset contains 12311 mesh CAD models from 40 classes, of which 9843 models were used for training and 2468 models were used for testing. The invention follows the experimental setting of DGCNN and other models, and for each model, 1024 points are uniformly sampled from a grid surface, and only the three-dimensional coordinates of the sampling points are used as the input data of the network.
Four DRG modules are used to extract local geometric features, and the features calculated by each DRG module are used for recalculation by the next module. For the DRG module, two different k neighborhood branches are taken here as 15 and 25, respectively. And then connecting the features obtained by each DRG module to obtain a 512-dimensional feature point cloud with 64+64+128+ 256. Global features are then obtained using global max pooling and average max pooling, respectively. Finally, two fully connected layers (512, 256) are used for feature classification.
All layers contain LeakyReLU and batch regularization. The experiment also compares the number of different k neighborhoods, selects the optimal number of k neighborhoods, and evaluates the model on the test data set. An SGD optimizer with a learning rate of 0.1 was used and the learning rate was attenuated to 0.001. The training data was selected as batch number 24 and the test data was selected as 16. The results of the experiment are shown in table 1.
TABLE 1
The invention also discloses a point cloud dynamic regional image segmentation method, which is used for convolution operation by adopting the point cloud dynamic regional image convolution method, and the characteristic map U is used as the characteristic obtained by each convolution operation.
In order to verify the effect of the point cloud dynamic region graph segmentation method disclosed by the invention, a partial segmentation task is performed on a ShapeNet data set. The task classifies each point in the point cloud into several part category labels of the object. The dataset contains 16881 3D shapes from 16 object classes, for a total of 50 parts, 2048 points were sampled in each training sample, again following the experimental protocol of DGCNN et al. The outputs of the three layers of DRGConv modules are connected, spliced into 2048 point features, and then feature transformed by MLP (256, 256, 128). The selection of batch number, activation function, learning rate, etc. is the same as the classification network.
The IOU of a shape is calculated by averaging the IOUs of different parts appearing in the shape using the same evaluation method as PointNet, and the IOU of the class is obtained by averaging the IOUs of all shapes belonging to the class. Finally, the average IOU (mIOU) is calculated by averaging the IOUs of all test shapes. By comparison with PointNet, PointNet + +, PointCNN, DGCNN, Kd-Net. The results of the experiment are shown in table 3.
TABLE 3
The above is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, several changes and modifications can be made without departing from the technical solution, and the technical solution of the changes and modifications should be considered as falling within the scope of the claims of the present application.
Claims (4)
1. A method for convolving a point cloud dynamic region map is characterized by comprising the following steps:
s1, acquiring three-dimensional point cloud data X, X ═ alpha1,α2,α3,…,αi,…,αn}，αiData representing the ith point, n representing the number of points in the three-dimensional point cloud data, alphai＝{xi,yi,zi}，xi、yiAnd ziDenotes alphaiThree-dimensional coordinates of (a);
s2, performing two independent k nearest neighbor operations on the three-dimensional point cloud data X to obtain two local feature maps y and z, wherein k values of the two independent k nearest neighbor operations are different;
s3, fusing the two local feature maps y and z to obtain fused information T, where T is Sum (y, z);
s4, pooling the fusion information T to obtain the characteristic communication information S1，s1＝MAX(T)；
S5, using full connection layer to feature communication information S1Performing compact dimensionality reduction to obtain compact features s2，s2＝FC(s1)；
S6, adaptively selecting branch dimension information of different areas from the compact features by using an attention mechanism, and normalizing the weights by using softmax to obtain normalized information a1And a2，
FC1() And FC2() Represent fully connected layer operations corresponding to y and z, respectively;
s7, multiplying the normalized information by the local feature map, and summing to obtain a feature map U, wherein U is Sum (a)1*y,a2*z)。
2. The point cloud dynamic region map convolution method of claim 1, wherein k values of two independent k neighbor operations in step S2 are 15 and 25, respectively.
3. A point cloud dynamic regional image classification method is characterized in that the point cloud dynamic regional image convolution method of claim 1 or 2 is adopted to carry out convolution operation, and the feature map U is used as a feature obtained by each convolution operation.
4. A point cloud dynamic regional image segmentation method is characterized in that the point cloud dynamic regional image convolution method of claim 1 or 2 is adopted to carry out convolution operation, and the feature map U is used as a feature obtained by each convolution operation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110261653.5A CN112967296B (en) | 2021-03-10 | 2021-03-10 | Point cloud dynamic region graph convolution method, classification method and segmentation method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110261653.5A CN112967296B (en) | 2021-03-10 | 2021-03-10 | Point cloud dynamic region graph convolution method, classification method and segmentation method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112967296A true CN112967296A (en) | 2021-06-15 |
| CN112967296B CN112967296B (en) | 2022-11-15 |
Family
ID=76277614
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110261653.5A Active CN112967296B (en) | 2021-03-10 | 2021-03-10 | Point cloud dynamic region graph convolution method, classification method and segmentation method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112967296B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113516663A (en) * | 2021-06-30 | 2021-10-19 | 同济大学 | Point cloud semantic segmentation method and device, electronic equipment and storage medium |
| CN113628217A (en) * | 2021-08-12 | 2021-11-09 | 江南大学 | Three-dimensional point cloud segmentation method based on image convolution and integrating direction and distance |
Citations (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150006117A1 (en) * | 2013-07-01 | 2015-01-01 | Here Global B.V. | Learning Synthetic Models for Roof Style Classification Using Point Clouds |
| CN106682233A (en) * | 2017-01-16 | 2017-05-17 | 华侨大学 | Method for Hash image retrieval based on deep learning and local feature fusion |
| CN107451616A (en) * | 2017-08-01 | 2017-12-08 | 西安电子科技大学 | Multi-spectral remote sensing image terrain classification method based on the semi-supervised transfer learning of depth |
| CN108198244A (en) * | 2017-12-20 | 2018-06-22 | 中国农业大学 | A kind of Apple Leaves point cloud compressing method and device |
| CN109035329A (en) * | 2018-08-03 | 2018-12-18 | 厦门大学 | Camera Attitude estimation optimization method based on depth characteristic |
| CN109063753A (en) * | 2018-07-18 | 2018-12-21 | 北方民族大学 | A kind of three-dimensional point cloud model classification method based on convolutional neural networks |
| CN109410321A (en) * | 2018-10-17 | 2019-03-01 | 大连理工大学 | Three-dimensional rebuilding method based on convolutional neural networks |
| CN110081890A (en) * | 2019-05-24 | 2019-08-02 | 长安大学 | A kind of dynamic K arest neighbors map-matching method of combination depth network |
| CN110188802A (en) * | 2019-05-13 | 2019-08-30 | 南京邮电大学 | SSD algorithm of target detection based on the fusion of multilayer feature figure |
| CN110197223A (en) * | 2019-05-29 | 2019-09-03 | 北方民族大学 | Point cloud data classification method based on deep learning |
| CN110443842A (en) * | 2019-07-24 | 2019-11-12 | 大连理工大学 | Depth map prediction technique based on visual angle fusion |
| CN111027559A (en) * | 2019-10-31 | 2020-04-17 | 湖南大学 | Point cloud semantic segmentation method based on expansion point convolution space pyramid pooling |
| CN111242208A (en) * | 2020-01-08 | 2020-06-05 | 深圳大学 | Point cloud classification method, point cloud segmentation method and related equipment |
| CN111476226A (en) * | 2020-02-29 | 2020-07-31 | 新华三大数据技术有限公司 | Text positioning method and device and model training method |
| CN111583263A (en) * | 2020-04-30 | 2020-08-25 | 北京工业大学 | Point cloud segmentation method based on joint dynamic graph convolution |
| CN111666836A (en) * | 2020-05-22 | 2020-09-15 | 北京工业大学 | High-resolution remote sensing image target detection method of M-F-Y type lightweight convolutional neural network |
| CN111723814A (en) * | 2020-06-05 | 2020-09-29 | 中国科学院自动化研究所 | Cross-image association based weak supervision image semantic segmentation method, system and device |
| CN111753698A (en) * | 2020-06-17 | 2020-10-09 | 东南大学 | Multi-mode three-dimensional point cloud segmentation system and method |
| CN111915619A (en) * | 2020-06-05 | 2020-11-10 | 华南理工大学 | Full convolution network semantic segmentation method for dual-feature extraction and fusion |
| CN112036447A (en) * | 2020-08-11 | 2020-12-04 | 复旦大学 | Zero-sample target detection system and learnable semantic and fixed semantic fusion method |
| CN112149725A (en) * | 2020-09-18 | 2020-12-29 | 南京信息工程大学 | Spectral domain graph convolution 3D point cloud classification method based on Fourier transform |
| CN112184548A (en) * | 2020-09-07 | 2021-01-05 | 中国科学院深圳先进技术研究院 | Image super-resolution method, device, equipment and storage medium |
| CN112329771A (en) * | 2020-11-02 | 2021-02-05 | 元准智能科技(苏州)有限公司 | Building material sample identification method based on deep learning |
-
2021
- 2021-03-10 CN CN202110261653.5A patent/CN112967296B/en active Active
Patent Citations (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150006117A1 (en) * | 2013-07-01 | 2015-01-01 | Here Global B.V. | Learning Synthetic Models for Roof Style Classification Using Point Clouds |
| CN106682233A (en) * | 2017-01-16 | 2017-05-17 | 华侨大学 | Method for Hash image retrieval based on deep learning and local feature fusion |
| CN107451616A (en) * | 2017-08-01 | 2017-12-08 | 西安电子科技大学 | Multi-spectral remote sensing image terrain classification method based on the semi-supervised transfer learning of depth |
| CN108198244A (en) * | 2017-12-20 | 2018-06-22 | 中国农业大学 | A kind of Apple Leaves point cloud compressing method and device |
| CN109063753A (en) * | 2018-07-18 | 2018-12-21 | 北方民族大学 | A kind of three-dimensional point cloud model classification method based on convolutional neural networks |
| CN109035329A (en) * | 2018-08-03 | 2018-12-18 | 厦门大学 | Camera Attitude estimation optimization method based on depth characteristic |
| CN109410321A (en) * | 2018-10-17 | 2019-03-01 | 大连理工大学 | Three-dimensional rebuilding method based on convolutional neural networks |
| CN110188802A (en) * | 2019-05-13 | 2019-08-30 | 南京邮电大学 | SSD algorithm of target detection based on the fusion of multilayer feature figure |
| CN110081890A (en) * | 2019-05-24 | 2019-08-02 | 长安大学 | A kind of dynamic K arest neighbors map-matching method of combination depth network |
| CN110197223A (en) * | 2019-05-29 | 2019-09-03 | 北方民族大学 | Point cloud data classification method based on deep learning |
| CN110443842A (en) * | 2019-07-24 | 2019-11-12 | 大连理工大学 | Depth map prediction technique based on visual angle fusion |
| CN111027559A (en) * | 2019-10-31 | 2020-04-17 | 湖南大学 | Point cloud semantic segmentation method based on expansion point convolution space pyramid pooling |
| CN111242208A (en) * | 2020-01-08 | 2020-06-05 | 深圳大学 | Point cloud classification method, point cloud segmentation method and related equipment |
| CN111476226A (en) * | 2020-02-29 | 2020-07-31 | 新华三大数据技术有限公司 | Text positioning method and device and model training method |
| CN111583263A (en) * | 2020-04-30 | 2020-08-25 | 北京工业大学 | Point cloud segmentation method based on joint dynamic graph convolution |
| CN111666836A (en) * | 2020-05-22 | 2020-09-15 | 北京工业大学 | High-resolution remote sensing image target detection method of M-F-Y type lightweight convolutional neural network |
| CN111723814A (en) * | 2020-06-05 | 2020-09-29 | 中国科学院自动化研究所 | Cross-image association based weak supervision image semantic segmentation method, system and device |
| CN111915619A (en) * | 2020-06-05 | 2020-11-10 | 华南理工大学 | Full convolution network semantic segmentation method for dual-feature extraction and fusion |
| CN111753698A (en) * | 2020-06-17 | 2020-10-09 | 东南大学 | Multi-mode three-dimensional point cloud segmentation system and method |
| CN112036447A (en) * | 2020-08-11 | 2020-12-04 | 复旦大学 | Zero-sample target detection system and learnable semantic and fixed semantic fusion method |
| CN112184548A (en) * | 2020-09-07 | 2021-01-05 | 中国科学院深圳先进技术研究院 | Image super-resolution method, device, equipment and storage medium |
| CN112149725A (en) * | 2020-09-18 | 2020-12-29 | 南京信息工程大学 | Spectral domain graph convolution 3D point cloud classification method based on Fourier transform |
| CN112329771A (en) * | 2020-11-02 | 2021-02-05 | 元准智能科技(苏州)有限公司 | Building material sample identification method based on deep learning |
Non-Patent Citations (2)
| Title |
|---|
| ZHENGLI ZHAI ET AL.: "MULTI-SCALE DYNAMIC GRAPH CONVOLUTION NETWORK FOR POINT CLOUDS CLASSIFICATION", 《IEEE ACCESS》 * |
| 于挺等: "基于K近邻卷积神经网络的点云模型识别与分类", 《激光与光电子学进展》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113516663A (en) * | 2021-06-30 | 2021-10-19 | 同济大学 | Point cloud semantic segmentation method and device, electronic equipment and storage medium |
| CN113628217A (en) * | 2021-08-12 | 2021-11-09 | 江南大学 | Three-dimensional point cloud segmentation method based on image convolution and integrating direction and distance |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112967296B (en) | 2022-11-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110322453B (en) | 3D point cloud semantic segmentation method based on position attention and auxiliary network | |
| CN111161364B (en) | Real-time shape completion and attitude estimation method for single-view depth map | |
| CN112967296B (en) | Point cloud dynamic region graph convolution method, classification method and segmentation method | |
| CN111768415A (en) | Image instance segmentation method without quantization pooling | |
| CN112785526B (en) | Three-dimensional point cloud restoration method for graphic processing | |
| CN111695494A (en) | Three-dimensional point cloud data classification method based on multi-view convolution pooling | |
| CN110751195B (en) | Fine-grained image classification method based on improved YOLOv3 | |
| CN112419191B (en) | Image motion blur removing method based on convolution neural network | |
| CN111401380A (en) | RGB-D image semantic segmentation method based on depth feature enhancement and edge optimization | |
| Rios et al. | Feature visualization for 3D point cloud autoencoders | |
| CN111652273A (en) | Deep learning-based RGB-D image classification method | |
| CN114419413A (en) | Method for constructing sensing field self-adaptive transformer substation insulator defect detection neural network | |
| WO2024021461A1 (en) | Defect detection method and apparatus, device, and storage medium | |
| CN116091823A (en) | Single-feature anchor-frame-free target detection method based on fast grouping residual error module | |
| CN112509021A (en) | Parallax optimization method based on attention mechanism | |
| CN114373104A (en) | Three-dimensional point cloud semantic segmentation method and system based on dynamic aggregation | |
| CN115830375A (en) | Point cloud classification method and device | |
| CN113537119B (en) | Transmission line connecting part detection method based on improved Yolov4-tiny | |
| CN116188882A (en) | Point cloud up-sampling method and system integrating self-attention and multipath path diagram convolution | |
| CN116704378A (en) | Homeland mapping data classification method based on self-growing convolution neural network | |
| CN113808006B (en) | Method and device for reconstructing three-dimensional grid model based on two-dimensional image | |
| CN113723468B (en) | Object detection method of three-dimensional point cloud | |
| CN112365456B (en) | Transformer substation equipment classification method based on three-dimensional point cloud data | |
| CN115272673A (en) | Point cloud semantic segmentation method based on three-dimensional target context representation | |
| CN115131245A (en) | Point cloud completion method based on attention mechanism |
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: 20230628 Address after: No. 1811, 18th Floor, Building 19, Section 1201, Lushan Avenue, Wan'an Street, Tianfu New District, Chengdu, Sichuan, China (Sichuan) Pilot Free Trade Zone, 610213, China Patentee after: Sichuan Jiulai Technology Co.,Ltd. Address before: No. 69 lijiatuo Chongqing District of Banan City Road 400054 red Patentee before: Chongqing University of Technology |
|
| TR01 | Transfer of patent right |