CN114723877A - Countermeasure sample generation method for three-dimensional sparse convolution network - Google Patents
Countermeasure sample generation method for three-dimensional sparse convolution network Download PDFInfo
- Publication number
- CN114723877A CN114723877A CN202210188045.0A CN202210188045A CN114723877A CN 114723877 A CN114723877 A CN 114723877A CN 202210188045 A CN202210188045 A CN 202210188045A CN 114723877 A CN114723877 A CN 114723877A
- Authority
- CN
- China
- Prior art keywords
- point cloud
- cloud data
- dimensional
- dimensional point
- value
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Software Systems (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computer Graphics (AREA)
- Geometry (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The method, the device and the storage medium for generating the countermeasure sample for the three-dimensional sparse convolution network, provided by the application, are used for acquiring input three-dimensional point cloud data, performing voxelization processing on the three-dimensional point cloud data to obtain an input voxel set, inputting the input voxel set into the three-dimensional sparse convolution network to obtain a predicted value of the input voxel set, converting the predicted value into a predicted value of the three-dimensional point cloud data, obtaining a loss value of the three-dimensional point cloud data by using a loss function based on a true value and the predicted value of the three-dimensional point cloud data, obtaining a gradient of the three-dimensional point cloud data by using the loss value through the three-dimensional sparse convolution network, and finally obtaining the countermeasure sample for the three-dimensional sparse convolution network through the gradient of the three-dimensional point cloud data. The method for generating the countermeasure sample for the three-dimensional sparse convolution network based on the dynamic sensing is adopted to carry out the countermeasure training on the three-dimensional sparse convolution network, the robustness of the network can be improved, and the accuracy of network analysis is further improved.
Description
Technical Field
The application relates to the technical field of computer vision and machine learning, in particular to a method and a device for generating confrontation samples for a three-dimensional sparse convolution network and a storage medium.
Background
Deep neural networks have been greatly developed in recent years, greatly improving the performance of computer vision tasks. However, the deep neural network is vulnerable to adversarial attacks, and a well-trained deep neural network model can be forced to make mistakes by performing carefully designed minimal disturbance on network input data, so that a huge safety risk is hidden in the deep neural network in practical application.
The existing adversarial attack method assumes that the network architecture is fixed in the whole attack process, and in this case, the adversarial disturbance which is learned by each step of the attack and aims at the current network is effective. However, for some specific networks whose structure includes input-dependent execution units to improve computational efficiency, such as conditional computations, dynamic deep neural networks, and three-dimensional sparse convolutional networks, the above assumption does not hold. That is, when one-step antagonistic perturbation is learned, the network architecture may change after the one-step antagonistic perturbation, and the attack learned according to the network architecture before the change may not be effective on the changed network architecture.
The three-dimensional sparse convolution network is a main network type of large-scale three-dimensional scene analysis, the sparsity of three-dimensional point cloud data is fully utilized in the structure of the three-dimensional sparse convolution network, and the network structure is determined by an input voxel set. Unlike many point-specific networks that directly take three-dimensional point cloud data as input, three-dimensional sparse convolutional networks first convert the point cloud into a plurality of voxels with point occupancy in the three-dimensional grid, i.e., sparse voxels. In order to maintain sparsity throughout the network, the center of each sparse convolution kernel is dedicated to a unique sparse voxel. Therefore, the network architecture of the three-dimensional sparse convolutional network depends on the position of the input voxel, and the attack operation learned at the current moment may be ineffective on the network architecture after the attack. The method can dynamically and perceptually generate the confrontation sample of the three-dimensional point cloud data aiming at the three-dimensional sparse convolution network.
Disclosure of Invention
The application provides a method and a device for generating a countermeasure sample for a three-dimensional sparse convolution network and a storage medium, and provides the method for generating the countermeasure sample for the three-dimensional sparse convolution network.
The embodiment of the first aspect of the application provides a countermeasure sample generation method for a three-dimensional sparse convolutional network, which includes:
acquiring input three-dimensional point cloud data;
performing voxelization processing on the three-dimensional point cloud data to obtain an input voxel set;
inputting the input voxel set into the three-dimensional sparse convolution network to obtain a predicted value of the input voxel set, and converting the predicted value into a predicted value of three-dimensional point cloud data;
obtaining a loss value of the three-dimensional point cloud data by using a loss function based on the real value and the predicted value of the three-dimensional point cloud data;
obtaining the gradient of the three-dimensional point cloud data by using the loss value;
generating a countermeasure sample of the three-dimensional point cloud data for the three-dimensional sparse convolutional network through the gradient of the three-dimensional point cloud data.
The embodiment of the second aspect of the present application provides a countermeasure sample generation apparatus for a three-dimensional sparse convolutional network, including:
the acquisition module is used for acquiring input three-dimensional point cloud data;
the processing module is used for carrying out voxelization processing on the three-dimensional point cloud data to obtain an input voxel set;
the acquisition module is further used for inputting the input voxel set into the three-dimensional sparse convolution network, acquiring a predicted value of the input voxel set and converting the predicted value into a predicted value of three-dimensional point cloud data;
the calculation module is used for obtaining a loss value of the three-dimensional point cloud data by using a loss function based on a real value and a predicted value of the three-dimensional point cloud data;
the processing module is further used for obtaining the gradient of the three-dimensional point cloud data by using the loss value;
and the generation module is used for generating a countermeasure sample of the three-dimensional point cloud data aiming at the three-dimensional sparse convolution network through the gradient of the three-dimensional point cloud data.
A non-transitory computer-readable storage medium as set forth in an embodiment of the third aspect of the present application, wherein the non-transitory computer-readable storage medium stores a computer program; which when executed by a processor implements the method as shown in the first aspect above.
A computer device according to an embodiment of the fourth aspect of the present application includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the processor executes the computer program, the method according to the first aspect of the present application can be implemented.
The technical scheme provided by the embodiment of the application at least has the following beneficial effects:
according to the countermeasure sample generation method and device for the three-dimensional sparse convolution network and the storage medium, input three-dimensional point cloud data are obtained, voxelization processing is conducted on the three-dimensional point cloud data to obtain an input voxel set, the input voxel set is input into the three-dimensional sparse convolution network to obtain a predicted value of the input voxel set, the predicted value is converted into a predicted value of the three-dimensional point cloud data, a loss value of the three-dimensional point cloud data is obtained through a loss function based on a true value and the predicted value of the three-dimensional point cloud data, a gradient of the three-dimensional point cloud data is obtained through the three-dimensional sparse convolution network through the loss value, and finally, an countermeasure sample for the three-dimensional sparse convolution network is obtained through the gradient of the three-dimensional point cloud data. The method comprises the steps of utilizing a loss value to obtain the gradient of three-dimensional point cloud data through a three-dimensional sparse convolution network, reconstructing the network gradient to enable the network gradient to contain potential dynamic change information of a network architecture, and accordingly predicting the change of the network architecture in the next step.
Additional aspects and advantages of the present application will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the present application.
Drawings
The foregoing and/or additional aspects and advantages of the present application will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings of which:
fig. 1 is a schematic flow chart of a method for generating a countermeasure sample for a three-dimensional sparse convolutional network according to an embodiment of the present application;
fig. 2 is a schematic structural diagram of a countermeasure sample generation apparatus for a three-dimensional sparse convolutional network according to an embodiment of the present application.
Detailed Description
Reference will now be made in detail to embodiments of the present application, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to the same or similar elements or elements having the same or similar function throughout. The embodiments described below with reference to the accompanying drawings are illustrative and intended to explain the present application and should not be construed as limiting the present application.
The method and apparatus for generating a countermeasure sample for a three-dimensional sparse convolutional network according to the embodiments of the present application are described below with reference to the drawings.
Example one
Fig. 1 is a schematic flowchart of a method for generating a countermeasure sample for a three-dimensional sparse convolutional network according to an embodiment of the present application, and as shown in fig. 1, the method may include:
In an embodiment of the present invention, the input three-dimensional point cloud data may be labeled with a category.
And 103, inputting the input voxel set into a three-dimensional sparse convolution network to obtain a predicted value of the input voxel set, and converting the predicted value into a predicted value of three-dimensional point cloud data.
Wherein, in an embodiment of the invention, the use of the first occupancy value in the three-dimensional sparse convolutional network represents the effectiveness of the sparse convolutional network.
Specifically, in the embodiment of the present invention, the three-dimensional sparse convolution network is:
wherein the content of the first and second substances,representing a given input set of voxels for a first occupancy valueIn the case of (1), whether the voxel x corresponding to the center of the current convolution kernel is valid or not is judged, and if so, whether the voxel x corresponding to the center of the current convolution kernel is valid or not is judgedMake the convolution valid, if not, thenThe convolution is invalidated and the result of the convolution is 0.Is a neighboring voxel of the voxel x,for voxels x within the convolution kernelqCorresponding convolution weight, fqIs voxel xqF' is the output feature vector of the convolution operation centered on voxel x.
And, in embodiments of the present invention, the three-dimensional sparse convolution network is dynamically varied and limits the scope of the convolution operation to valid voxels and to invalid voxels adjacent to the valid voxels, i.e. these adjacent invalid voxels may become valid after an attack step. On this basis, the network gradient needs to be reconstructed so that the gradient can include the change of the first occupancy value caused by the input perturbation, so that the learned attack is dynamically perceived.
Further, in the embodiment of the present invention, the three-dimensional sparse convolution network may be applied to the field of different three-dimensional point cloud data. Specifically, in an embodiment of the present invention, if three-dimensional point cloud data with category labels are used to perform three-dimensional point cloud data classification training on a three-dimensional sparse convolution network, the obtained three-dimensional sparse convolution network is used to classify the three-dimensional point cloud data. In another embodiment of the present invention, if the three-dimensional sparse convolution network is trained to perform three-dimensional point cloud data segmentation by using three-dimensional point cloud data with category labels, the obtained three-dimensional sparse convolution network is used to segment the three-dimensional point cloud data.
And step 104, obtaining a loss value of the three-dimensional point cloud data by using a loss function based on the real value and the predicted value of the three-dimensional point cloud data.
And 105, obtaining the gradient of the three-dimensional point cloud data by utilizing the loss value through a three-dimensional sparse convolution network.
In an embodiment of the present invention, the method for obtaining the gradient of the three-dimensional point cloud data through the three-dimensional sparse convolution network by using the loss value may include: and performing partial differential calculation on the three-dimensional point cloud data through a three-dimensional sparse convolution network by utilizing the loss value to obtain a gradient.
And, in an embodiment of the present invention, the partial differential calculation of the three-dimensional point cloud data by the three-dimensional sparse convolution network using the loss value comprises:
it should be noted that, in the embodiment of the present invention, the partial differential calculation of the three-dimensional sparse convolution network without dynamic sensing is not performed as described aboveIs calculated therebyThe calculation of (a) is the difference between the dynamically perceived three-dimensional sparse convolutional network and the non-dynamically perceived three-dimensional sparse convolutional network.
Further, in embodiments of the present disclosure, the first occupancy value is paired using a micro-computable algorithmAnd carrying out differential calculation to obtain the gradient of the three-dimensional point cloud data. Wherein, in the embodiment of the invention, the first occupation value is processed by utilizing a micro algorithmThe method of performing a differential calculation may comprise calculating a first occupancy valueReconstructing to obtain a second occupancy valueAnd a differential calculation is performed on the second occupancy value.
Further, in an embodiment of the present invention, the first occupancy value is setReconstructing to obtain a second occupancy valueThe method of (3) may comprise the steps of:
step 1, constructing correlation valuesTo represent a collection of point cloudsAt a certain point in the middleDegree of association with voxel x, associationValue ofComprises the following steps:
wherein the content of the first and second substances,is a pointA distance vector to the voxel x center coordinate,as a distance vectorThe ith term in (i), λ, is a given parameter to control the slope of the boundary.
Step 2, obtaining a second occupancy value by utilizing the correlation valueSecond occupancy valueComprises the following steps:
wherein the content of the first and second substances,for a set of point clouds around voxel xThe inner part is adjacent.
In an embodiment of the present invention, the second occupancy value obtained as described aboveWhereinInputting a point cloud coordinate set so that the second occupancy value is directly determined by the original point cloud coordinate inputAnd is
Further, in the embodiment of the present disclosure, after the second occupancy value is obtained, a differential calculation is performed on the second occupancy value, and the result is substituted into the above partial differential calculation formula to obtain the gradient of the three-dimensional point cloud data.
And 106, generating a countermeasure sample of the three-dimensional point cloud data aiming at the three-dimensional sparse convolution network through the gradient of the three-dimensional point cloud data.
In an embodiment of the present invention, a method for generating a countermeasure sample of three-dimensional point cloud data for a three-dimensional sparse convolution network through a gradient of the three-dimensional point cloud data includes: and advancing the three-dimensional point cloud data by one step length in the gradient direction of the three-dimensional point cloud data to obtain a confrontation sample corresponding to the three-dimensional point cloud data.
The method for generating the countermeasure sample for the three-dimensional sparse convolution network comprises the steps of obtaining input three-dimensional point cloud data, conducting voxelization on the three-dimensional point cloud data to obtain an input voxel set, inputting the input voxel set into the three-dimensional sparse convolution network to obtain a predicted value of the input voxel set, converting the predicted value into a predicted value of the three-dimensional point cloud data, obtaining a loss value of the three-dimensional point cloud data by using a loss function based on a true value and the predicted value of the three-dimensional point cloud data, obtaining a gradient of the three-dimensional point cloud data by using the loss value through the three-dimensional sparse convolution network, and finally obtaining the countermeasure sample for the three-dimensional sparse convolution network through the gradient of the three-dimensional point cloud data. The method comprises the steps of utilizing a loss value to obtain the gradient of three-dimensional point cloud data through a three-dimensional sparse convolution network, reconstructing the network gradient to enable the network gradient to contain potential dynamic change information of a network architecture, and accordingly predicting the change of the network architecture in the next step.
Example two
Further, fig. 2 is a schematic structural diagram of a countermeasure sample generation apparatus for a three-dimensional sparse convolutional network according to an embodiment of the present application, and as shown in fig. 2, the countermeasure sample generation apparatus may include:
an obtaining module 201, configured to obtain input three-dimensional point cloud data;
the processing module 202 is configured to perform voxelization processing on the three-dimensional point cloud data to obtain an input voxel set;
the obtaining module 201 is further configured to input the input voxel set into the three-dimensional sparse convolution network, obtain a predicted value of the input voxel set, and convert the predicted value into a predicted value of the three-dimensional point cloud data;
a calculating module 203, configured to obtain a loss value of the input three-dimensional point cloud data by using a loss function based on a true value and a predicted value of the three-dimensional point cloud data;
the processing module 202 is further configured to obtain a gradient of the three-dimensional point cloud data by using the loss value;
a generating module 204, configured to generate a countermeasure sample of the three-dimensional point cloud data for the three-dimensional sparse convolutional network through a gradient of the three-dimensional point cloud data.
To implement the above embodiments, the present disclosure also proposes a non-transitory computer-readable storage medium.
A non-transitory computer-readable storage medium provided by an embodiment of the present disclosure stores a computer program; the computer program, when executed by a processor, is capable of implementing the method as shown in fig. 1.
In order to implement the above embodiments, the present disclosure also provides a computer device.
The computer equipment provided by the embodiment of the disclosure comprises a memory, a processor and a computer program which is stored on the memory and can run on the processor; the processor, when executing the program, is capable of implementing the method as shown in fig. 1.
In the countermeasure sample generation device for the three-dimensional sparse convolution network, input three-dimensional point cloud data are obtained, voxelization processing is conducted on the three-dimensional point cloud data to obtain an input voxel set, the input voxel set is input into the three-dimensional sparse convolution network to obtain a predicted value of the input voxel set, the predicted value is converted into a predicted value of the three-dimensional point cloud data, a loss value of the three-dimensional point cloud data is obtained through a loss function based on a true value and the predicted value of the three-dimensional point cloud data, a gradient of the three-dimensional point cloud data is obtained through the three-dimensional sparse convolution network through the loss value, and finally, an countermeasure sample for the three-dimensional sparse convolution network is obtained through the gradient of the three-dimensional point cloud data. The method comprises the steps of utilizing a loss value to obtain the gradient of three-dimensional point cloud data through a three-dimensional sparse convolution network, reconstructing the network gradient to enable the network gradient to contain potential dynamic change information of a network architecture, and accordingly predicting the change of the network architecture in the next step.
In the description herein, reference to the description of the term "one embodiment," "some embodiments," "an example," "a specific example," or "some examples," etc., means 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 application. In this specification, the schematic representations of the terms used above are not necessarily intended to 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. Furthermore, various embodiments or examples and features of different embodiments or examples described in this specification can be combined and combined by one skilled in the art without contradiction.
Any process or method descriptions in flow charts or otherwise described herein may be understood as representing modules, segments, or portions of code which include one or more executable instructions for implementing steps of a custom logic function or process, and alternate implementations are included within the scope of the preferred embodiment of the present application in which functions may be executed out of order from that shown or discussed, including substantially concurrently or in reverse order, depending on the functionality involved, as would be understood by those reasonably skilled in the art of the present application.
Although embodiments of the present application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present application, and that variations, modifications, substitutions and alterations may be made to the above embodiments by those of ordinary skill in the art within the scope of the present application.
Claims (10)
1. A method of antagonistic sample generation for a three-dimensional sparse convolutional network, the method comprising:
acquiring input three-dimensional point cloud data;
performing voxelization processing on the three-dimensional point cloud data to obtain an input voxel set;
inputting the input voxel set into the three-dimensional sparse convolution network to obtain a predicted value of the input voxel set, and converting the predicted value into a predicted value of the three-dimensional point cloud data;
obtaining a loss value of the three-dimensional point cloud data by using a loss function based on the real value and the predicted value of the three-dimensional point cloud data;
obtaining the gradient of the three-dimensional point cloud data by utilizing the loss value through the three-dimensional sparse convolution network;
generating a countermeasure sample of the three-dimensional point cloud data for the three-dimensional sparse convolutional network through the gradient of the three-dimensional point cloud data.
2. The method of claim 1, wherein the three-dimensional sparse convolutional network comprises representing the effectiveness of the sparse convolutional network using a first occupancy value,
the three-dimensional sparse convolution network is as follows:
wherein the content of the first and second substances,representing a given input set of voxels for a first occupancy valueIn the case of (1), whether the voxel x corresponding to the center of the current convolution kernel is valid or not is judged, and if so, whether the voxel x corresponding to the center of the current convolution kernel is valid or not is judgedMake the convolution valid, if not, thenThe convolution is invalidated and the convolution result is 0.Are the neighbors of the voxel x,for voxels x within the convolution kernelqCorresponding convolution weight, fqIs voxel xqF' is the output feature vector of the convolution operation centered on voxel x.
3. The method of claim 1, wherein the obtaining the gradient of the three-dimensional point cloud data through the three-dimensional sparse convolutional network using the loss value comprises performing a partial differential calculation on the three-dimensional point cloud data through the three-dimensional sparse convolutional network using the loss value to obtain a gradient,
the partial differential calculation includes:
6. The method as recited in claim 3, wherein said pair of first occupancy valuesReconstructing to obtain a second occupancy valueThe method comprises the following steps:
building relevance valuesTo represent a collection of point cloudsAt a certain point in the middleDegree of association with voxel x, the association valueComprises the following steps:
wherein the content of the first and second substances,is a pointA distance vector to the voxel x center coordinate,as a distance vectorThe ith term, λ is a given parameter to control the boundary slope;
obtaining a second occupancy value by using the correlation valueThe second occupancy valueComprises the following steps:
7. The method of claim 1, wherein generating the countermeasure sample of the three-dimensional point cloud data against the three-dimensional sparse convolutional network by the gradient of the three-dimensional point cloud data comprises obtaining the countermeasure sample corresponding to the three-dimensional point cloud data by advancing the three-dimensional point cloud data by one step in the gradient direction of the three-dimensional point cloud data.
8. A countermeasure sample generation apparatus for a three-dimensional sparse convolutional network, comprising:
the acquisition module is used for acquiring input three-dimensional point cloud data;
the processing module is used for carrying out voxelization processing on the three-dimensional point cloud data to obtain an input voxel set;
the acquisition module is further configured to input the input voxel set into the three-dimensional sparse convolutional network, obtain a predicted value of the input voxel set, and convert the predicted value into a predicted value of the three-dimensional point cloud data;
the computing module is used for obtaining a loss value of the input three-dimensional point cloud data by using a loss function based on a real value and a predicted value of the three-dimensional point cloud data;
the processing module is further used for obtaining the gradient of the three-dimensional point cloud data by using the loss value;
and the generation module is used for generating a countermeasure sample of the three-dimensional point cloud data aiming at the three-dimensional sparse convolution network through the gradient of the three-dimensional point cloud data.
9. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the method according to any one of claims 1-7 when executing the program.
10. A non-transitory computer-readable storage medium, on which a computer program is stored, which, when executed by a processor, implements the method of any one of claims 1-7.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210188045.0A CN114723877A (en) | 2022-02-28 | 2022-02-28 | Countermeasure sample generation method for three-dimensional sparse convolution network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210188045.0A CN114723877A (en) | 2022-02-28 | 2022-02-28 | Countermeasure sample generation method for three-dimensional sparse convolution network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114723877A true CN114723877A (en) | 2022-07-08 |
Family
ID=82236064
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210188045.0A Pending CN114723877A (en) | 2022-02-28 | 2022-02-28 | Countermeasure sample generation method for three-dimensional sparse convolution network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114723877A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115880547A (en) * | 2023-03-02 | 2023-03-31 | 宁波微科光电股份有限公司 | Foreign matter detection method and device based on image point cloud data and storage medium thereof |
-
2022
- 2022-02-28 CN CN202210188045.0A patent/CN114723877A/en active Pending
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115880547A (en) * | 2023-03-02 | 2023-03-31 | 宁波微科光电股份有限公司 | Foreign matter detection method and device based on image point cloud data and storage medium thereof |
CN115880547B (en) * | 2023-03-02 | 2023-11-21 | 宁波微科光电股份有限公司 | Foreign matter detection method and device based on image point cloud data and storage medium thereof |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110705029B (en) | Flow field prediction method of oscillating flapping wing energy acquisition system based on transfer learning | |
CN115834433B (en) | Data processing method and system based on Internet of things technology | |
CN109785409B (en) | Image-text data fusion method and system based on attention mechanism | |
CN116783620A (en) | Efficient three-dimensional object detection from point clouds | |
CN113435520A (en) | Neural network training method, device, equipment and computer readable storage medium | |
CN108764541B (en) | Wind energy prediction method combining space characteristic and error processing | |
Han et al. | L-Net: lightweight and fast object detector-based ShuffleNetV2 | |
Alsanad et al. | Real-time fuel truck detection algorithm based on deep convolutional neural network | |
CN114723877A (en) | Countermeasure sample generation method for three-dimensional sparse convolution network | |
Zhang et al. | A graph-voxel joint convolution neural network for ALS point cloud segmentation | |
CN112906800B (en) | Image group self-adaptive collaborative saliency detection method | |
CN114241314A (en) | Remote sensing image building change detection model and algorithm based on CenterNet | |
CN111860465A (en) | Remote sensing image extraction method, device, equipment and storage medium based on super pixels | |
Hu et al. | Unifying label propagation and graph sparsification for hyperspectral image classification | |
CN116292130A (en) | Wind driven generator state monitoring system and method based on cooperative analysis | |
Mujtaba et al. | Automatic solar panel detection from high-resolution orthoimagery using deep learning segmentation networks | |
CN113610015A (en) | Attitude estimation method, device and medium based on end-to-end rapid ladder network | |
CN113177546A (en) | Target detection method based on sparse attention module | |
CN108764514B (en) | Photovoltaic power generation power prediction method based on parallel operation | |
CN114882292B (en) | Remote sensing image ocean target identification method based on cross-sample attention mechanism graph neural network | |
CN113743189B (en) | Human body posture recognition method based on segmentation guidance | |
CN117404853B (en) | External circulating water cooling system and method for tunnel boring machine | |
CN116030347B (en) | High-resolution remote sensing image building extraction method based on attention network | |
CN116052006B (en) | Building edge optimization method based on multitask learning and dual lottery hypothesis | |
CN117351374B (en) | Remote sensing image saliency target detection method, system, equipment and medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |