CN112801910B - Channel state information image denoising method and indoor positioning model - Google Patents
Channel state information image denoising method and indoor positioning model Download PDFInfo
- Publication number
- CN112801910B CN112801910B CN202110171211.1A CN202110171211A CN112801910B CN 112801910 B CN112801910 B CN 112801910B CN 202110171211 A CN202110171211 A CN 202110171211A CN 112801910 B CN112801910 B CN 112801910B
- Authority
- CN
- China
- Prior art keywords
- random
- long
- block
- range
- state information
- 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
- 238000000034 method Methods 0.000 title claims abstract description 15
- 238000005070 sampling Methods 0.000 claims abstract description 18
- 238000013507 mapping Methods 0.000 claims description 46
- 239000011159 matrix material Substances 0.000 claims description 10
- 238000011176 pooling Methods 0.000 claims description 6
- 238000012549 training Methods 0.000 claims description 5
- 239000000126 substance Substances 0.000 claims description 3
- 238000013528 artificial neural network Methods 0.000 description 5
- 230000004807 localization Effects 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 4
- 230000015556 catabolic process Effects 0.000 description 3
- 238000006731 degradation reaction Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000010339 dilation Effects 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 238000010295 mobile communication Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 210000004205 output neuron Anatomy 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- 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
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4038—Image mosaicing, e.g. composing plane images from plane sub-images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/20—Image enhancement or restoration using local operators
- G06T5/30—Erosion or dilatation, e.g. thinning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2200/00—Indexing scheme for image data processing or generation, in general
- G06T2200/32—Indexing scheme for image data processing or generation, in general involving image mosaicing
-
- 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]
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a channel state information image denoising method and an indoor positioning model, comprising the following steps: the denoising autoencoder is used for obtaining a denoised channel state information image; the positioning network is used for carrying out indoor positioning according to the denoised channel state information image; the denoising self-encoder comprises a Bernoulli sampling module used for carrying out Bernoulli sampling on an input channel state information image to obtain an image pair; a plurality of encoding blocks for encoding the image pair into a hidden layer representation; a plurality of decoding blocks for decoding the hidden representations into a clean image; denoising reconstructing an input picture from an encoder; the positioning network is used for indoor positioning, the positioning performance can be improved by effectively utilizing low-grade characteristics, the information flow of the positioning network can be enhanced, and overfitting is relieved.
Description
Technical Field
The invention belongs to the technical field of intelligent information processing, and particularly relates to a channel state information image denoising method and an indoor positioning model.
Background
With the development of mobile communication technology and the popularization of intelligent terminals, people have made urgent demands for location-based services. Indoor positioning technology has wide application, such as indoor navigation, people flow monitoring and trajectory tracking. Since WiFi devices have been deployed in a large scale, an indoor positioning technology based on channel state information is expected to become a commercial technology. With the development of deep learning, the convolutional neural network can effectively extract image features and classify, so that channel state information is converted into images, and an indoor positioning problem is also converted into a problem about image classification.
Since gaussian noise is always present in a wireless link, the noise in training samples often causes problems of overfitting and network performance degradation, and therefore, the noise in an image needs to be removed firstly. Most existing denoising methods require the use of an image pair including a noise image and a clean image, however, a clean channel state information image is difficult to obtain. Therefore, only a clean channel state information image can be recovered from a channel state information image containing noise as much as possible.
How to design a neural network suitable for the channel state information image is still a problem to be studied.
Disclosure of Invention
The purpose of the invention is as follows: in order to solve the problem of low indoor positioning precision and accuracy, the invention provides a channel state information image denoising method and an indoor positioning model.
The technical scheme is as follows: a channel state information image denoising self-encoder sequentially comprises:
the Bernoulli sampling module is used for carrying out Bernoulli sampling on the input channel state information image to obtain an image pair;
a plurality of encoding blocks for encoding the image pair into a hidden layer representation;
and the decoding blocks are used for decoding the hidden layer representation to obtain a denoised image.
Further, the image pair is represented as
Wherein:
wherein, l represents element-by-element multiplication, b m A bernoulli matrix is represented, each element is set to 0 with probability p, and n represents the input channel state information image.
Further, the encoding block includes: the first coding block is composed of a partial convolution layer and a maximum pooling layer, the second coding block is composed of a partial convolution layer and a maximum pooling layer, and the third coding block is composed of a partial convolution layer.
Further, the decoding block includes: the decoding device comprises a first decoding block consisting of a convolutional layer, an up-sampling layer, a convolutional layer and two splicing operations, a second decoding block consisting of an up-sampling layer, a convolutional layer and two splicing operations, and a third decoding block consisting of three convolutional layers and two splicing operations.
The invention also discloses a channel state information image denoising method, which comprises the following steps:
step 1: constructing and training a denoising autoencoder, wherein the denoising autoencoder is a channel state information image denoising autoencoder;
step 2: collecting channel state information data, and converting the channel state information data into an RGB image;
and 3, step 3: and inputting the RGB image to a denoising autoencoder to obtain a denoised image.
Further, in step 1, the denoising autoencoder adopts a mean square error loss function for training:
where M represents M pairs of images,
denoised autoencoder representing the need for learning, b m Representing a bernoulli matrix, where each element is set to 0 with a probability p,
and
representing the image obtained after bernoulli sampling.
Further, in step 2, channel state information data is collected by using the TP-Link router and the Intel 5300 network card as a sending and receiving device.
The invention also discloses an indoor positioning model, which comprises:
the denoising autoencoder is used for obtaining a denoised channel state information image;
the positioning network is used for carrying out indoor positioning according to the denoised channel state information image;
the denoising autoencoder is a channel state information image denoising autoencoder;
the positioning network sequentially comprises a convolution layer, a first long-range random short circuit connecting block, a second long-range random short circuit connecting block, a third long-range random short circuit connecting block, an expansion convolution layer, a fourth long-range random short circuit connecting block and a full connecting layer; the first long-range random short circuit connecting block, the second long-range random short circuit connecting block, the third long-range random short circuit connecting block and the fourth long-range random short circuit connecting block all comprise random identity mapping blocks and random convolution mapping blocks.
Furthermore, the first long-range random short circuit connection block consists of a random convolution mapping block, two random identity mapping blocks and a long-range splicing operation; the second long-range random short circuit connecting block consists of a random convolution mapping block, three random identity mapping blocks and a long-range splicing operation; the third long-range random short circuit connecting block consists of a random convolution mapping block, five random identity mapping blocks and a long-range splicing operation; the fourth long-range random short circuit connection block is composed of a random convolution mapping block, two random identity mapping blocks and a long-range splicing operation.
Further, the random identity mapping block and the random convolution mapping block are both composed of three convolution layers and a short circuit connection;
the random identity mapping block is represented as:
the random convolution mapping block is represented as:
wherein the content of the first and second substances,
and
respectively representing input and output, w b Convolution kernels representing hidden layers, w s Representing the convolution kernel in a long-range stitching operation, B represents the bernoulli matrix,
representing the residual map that needs to be learnt.
Has the advantages that: according to the method, the input picture is reconstructed through the denoising neural network, indoor positioning is carried out through the residual error network, the denoising performance can be improved by effectively utilizing low-level characteristics, the information flow of the positioning network can be enhanced, and overfitting is relieved; compared with the prior art, the method has the following advantages:
(1) according to the invention, the denoising autoencoder is used for removing Gaussian noise in the channel state information image, so that the problems of over-fitting, performance reduction and the like of a positioning network are effectively avoided, and the denoising effect is improved by multiplexing shallow layer features;
(2) the invention uses the residual error network to carry out indoor positioning, can effectively solve the problems of overfitting, network degradation, gradient disappearance and the like of a deep neural network by adding a random matrix and long-range random short circuit connection into the residual error network, and adds an expansion convolution layer at the tail end of the positioning network to expand the receptive field, thereby effectively improving the precision and the accuracy of indoor positioning.
Drawings
FIG. 1 is a general flow diagram of the present invention;
FIG. 2 is a diagram of a denoised neural network architecture according to the present invention;
FIG. 3 is a diagram of a positioning neural network (ResFi) architecture of the present invention;
fig. 4 is a comparison of the performance of the positioning algorithm of the present invention with existing algorithms.
Detailed Description
The technical solution of the present invention will be further explained with reference to the accompanying drawings and examples.
Example 1:
referring to fig. 2, the channel state information image denoising method based on an autoencoder of the present embodiment specifically includes the following steps:
step 1: constructing and training a denoising autoencoder; the denoising self-encoder comprises a Bernoulli sampling module, a plurality of encoding blocks for encoding an input image into a hidden layer representation and a plurality of decoding blocks for restoring the hidden layer representation into a clean image; in this embodiment, the bernoulli sampling module is configured to perform bernoulli sampling on an input image with a size of 30 × 30 × 3 to obtain an image pair
Wherein:
wherein, l represents element-by-element multiplication, b m Representing a bernoulli matrix with each element set to 0 with probability p and n representing the input image.
In the embodiment, three coding blocks are used
Encoding the hidden representations, and restoring the hidden representations into a clean picture by three decoding blocks; specifically, the method comprises the following steps:
the first coding block consists of a partial convolution layer and a maximum pooling layer, the size of the convolution kernel is 3 multiplied by 3, and the number of channels is 32;
the second coding block consists of a partial convolution layer and a maximum pooling layer, the size of the convolution kernel is 3 multiplied by 3, and the number of channels is 64;
the last coding block is composed of a part of convolution layers, the size of the convolution kernel is 3 multiplied by 3, the number of channels is 128, and the size of the output characteristic mapping is 8 multiplied by 128.
The first decoding block consists of a convolution layer, an up-sampling layer, a convolution layer and two termination operations, the size of a convolution kernel is 3 multiplied by 3, and the number of channels is 128;
the second decoding block consists of an up-sampling layer, a convolution layer and two collocation operations, the convolution kernel size is 3 multiplied by 3, and the number of channels is 64;
the last coding block consists of three convolution layers and two Concatenation operations, the sizes of the convolution kernels are all 3 x 3, and the number of channels is 48, 24 and 3 respectively.
The denoised autoencoder is trained using a mean square error loss function:
where M represents M pairs of images,
denoised autoencoder representing the need for learning, b m Representing a bernoulli matrix, where each element is set to 0 with a probability p,
and
representing the images obtained after bernoulli sampling.
Step 2: collecting CSI data by using a TP-Link router and an Intel 5300 network card as a sending and receiving device, and converting the CSI data into an RGB image; in some embodiments, a single transmit and a triple receive are used, so that there are three wireless links, each link has 30 subcarriers, 30000 data packets are collected, each data packet has a size of 30 × 1 × 3, each data packet has a group of 30 × 30 × 3, each layer corresponds to RGB values, and thus can be converted into RGB images.
And 3, step 3: and (3) inputting the RGB image into the denoising autoencoder constructed in the step (1) to obtain a denoised image.
Example 2:
the present embodiment provides an indoor positioning method based on a residual error network on the basis of embodiment 1, and with reference to fig. 3, the method specifically includes the following steps:
step 1: the method comprises the following steps of constructing an indoor positioning model based on a residual error network, specifically, the indoor positioning model sequentially comprises:
a convolution layer, the input of which is a clean image obtained after denoising in embodiment 1, the convolution kernel size is 3 × 3 × 64;
the first long-range random short circuit connecting block is composed of a random convolution mapping block, two random identity mapping blocks and a long-range collocation operation, wherein the channel numbers of the random convolution mapping blocks are respectively 64, 64 and 128; the channel numbers of the random identity mapping blocks are respectively 64, 64 and 128;
a second long-range random short circuit connection block, which is composed of a random convolution mapping block, three random identity mapping blocks and a long-range localization operation; the number of channels of the random convolution mapping block is respectively 128, 128 and 256, and the number of channels of the random identity mapping block is respectively 128, 128 and 256;
a third long-range random short-circuit connection block, which is composed of a random convolution mapping block (the channels are respectively 256, 256 and 512), five random identity mapping blocks (the channels are respectively 256, 256 and 512) and a long-range localization operation;
an expansion convolution layer with expansion rate of 2, convolution kernel size of 3 x 3 and channel number of 512; the dilated convolution can be used to increase the receptive field and place it at the end of the localization network;
a fourth long-range random short-circuit connection block, which is composed of a random convolution mapping block (channels are 512, 1024 respectively), two random identity mapping blocks (channels are 512, 1024 respectively) and a long-range localization operation;
a fully connected layer having 10 output neurons.
The above mentioned Concatenation operation means to concatenate the input of the module and the output of the network, which is distinguished from the addition operation: addition is the addition of input and output element by element, the output dimension cannot be increased, the output dimension can be increased by concationration, and the long range refers to large span.
The above mentioned dilation convolution layer will increase the computation significantly, the net shallow feature map is larger, the deep feature map is smaller, so it is placed at the end to reduce the computation, and it is not placed behind the fourth long range random short-circuited block because the feature map is too small to do dilation convolution.
According to the structure, the long-range random short circuit connection block in the indoor positioning model can map shallow features to a network deep layer, so that information flow is further enhanced, and the problem of network degradation is solved. The long-range random short circuit connection blocks mentioned above are composed of two random residual blocks, one is a random identity mapping block, and the other is a random convolution mapping block. The two random residual blocks are composed of three convolution layers and a short-circuit connection, the sizes of convolution kernels are 1 multiplied by 1, 3 multiplied by 3 and 1 multiplied by 1 respectively, a Bernoulli matrix B is added into the short-circuit connection, and learned feature mapping is discarded randomly; short-circuit connection is a common operation of deep learning, and the input and the output are added, and there are two ways: 1. concatenation 2, element-wise addition. As shown in formulas (3) and (4), the second is used here. Expression (3) is an expression form of the random identity mapping block, and expression (4) is an expression form of the random convolution mapping block:
wherein the content of the first and second substances,
and
respectively representing an input layer and an output layer, w b Convolution kernels representing hidden layers, w s Represents the convolution kernel in the long-range localization operation,
representing a residual map indicating the need for learning.
The performance of the indoor positioning network (ResFi) provided by the present invention is now compared with the performance of the mainstream network, including the average positioning error and standard deviation of positioning. Observing table 1 and fig. 4, it can be found that the indoor positioning network (ResFi) provided by the present invention has the best performance and higher positioning accuracy compared with other networks.
Table 1 shows the comparison of the performance of the indoor positioning network (ResFi) and the mainstream network provided by the present invention
| Network | Mean error (m) | Standard deviation (m) |
| DANN | 2.3910 | 1.6507 |
| DeepFi | 2.1082 | 1.4821 |
| ConFi | 1.9365 | 1.3554 |
| ResFi | 1.7873 | 1.2806 |
Claims (8)
1. An indoor positioning model, its characterized in that: the method comprises the following steps:
the denoising autoencoder is used for obtaining a denoised channel state information image;
the positioning network is used for carrying out indoor positioning according to the denoised channel state information image;
wherein, the denoising self-encoder sequentially comprises:
the Bernoulli sampling module is used for carrying out Bernoulli sampling on the input channel state information image to obtain an image pair;
a plurality of encoding blocks for encoding the image pair into a hidden layer representation;
the decoding blocks are used for decoding the hidden layer representation to obtain a denoised image;
the positioning network sequentially comprises a convolution layer, a first long-range random short circuit connecting block, a second long-range random short circuit connecting block, a third long-range random short circuit connecting block, an expansion convolution layer, a fourth long-range random short circuit connecting block and a full connecting layer; the first long-range random short circuit connecting block, the second long-range random short circuit connecting block, the third long-range random short circuit connecting block and the fourth long-range random short circuit connecting block comprise random identity mapping blocks and random convolution mapping blocks;
the first long-range random short circuit connecting block consists of a random convolution mapping block, two random identity mapping blocks and a long-range splicing operation; the second long-range random short circuit connecting block consists of a random convolution mapping block, three random identity mapping blocks and a long-range splicing operation; the third long-range random short circuit connecting block consists of a random convolution mapping block, five random identity mapping blocks and a long-range splicing operation; the fourth long-range random short circuit connection block is composed of a random convolution mapping block, two random identity mapping blocks and a long-range splicing operation.
2. The indoor positioning model of claim 1, wherein: the random identity mapping block and the random convolution mapping block are both composed of three convolution layers and a short circuit connection;
the random identity mapping block is represented as:
the random convolution mapping block is represented as:
wherein the content of the first and second substances,
and
respectively representing input and output, w b Convolution kernel representing hidden layer, w s Representing the convolution kernel in a long-range stitching operation, B represents the bernoulli matrix,
representing the residual map that needs to be learnt.
3. The indoor positioning model of claim 1, wherein: said pair of images being represented as
Wherein:
wherein, l represents element-by-element multiplication, b m A bernoulli matrix is represented, each element is set to 0 with probability p, and n represents the input channel state information image.
4. The indoor positioning model of claim 1, wherein: the coding block includes: the first coding block is composed of a partial convolutional layer and a max-pooling layer, the second coding block is composed of a partial convolutional layer and a max-pooling layer, and the third coding block is composed of a partial convolutional layer.
5. The indoor positioning model of claim 1, wherein: the decoding block includes: the decoding device comprises a first decoding block consisting of a convolutional layer, an up-sampling layer, a convolutional layer and two splicing operations, a second decoding block consisting of an up-sampling layer, a convolutional layer and two splicing operations, and a third decoding block consisting of three convolutional layers and two splicing operations.
6. The indoor positioning model of claim 1, wherein: the denoising autoencoder is obtained by adopting mean square error loss function training;
where M represents M pairs of images,
de-noising auto-encoder representing the need for de-learning, b m Representing a bernoulli matrix, where each element is set to 0 with a probability p,
and
representing the images obtained after bernoulli sampling.
7. The indoor positioning model of claim 6, wherein: the denoising autoencoder is configured to obtain a denoised channel state information image, and specifically includes:
collecting channel state information data, and converting the channel state information data into an RGB image;
and inputting the RGB image to a denoising autoencoder to obtain a denoised image.
8. The indoor positioning model of claim 7, wherein: the acquiring of the channel state information data specifically includes:
channel state information data is collected by using a TP-Link router and an Intel 5300 network card as transmitting and receiving devices.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110171211.1A CN112801910B (en) | 2021-02-08 | 2021-02-08 | Channel state information image denoising method and indoor positioning model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110171211.1A CN112801910B (en) | 2021-02-08 | 2021-02-08 | Channel state information image denoising method and indoor positioning model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112801910A CN112801910A (en) | 2021-05-14 |
| CN112801910B true CN112801910B (en) | 2022-09-23 |
Family
ID=75814790
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110171211.1A Active CN112801910B (en) | 2021-02-08 | 2021-02-08 | Channel state information image denoising method and indoor positioning model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112801910B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115849519B (en) * | 2022-12-30 | 2024-03-22 | 浙江致远环境科技股份有限公司 | Organic modularized electrocatalytic oxidation treatment device |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111127331A (en) * | 2019-10-22 | 2020-05-08 | 广东启迪图卫科技股份有限公司 | Image denoising method based on pixel-level global noise estimation coding and decoding network |
| CN111669820A (en) * | 2020-05-27 | 2020-09-15 | 南京邮电大学 | Density peak value abnormity detection method and intelligent passive indoor positioning method |
-
2021
- 2021-02-08 CN CN202110171211.1A patent/CN112801910B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111127331A (en) * | 2019-10-22 | 2020-05-08 | 广东启迪图卫科技股份有限公司 | Image denoising method based on pixel-level global noise estimation coding and decoding network |
| CN111669820A (en) * | 2020-05-27 | 2020-09-15 | 南京邮电大学 | Density peak value abnormity detection method and intelligent passive indoor positioning method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112801910A (en) | 2021-05-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110232394B (en) | Multi-scale image semantic segmentation method | |
| CN108376387B (en) | Image deblurring method based on aggregation expansion convolution network | |
| CN111340814A (en) | Multi-mode adaptive convolution-based RGB-D image semantic segmentation method | |
| CN112036454B (en) | Image classification method based on multi-core dense connection network | |
| CN112511555A (en) | Private encryption protocol message classification method based on sparse representation and convolutional neural network | |
| CN112668459B (en) | Rolling bearing fault diagnosis method based on supervised contrast learning and convolutional neural network | |
| CN110446173B (en) | Efficient and energy-saving satellite-borne wireless sensor network data compression method | |
| CN113642445B (en) | Hyperspectral image classification method based on full convolution neural network | |
| CN112801910B (en) | Channel state information image denoising method and indoor positioning model | |
| US11700156B1 (en) | Intelligent data and knowledge-driven method for modulation recognition | |
| CN111224905B (en) | Multi-user detection method based on convolution residual error network in large-scale Internet of things | |
| CN110796027A (en) | Sound scene recognition method based on compact convolution neural network model | |
| CN112867010A (en) | Radio frequency fingerprint embedded real-time identification method and system based on convolutional neural network | |
| CN111159250B (en) | Mobile terminal user behavior detection method based on nested deep twin neural network | |
| CN112418229A (en) | Unmanned ship marine scene image real-time segmentation method based on deep learning | |
| CN116434039A (en) | Target detection method based on multiscale split attention mechanism | |
| CN113139577B (en) | Deep learning image classification method and system based on deformable convolution network | |
| CN111414941B (en) | Point cloud convolution neural network based on feature multiplexing | |
| CN112487718B (en) | Satellite image inversion radar combined reflectivity method based on deep learning | |
| CN113807370A (en) | Data processing method, device, equipment, storage medium and computer program product | |
| CN111026835B (en) | Chat subject detection method, device and storage medium | |
| CN113095356A (en) | Light weight type neural network and image processing method and device | |
| CN111368922B (en) | Point cloud processing network architecture for object classification | |
| KR20210038027A (en) | Method for Training to Compress Neural Network and Method for Using Compressed Neural Network | |
| CN114255348B (en) | Insulator aging and fouling spectrum classification method for improving B _ CNN |
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 |