CN116091362A - Remote sensing image data enhancement method based on variation self-encoder - Google Patents
Remote sensing image data enhancement method based on variation self-encoder Download PDFInfo
- Publication number
- CN116091362A CN116091362A CN202310334749.9A CN202310334749A CN116091362A CN 116091362 A CN116091362 A CN 116091362A CN 202310334749 A CN202310334749 A CN 202310334749A CN 116091362 A CN116091362 A CN 116091362A
- Authority
- CN
- China
- Prior art keywords
- remote sensing
- encoder
- sensing image
- cloud
- hidden variable
- 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
- G06T5/00—Image enhancement or restoration
-
- 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
- G06T2207/10032—Satellite or aerial image; Remote sensing
-
- 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]
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Abstract
The invention provides a remote sensing image data enhancement method based on a variation self-encoder, which belongs to the field of remote sensing image processing, and comprises the steps of firstly constructing a remote sensing image data set, and structurally compiling remote sensing images with different backgrounds, latitude distribution, imaging time and different degrees of cloud and snow according to attributes; secondly, constructing an encoder, a hidden variable layer and a decoder to form a variable self-encoder network; then extracting the change attribute, inputting the remote sensing images with different change attributes and the same other attributes into a change self-encoder network in 5 batches to participate in network parameter training; and finally resampling hidden variables in different hidden variable clusters, taking the resampled value as the input of a decoder, and decoding to generate the remote sensing image with specific attribute change. Compared with the traditional remote sensing image data enhancement method, the method can generate images with gradually changing background, latitude distribution, imaging time, cloud and fog conditions and snow conditions according to task intention, and reduces the randomness of data enhancement.
Description
Technical Field
The invention belongs to the field of remote sensing image processing, and particularly relates to a remote sensing image data enhancement method based on a variation self-encoder.
Background
The development of artificial intelligence technologies such as deep learning greatly improves the performance level of tasks such as natural language processing, computer vision, automatic driving and the like, however, a current reliable deep learning model is generally obtained by training in a supervised learning mode, a large number of training samples are needed for model training in the supervised learning mode, and in order to expand the data volume required by the supervised learning, reduce the cost of data labeling, improve the robustness and generalization capability of the model, and provide additional training data for the supervised learning method in a data enhancement mode.
The conventional data enhancement mode in the fields of image recognition and target detection generally adopts the modes of random cutting, random overturning, brightness adjustment, contrast adjustment, random splicing and the like, so that the distribution range of training data in a high-dimensional space is expanded to a certain extent, but the distribution range of real data cannot be reflected, particularly in the field of remote sensing image processing, the conventional data enhancement mode cannot reflect the change of a remote sensing image due to the influence of factors such as background, rain, snow, latitude and the like.
Disclosure of Invention
The object of the invention is to generate enhancement data which is closer to real data on the one hand and to be able to select specific properties which need to be enhanced in accordance with the purpose of downstream tasks on the other hand.
The invention adopts the technical scheme that:
a remote sensing image data enhancement method based on a variation self-encoder comprises the following steps:
step 1, structuring and compiling remote sensing images with different backgrounds, different latitude distributions, different imaging times and different degrees of cloud and snow according to attributes to form a remote sensing image data set;
step 2, constructing a variable self-encoder network, which comprises an encoder, an hidden variable layer and a decoder, wherein the encoder and the decoder adopt a multi-layer convolutional neural network, the structures of the encoder and the decoder are symmetrical structures, the hidden variable layer comprises 5 hidden variable clusters, each hidden variable cluster comprises 10 hidden variables, and each hidden variable obeys Gaussian distribution;
step 3, respectively extracting one attribute of the background, latitude distribution, imaging time, cloud and fog conditions and snow conditions as a change attribute, inputting remote sensing images with different change attributes and the same other attributes into a change self-encoder network in 5 batches for parameter training, sequentially activating corresponding hidden variable cluster parameters according to the change attributes during training, and freezing other hidden variable cluster parameters;
and 4, resampling hidden variables in different hidden variable clusters according to the downstream task demand, taking the resampled value as the input of a decoder, and decoding to generate the remote sensing image with specific attribute change.
Further, the remote sensing image data set construction method in the step 1 adopts a tree structure to carry out structural assembly on the attribute of the remote sensing image, the depth of the tree structure is 5, the tree structure comprises 5 layers of nodes, the 1 st layer of nodes reflect imaging background, the 2 nd layer of nodes reflect latitude, the 3 rd layer of nodes reflect imaging time, the 4 th layer of nodes reflect cloud and fog conditions, the 5 th layer of nodes reflect snow conditions, wherein the 1 st layer comprises forest lands, deserts, grasslands, bare lands and water bodies, the 2 nd layer further subdivides each node of the first layer into high latitude, medium latitude and low latitude according to the latitude, the 3 rd layer categorizes the tree structure into spring, summer, autumn and winter according to the imaging time, the 4 th layer categorizes the tree structure into cloud content of more than 50%, cloud content of less than 50% and no cloud and fog according to the cloud and fog conditions, the 5 th layer categorizes the snow content of more than 50% and snow content of no snow, the number of the nodes of the 5 layers of the tree structure are respectively 5, 15, 60, 180 and 540, and the remote sensing image is stored in each node according to the attribute.
Further, the encoder of the variable self-encoder network in the step 2 consists of a layer 3 convolutional neural network and a layer 1 encoding vector, the decoder is structurally symmetrical to the encoder, and the decoder and the encoder share the weight of the encoding vector.
Further, in step 2, the mean and variance of the gaussian distribution are output by the encoder, the decoder performs random sampling according to the gaussian distribution, and the sampling result is combined with the encoding vector to decode to generate an output image.
Further, the parameter training in the step 3 is specifically:
extracting the background, latitude distribution, imaging time and cloud conditions from a remote sensing image data set, inputting data with different snow conditions into a variable self-encoder network, activating parameters of a first hidden variable cluster, freezing parameters of other hidden variable clusters, and training by adopting an optimization algorithm until convergence;
then extracting the background, latitude distribution, imaging time and snow condition from the remote sensing image data set, inputting data with different cloud and fog conditions into a variable self-encoder network, activating parameters of a second hidden variable cluster, freezing parameters of other hidden variable clusters, and training by adopting an optimization algorithm until convergence;
then, extracting the background, latitude distribution, snow accumulation condition and cloud condition from the remote sensing image data set, inputting data with different imaging time to a variable self-encoder network, activating parameters of a third hidden variable cluster, freezing parameters of other hidden variable clusters, and training by adopting an optimization algorithm until convergence;
then, extracting the background, snow condition, imaging time and cloud condition from the remote sensing image data set, inputting data with different latitude distribution into a variable self-encoder network, activating parameters of a fourth hidden variable cluster, freezing parameters of other hidden variable clusters, and training by adopting an optimization algorithm until convergence;
and then extracting the snow state, latitude distribution, imaging time and cloud state from the remote sensing image data set, inputting data with different backgrounds into a variable self-encoder, activating parameters of a fifth hidden variable cluster, freezing parameters of other hidden variable clusters, and training by adopting an optimization algorithm until convergence.
Further, the step 4 specifically includes: and according to the downstream task demand, respectively and independently sampling or combining sampling the first hidden variable cluster to the fifth hidden variable cluster, taking the resampling value as the input of a decoder, and generating a remote sensing image with the background, latitude distribution, imaging time, cloud and fog conditions and snow accumulation conditions changed along with the background, latitude distribution, imaging time, cloud and fog conditions by the decoder.
Compared with the prior art, the invention has the advantages that:
(1) According to the invention, the remote sensing image is assembled through the tree structure to obtain the remote sensing image data set, so that the image generation training is conveniently carried out on a plurality of generation models such as a self-encoder, a generation countermeasure network and the like;
(2) According to the invention, the hidden variable clusters are introduced into the variable self-encoder, so that the encoding result of the encoder is stronger in structure, and the risk that the encoding result has no practical meaning is reduced;
(3) According to the invention, through improving the training process, the decoder can generate images with gradually changed background, latitude distribution, imaging time, cloud and fog conditions and snow conditions according to task intention, so that randomness is reduced.
Drawings
Fig. 1 is a tree structure constructed by a remote sensing image dataset according to the present invention.
Fig. 2 is a block diagram of a variable self-encoder according to the present invention.
Detailed Description
The technical scheme of the invention is clearly and completely described below with reference to the accompanying drawings, and the invention provides a remote sensing image data enhancement method based on a variation self-encoder.
Step 1, constructing a remote sensing image data set, namely structurally assembling remote sensing images with different backgrounds, different latitude distributions, different imaging times and different degrees of cloud and snow according to attributes, wherein the data set constructing method adopts a tree structure to structurally assemble the attributes of the remote sensing images, as shown in figure 1, the depth of the tree structure is 5, the nodes with 5 layers are included, the 1 st layer node reflects the imaging background, the 2 nd layer node reflects the latitude, the 3 rd layer node reflects the imaging time, the 4 th layer node reflects the cloud condition, the 5 th layer node reflects the snow condition, wherein the 1 st layer comprises woodland, desert, grassland, bare land and water body, the method comprises the steps of (1) dividing each node of a first level into high latitude, medium latitude and low latitude according to the latitude, dividing the first level into spring, summer, autumn and winter according to imaging time, dividing the first level into a large amount of cloud and fog (with cloud content of more than 50%), a small amount of cloud and fog (with cloud content of less than 50%) and no cloud according to the cloud and fog condition, dividing the second level into a large amount of snow (with snow content of more than 50%), a small amount of snow (with snow content of less than 50%) and no snow according to the snow condition, wherein the number of nodes of 5 layers of nodes of a tree structure is 5, 15, 60, 180 and 540 respectively, and collecting not less than 10 remote sensing images by each leaf node to form a remote sensing image data set;
step 2, constructing a variable self-encoder network, wherein the variable self-encoder network consists of an encoder, an hidden variable layer and a decoder, the encoder and the decoder are respectively a multi-layer convolutional neural network as shown in fig. 2, the encoder consists of a 3-layer convolutional neural network and a 1-layer code vector, the decoder is symmetrical to the encoder in structure, the decoder shares the weight of the code vector with the encoder, does not share the weight of the convolutional neural network, the hidden variable layer comprises 5 hidden variable clusters, each hidden variable cluster comprises 10 hidden variables, each hidden variable follows Gaussian distribution, the average value and the variance of the Gaussian distribution are output by the encoder, the decoder randomly samples according to the Gaussian distribution, and decodes the sampling result by combining the code vector to generate an output image;
step 3, parameter training of a variable self-encoder, extracting the same background, latitude distribution, imaging time and cloud and fog conditions from a remote sensing image data set, inputting data with different snow accumulation conditions into the variable self-encoder, activating parameters of hidden variable clusters 1, freezing other hidden variable cluster parameters, and training by adopting an optimization algorithm until convergence;
then extracting the background, latitude distribution, imaging time and snow condition from the remote sensing image data set, inputting data with different cloud and fog conditions into a variable self-encoder, activating parameters of hidden variable clusters 2, freezing other hidden variable cluster parameters, and training by adopting an optimization algorithm until convergence;
then, extracting the background, latitude distribution, snow accumulation condition and cloud condition from the remote sensing image data set, inputting data with different imaging time to a variable self-encoder, activating parameters of hidden variable clusters 3, freezing other hidden variable cluster parameters, and training by adopting an optimization algorithm until convergence;
then, extracting the background, snow condition, imaging time and cloud condition from the remote sensing image data set, inputting data with different latitude distribution into a variable self-encoder, activating parameters of hidden variable clusters 4, freezing other hidden variable cluster parameters, and training by adopting an optimization algorithm until convergence;
and then extracting the snow state, latitude distribution, imaging time and cloud state from the remote sensing image data set, inputting data with different backgrounds into a variable self-encoder, activating parameters of hidden variable clusters 5, freezing other hidden variable cluster parameters, and training by adopting an optimization algorithm until convergence.
And 4, enhancing remote sensing image data, namely resampling hidden variables in different hidden variable clusters according to downstream task demands, inputting resampling values as a decoder, and respectively and independently or combined sampling hidden variable clusters 1 to 5 according to task demands, wherein the decoder generates remote sensing images with background, latitude distribution, imaging time, cloud and fog conditions and snow conditions changed along with the background, latitude distribution, imaging time, cloud and fog conditions, so that the aim of providing data enhancement service for downstream deep learning tasks is fulfilled.
Claims (6)
1. The remote sensing image data enhancement method based on the variation self-encoder is characterized by comprising the following steps of:
step 1, structuring and compiling remote sensing images with different backgrounds, different latitude distributions, different imaging times and different degrees of cloud and snow according to attributes to form a remote sensing image data set;
step 2, constructing a variable self-encoder network, which comprises an encoder, an hidden variable layer and a decoder, wherein the encoder and the decoder adopt a multi-layer convolutional neural network, the structures of the encoder and the decoder are symmetrical structures, the hidden variable layer comprises 5 hidden variable clusters, each hidden variable cluster comprises 10 hidden variables, and each hidden variable obeys Gaussian distribution;
step 3, respectively extracting one attribute of the background, latitude distribution, imaging time, cloud and fog conditions and snow conditions as a change attribute, inputting remote sensing images with different change attributes and the same other attributes into a change self-encoder network in 5 batches for parameter training, sequentially activating corresponding hidden variable cluster parameters according to the change attributes during training, and freezing other hidden variable cluster parameters;
and 4, resampling hidden variables in different hidden variable clusters according to the downstream task demand, taking the resampled value as the input of a decoder, and decoding to generate the remote sensing image with specific attribute change.
2. The remote sensing image data enhancement method based on the variable-classification self-encoder according to claim 1, wherein the remote sensing image data set construction method in the step 1 adopts a tree structure to carry out structural assembly on the attributes of the remote sensing image, the depth of the tree structure is 5, the nodes comprise 5 layers of nodes, the 1 st layer of nodes reflect imaging background, the 2 nd layer of nodes reflect latitude, the 3 rd layer of nodes reflect imaging time, the 4 th layer of nodes reflect cloud and fog conditions, the 5 th layer of nodes reflect snow and snow conditions, the 1 st layer of nodes comprise forest lands, deserts, grasslands, bare lands and water bodies, the 2 nd layer of nodes further subdivide each node of the first layer into high latitude, medium latitude and low latitude according to the latitude, the 3 rd layer of nodes are divided into spring, summer, autumn and winter according to imaging time, the 4 th layer of nodes comprise more than 50% of cloud, less than 50% of cloud and no cloud according to snow and the number of the 5 th layer of nodes are divided into more than 50% of cloud and less than 50% of snow and the 5 th layer of nodes are respectively stored in the remote sensing image in the middle of each node according to the snow and snow conditions, the node is respectively stored in the nodes according to the attributes of 5, 60, 180 and the nodes.
3. The method of claim 1, wherein the encoder of the variable self-encoder network in step 2 consists of a layer 3 convolutional neural network and a layer 1 encoded vector, the decoder is structurally symmetrical to the encoder, and the decoder and the encoder share the weights of the encoded vectors.
4. The method for enhancing remote sensing image data based on a variance self-encoder as claimed in claim 1, wherein in the step 2, the mean and variance of the gaussian distribution are outputted by the encoder, the decoder performs random sampling according to the gaussian distribution, and the sampling result is decoded by combining with the encoding vector to generate the output image.
5. The remote sensing image data enhancement method based on a variation self-encoder as set forth in claim 1, wherein the parameter training in step 3 is specifically:
extracting the background, latitude distribution, imaging time and cloud conditions from a remote sensing image data set, inputting data with different snow conditions into a variable self-encoder network, activating parameters of a first hidden variable cluster, freezing parameters of other hidden variable clusters, and training by adopting an optimization algorithm until convergence;
then extracting the background, latitude distribution, imaging time and snow condition from the remote sensing image data set, inputting data with different cloud and fog conditions into a variable self-encoder network, activating parameters of a second hidden variable cluster, freezing parameters of other hidden variable clusters, and training by adopting an optimization algorithm until convergence;
then, extracting the background, latitude distribution, snow accumulation condition and cloud condition from the remote sensing image data set, inputting data with different imaging time to a variable self-encoder network, activating parameters of a third hidden variable cluster, freezing parameters of other hidden variable clusters, and training by adopting an optimization algorithm until convergence;
then, extracting the background, snow condition, imaging time and cloud condition from the remote sensing image data set, inputting data with different latitude distribution into a variable self-encoder network, activating parameters of a fourth hidden variable cluster, freezing parameters of other hidden variable clusters, and training by adopting an optimization algorithm until convergence;
and then extracting the snow state, latitude distribution, imaging time and cloud state from the remote sensing image data set, inputting data with different backgrounds into a variable self-encoder, activating parameters of a fifth hidden variable cluster, freezing parameters of other hidden variable clusters, and training by adopting an optimization algorithm until convergence.
6. The remote sensing image data enhancement method based on a variation self-encoder as set forth in claim 5, wherein the step 4 is specifically: and according to the downstream task demand, respectively and independently sampling or combining sampling the first hidden variable cluster to the fifth hidden variable cluster, taking the resampling value as the input of a decoder, and generating a remote sensing image with the background, latitude distribution, imaging time, cloud and fog conditions and snow accumulation conditions changed along with the background, latitude distribution, imaging time, cloud and fog conditions by the decoder.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310334749.9A CN116091362B (en) | 2023-03-31 | 2023-03-31 | Remote sensing image data enhancement method based on variation self-encoder |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310334749.9A CN116091362B (en) | 2023-03-31 | 2023-03-31 | Remote sensing image data enhancement method based on variation self-encoder |
Publications (2)
Publication Number | Publication Date |
---|---|
CN116091362A true CN116091362A (en) | 2023-05-09 |
CN116091362B CN116091362B (en) | 2023-06-09 |
Family
ID=86210447
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310334749.9A Active CN116091362B (en) | 2023-03-31 | 2023-03-31 | Remote sensing image data enhancement method based on variation self-encoder |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116091362B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117710371A (en) * | 2024-02-05 | 2024-03-15 | 成都数之联科技股份有限公司 | Method, device, equipment and storage medium for expanding defect sample |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20200361083A1 (en) * | 2019-05-15 | 2020-11-19 | Nvidia Corporation | Grasp generation using a variational autoencoder |
CN114170671A (en) * | 2021-09-16 | 2022-03-11 | 上海大学 | Massage manipulation identification method based on deep learning |
US20220215262A1 (en) * | 2021-01-05 | 2022-07-07 | Capital One Services, Llc | Augmenting Datasets with Synthetic Data |
CN115115028A (en) * | 2022-06-21 | 2022-09-27 | 国网湖北省电力有限公司信息通信公司 | Mixed data generation method based on generation countermeasure network |
CN115294510A (en) * | 2022-02-22 | 2022-11-04 | 北京京东尚科信息技术有限公司 | Network training and recognition method and device, electronic equipment and medium |
CN115372960A (en) * | 2022-07-11 | 2022-11-22 | 西北工业大学 | Sky wave radar ground and sea clutter data enhancement method for improving generation of countermeasure network |
-
2023
- 2023-03-31 CN CN202310334749.9A patent/CN116091362B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20200361083A1 (en) * | 2019-05-15 | 2020-11-19 | Nvidia Corporation | Grasp generation using a variational autoencoder |
US20220215262A1 (en) * | 2021-01-05 | 2022-07-07 | Capital One Services, Llc | Augmenting Datasets with Synthetic Data |
CN114170671A (en) * | 2021-09-16 | 2022-03-11 | 上海大学 | Massage manipulation identification method based on deep learning |
CN115294510A (en) * | 2022-02-22 | 2022-11-04 | 北京京东尚科信息技术有限公司 | Network training and recognition method and device, electronic equipment and medium |
CN115115028A (en) * | 2022-06-21 | 2022-09-27 | 国网湖北省电力有限公司信息通信公司 | Mixed data generation method based on generation countermeasure network |
CN115372960A (en) * | 2022-07-11 | 2022-11-22 | 西北工业大学 | Sky wave radar ground and sea clutter data enhancement method for improving generation of countermeasure network |
Non-Patent Citations (1)
Title |
---|
LIBAO ZHANG ET AL.: "Remote Sensing Image Generation Baed on Attention Mechanism and VAE-MSGAN for ROI Extraction", IEEE * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117710371A (en) * | 2024-02-05 | 2024-03-15 | 成都数之联科技股份有限公司 | Method, device, equipment and storage medium for expanding defect sample |
CN117710371B (en) * | 2024-02-05 | 2024-04-26 | 成都数之联科技股份有限公司 | Method, device, equipment and storage medium for expanding defect sample |
Also Published As
Publication number | Publication date |
---|---|
CN116091362B (en) | 2023-06-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Gai et al. | New image denoising algorithm via improved deep convolutional neural network with perceptive loss | |
CN112418027A (en) | Remote sensing image road extraction method for improving U-Net network | |
CN116091362B (en) | Remote sensing image data enhancement method based on variation self-encoder | |
CN111445476B (en) | Monocular depth estimation method based on multi-mode unsupervised image content decoupling | |
CN110570035B (en) | People flow prediction system for simultaneously modeling space-time dependency and daily flow dependency | |
CN113065649B (en) | Complex network topology graph representation learning method, prediction method and server | |
Deng et al. | Compressing explicit voxel grid representations: fast nerfs become also small | |
CN115619743A (en) | Construction method and application of OLED novel display device surface defect detection model | |
CN114359292A (en) | Medical image segmentation method based on multi-scale and attention | |
CN114663439A (en) | Remote sensing image land and sea segmentation method | |
CN115330620A (en) | Image defogging method based on cyclic generation countermeasure network | |
CN111627055A (en) | Scene depth completion method based on semantic segmentation | |
Wang et al. | Filter clustering for compressing cnn model with better feature diversity | |
CN108537132B (en) | Road segmentation method of depth automatic encoder based on supervised learning | |
CN114359902A (en) | Three-dimensional point cloud semantic segmentation method based on multi-scale feature fusion | |
CN113487115A (en) | Prediction method and system for steam flooding reservoir temperature field | |
Li et al. | Towards communication-efficient digital twin via ai-powered transmission and reconstruction | |
CN116912661A (en) | Target track prediction method and system with domain generalization capability | |
CN114331883A (en) | Point cloud completion method based on local covariance optimization | |
CN113255883B (en) | Weight initialization method based on power law distribution | |
CN113222016B (en) | Change detection method and device based on cross enhancement of high-level and low-level features | |
CN114092579B (en) | Point cloud compression method based on implicit neural network | |
CN112329799A (en) | Point cloud colorization algorithm | |
CN116109944B (en) | Satellite image cloud target extraction method based on deep learning network architecture | |
CN112991473B (en) | Neural network coding and decoding method and system based on cube template |
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 |