CN112580645A - Unet semantic segmentation method based on convolutional sparse coding - Google Patents

Unet semantic segmentation method based on convolutional sparse coding Download PDF

Info

Publication number
CN112580645A
CN112580645A CN202011445030.5A CN202011445030A CN112580645A CN 112580645 A CN112580645 A CN 112580645A CN 202011445030 A CN202011445030 A CN 202011445030A CN 112580645 A CN112580645 A CN 112580645A
Authority
CN
China
Prior art keywords
unet
model
csc
network
segmentation
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
Application number
CN202011445030.5A
Other languages
Chinese (zh)
Other versions
CN112580645B (en
Inventor
王倪传
唐海童
卢霞
颜虹杰
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Jiangsu Ocean University
Original Assignee
Jiangsu Ocean University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Jiangsu Ocean University filed Critical Jiangsu Ocean University
Priority to CN202011445030.5A priority Critical patent/CN112580645B/en
Publication of CN112580645A publication Critical patent/CN112580645A/en
Application granted granted Critical
Publication of CN112580645B publication Critical patent/CN112580645B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
    • G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • 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
    • G06T9/00—Image coding
    • G06T9/002—Image coding using neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
    • G06V10/467—Encoded features or binary features, e.g. local binary patterns [LBP]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/513—Sparse representations
    • 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
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T10/00—Road transport of goods or passengers
    • Y02T10/10—Internal combustion engine [ICE] based vehicles
    • Y02T10/40—Engine management systems

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Multimedia (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Mathematical Physics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Software Systems (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a Unet semantic segmentation method based on convolution sparse coding, which combines the convolution sparse coding with a coding network in an Unet model to form a coder of a CSC-Unet model so as to obtain the global information of an image; combining the convolution sparse coding with a decoding network in Unet to form a decoder of a CSC-Unet model so as to obtain the position information of the image; and use of a hopping structure allows global information to be combined with positional information to produce accurate and fine segmentation. The method firstly preprocesses the training pictures and labels in the data set, such as cutting, data enhancement and the like, and then reads in the CSC-Unet segmentation model for training. After training is finished, the test samples and labels in the data set are read into the CSC-Unet segmentation model, and the stored best weight values are loaded into the model, so that the purpose of accurate semantic segmentation by using the model is achieved.

Description

Unet semantic segmentation method based on convolutional sparse coding
Technical Field
The invention relates to the field of semantic segmentation of images, in particular to a Unet semantic segmentation method based on convolutional sparse coding.
Background
Over the past decades, sparse and redundant representation areas have made a significant leap. It grows into a mature discipline with great influence. The sparse domain brings the idea that natural signals can be described as a linear combination consisting of only a few members or components (often called atoms). Sparse domain models are increasingly becoming the core of signal and image processing and machine learning applications, producing the most advanced results in a variety of tasks and many different domains. In recent years, the offspring model Convolutional Sparse Coding (CSC) in the sparse domain and the multi-layered convolutional sparse coding (ML-CSC) have achieved significant results in many different domains ranging from signal and image processing to machine learning. On the other hand, semantic segmentation is one of the key tasks in computer vision. In reality, more and more application scenes need to infer relevant knowledge or semantics (i.e., concrete to abstract processes) from imagery. As a core problem of computer vision, semantic segmentation is increasingly important for scene understanding. Based on the above, the invention provides the Unet semantic segmentation method based on the convolution sparse coding, which can better capture the semantic information and the representation information of the image so as to more accurately perform semantic segmentation on the image.
Disclosure of Invention
The invention aims to obtain a model with more accurate segmentation effect by combining convolutional sparse coding and a Unet segmentation model based on the obvious effect of the convolutional sparse coding in many different fields.
In order to achieve the above object, the present invention provides a method for performing semantic segmentation on a Unet based on convolutional sparse coding, the method comprising the following steps:
s1: reading training samples and labels in a data set into a CSC-Unet semantic segmentation network, and preprocessing training pictures and labels according to actual needs, such as cutting and normalization;
s2: combining the convolution sparse coding with a coding network in Unet to form a coder of a CSC-Unet model so as to obtain global information of the image;
s3: combining the convolutional sparse coding with a decoding network in Unet to form a decoder of a CSC-Unet model so as to obtain the position information of the image, and using a hopping structure to enable the global information to be combined with the position information so as to generate accurate and fine segmentation;
s4: and carrying out post-processing on a result obtained by the CSC-Unet model to obtain a visualized semantic segmentation map.
As a preferred technical solution of the present invention, the S1 and S4 are respectively preprocessing and post-processing of the image, and the S2 and S3 are methods for proposing a CSC-Unet semantic segmentation based on convolutional sparse coding and Unet segmentation model formation.
As a preferred embodiment of the present invention, the step S1: the data preprocessing part can process the data by the following steps:
s1.1, preprocessing data, such as normalization, standardization, cutting and data enhancement, is beneficial to training of a deep network, accelerates a convergence process, avoids an overfitting problem and enhances the generalization capability of a model;
s1.2, preprocessing data and reading the data into a network: and reading the training samples and the labels in the data set into the convolutional neural network according to the size of the batch-size.
As a preferred embodiment of the present invention, the step S2: the design process of the encoder of the CSC-Unet model can be composed of the following steps:
s2.1, designing a two-layer convolution sparse coding model network model to form an ML-CSC module: let the original signal X satisfy two-layer convolution sparse model, which can be expressed as X ═ D1Γ1,Γ1＝D2Γ2Wherein
Figure BDA0002824083010000021
Figure BDA0002824083010000022
S2.2, solving of ML-CSC problem: how to find gamma1And Γ2Can be seen as a Depth Coding Problem (DCP): in | | | Y-D1Γ1||2≤ε,
Figure BDA0002824083010000023
Γ1＝D2Γ2,
Figure BDA0002824083010000024
Solving under the condition
Figure BDA0002824083010000025
Where Y is the original signal X mixed with the noise E, i.e., Y ═ X + E (| E | | non-calculation circuit)2Epsilon is less than or equal to epsilon). Solving the depth coding problem with a layered basis pursuit algorithm (LBP) can yield:
Figure BDA0002824083010000031
wherein
Figure BDA0002824083010000032
S2.3, solving the LBP problem: an approximate solution to the LBP problem can be found using a multi-layered iterative soft threshold algorithm (ML-ISTA):
Figure BDA0002824083010000033
where T is the number of iterations, TλIs a threshold operator; if we further assume that the representation coefficients are non-negative, the approximate solution can be written as:
Figure BDA0002824083010000034
wherein
Figure BDA0002824083010000035
WkFor the convolution operation, when t is 0, i.e. there is no iteration number, the ML-CSC module is equivalent to two convolution operations, and the convolution coefficients are:W1and W2When the iteration time t is 1,
Figure BDA0002824083010000036
the number of learnable parameters is not increased in the iterative process;
s2.4, combining the ML-CSC module with a coding end of a traditional Unet segmentation network.
As a preferred embodiment of the present invention, the step S3: the design process of the encoder of the CSC-Unet model and the jump connection can be composed of the following steps:
s3.1, combining the ML-CSC module with a decoding end of a traditional Unet segmentation network;
s3.2, using a jump structure to enable the global information to be combined with the position information, so as to generate accurate and fine segmentation;
and S3.3, selecting NLL _ LOSS as a LOSS function, activating a log-SoftMax function for input parameters, and selecting an Adam optimization function which is an optimization method of self-adaptive learning rate as an optimization function for optimization.
As a preferred technical solution of the present invention, in S4, a result obtained by the CSC-Unet model is post-processed to obtain a visualized accurate semantic segmentation map; the process can be composed of the following steps:
s4.1, performing precision operation on the result output by the model and the real label to obtain a confusion matrix so as to obtain corresponding index measurement parameters such as average intersection ratio (Miou), pixel precision, average pixel precision and the like, and further measure the network performance;
and S4.2, storing the prediction result of the model in a picture form, and visually feeling the accuracy of the segmentation.
The invention has the beneficial effects that: compared with the segmentation method in the prior art, the method has the advantages that the global information of the original image can be better captured at the encoding end, the position information of the original image can be better captured at the decoding end, and finally the captured global information and the position information are combined to obtain more accurate semantic segmentation of the image.
Drawings
FIG. 1 is a simplified flowchart of the Unet semantic segmentation method based on convolutional sparse coding according to the present invention;
fig. 2-3 are models of the Unet semantic segmentation method based on convolutional sparse coding of the present invention.
Detailed Description
The following detailed description of the preferred embodiments of the present invention, taken in conjunction with the accompanying drawings, will make the advantages and features of the invention more readily understood by those skilled in the art, and thus will more clearly and distinctly define the scope of the invention.
Example (b): referring to fig. 1-3, the present invention provides a technical solution: the following further describes specific embodiments of the present invention with reference to the drawings.
As shown in fig. 1, the method for performing semantic segmentation on a pnet based on convolutional sparse coding of the present invention includes the following steps:
s1: the data preprocessing part can process the data by the following steps:
s1.1, preprocessing the data, such as normalization, standardization, cutting, data enhancement and the like. This facilitates deep network training, speeds up the convergence process, while also avoiding the over-fitting problem and enhancing the generalization capability of the model.
S1.2, preprocessing data and reading the data into a network: and reading the training samples and the labels in the data set into the convolutional neural network according to the size of the batch-size.
S2: the design process of the encoder of the CSC-Unet model can be composed of the following steps:
s2.1, designing a two-layer convolution sparse coding model network model to form an ML-CSC module: let the original signal X satisfy two-layer convolution sparse model, which can be expressed as X ═ D1Γ1,Γ1＝D2Γ2Wherein
Figure BDA0002824083010000041
Figure BDA0002824083010000051
S2.2, solving of ML-CSC problem: how to find gamma1And Γ2Can be seen as a Depth Coding Problem (DCP): in | | | Y-D1Γ1||2≤ε,
Figure BDA0002824083010000052
Γ1＝D2Γ2,
Figure BDA0002824083010000053
Solving under the condition
Figure BDA0002824083010000054
Where Y is the original signal X mixed with the noise E, i.e., Y ═ X + E (| E | | non-calculation circuit)2Epsilon is less than or equal to epsilon). Solving the depth coding problem with a layered basis pursuit algorithm (LBP) can yield:
Figure BDA0002824083010000055
wherein
Figure BDA0002824083010000056
S2.3, solving the LBP problem: an approximate solution to the LBP problem can be found using a multi-layered iterative soft threshold algorithm (ML-ISTA):
Figure BDA0002824083010000057
where T is the number of iterations, TλIs a threshold operator. If we further assume that the representation coefficients are non-negative, the approximate solution can be written as:
Figure BDA0002824083010000058
wherein
Figure BDA0002824083010000059
WkFor convolution operation, when t is 0, i.e. there are no iterations, the ML-CSC module is equivalent to two convolution operationsThe convolution coefficients are: w1And W2When the iteration time t is 1,
Figure BDA00028240830100000510
as shown in fig. 2, and does not increase the number of learnable parameters in the iterative process.
S2.4, combining the ML-CSC module with the coding end of the traditional Unet segmentation network, as shown in figure 3.
S3: the design process of the encoder of the CSC-Unet model and the jump connection can be composed of the following steps:
s3.1, combining the ML-CSC module with a decoding end of a traditional Unet segmentation network, as shown in figure 3;
s3.2, using a jump structure to enable the global information to be combined with the position information, so as to generate accurate and fine segmentation;
and S3.3, selecting NLL _ LOSS as a LOSS function, activating a log-SoftMax function for input parameters, and selecting an Adam optimization function which is an optimization method of self-adaptive learning rate as an optimization function for optimization.
S4: and carrying out post-processing on a result obtained by the CSC-Unet model to obtain a visualized accurate semantic segmentation map. The process can be composed of the following steps:
s4.1, performing precision operation on the result output by the model and the real label to obtain a confusion matrix so as to obtain corresponding index measurement parameters such as average intersection ratio (Miou), pixel precision, average pixel precision and the like, and further measure the network performance;
and S4.2, storing the prediction result of the model in a picture form, and visually feeling the accuracy of the segmentation.
The method comprises the steps of preprocessing training pictures and labels in a data set, such as cutting and data enhancement, reading in a CSC-Unet segmentation model for training, reading test samples and labels in the data set into a convolutional neural network after all training is finished, and loading the stored best weight values into the model, so that the aim of accurate semantic segmentation is fulfilled. The invention is very beneficial to the research of geographic information systems, unmanned vehicle driving, medical image analysis, robots, image search engines and the like.
The above-mentioned embodiments only express several embodiments of the present invention, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the inventive concept, which falls within the scope of the present invention.

Claims (6)

1. The Unet semantic segmentation method based on convolutional sparse coding is characterized by comprising the following steps of:
s1: reading training samples and labels in a data set into a CSC-Unet semantic segmentation network, and preprocessing training pictures and labels according to actual needs, such as cutting and normalization;
s2: combining the convolution sparse coding with a coding network in Unet to form a coder of a CSC-Unet model so as to obtain global information of the image;
s3: combining the convolutional sparse coding with a decoding network in Unet to form a decoder of a CSC-Unet model so as to obtain the position information of the image, and using a hopping structure to enable the global information to be combined with the position information so as to generate accurate and fine segmentation;
s4: and carrying out post-processing on a result obtained by the CSC-Unet model to obtain a visualized semantic segmentation map.
2. The method for semantic segmentation of the Unet network based on convolutional sparse coding as claimed in claim 1, wherein: the S1 and S4 are respectively preprocessing and post-processing of the image, and the S2 and S3 are methods for proposing CSC-Unet semantic segmentation based on convolutional sparse coding and Unet segmentation model formation.
3. The method for semantic segmentation of the Unet network based on convolutional sparse coding as claimed in claim 1, wherein: the S1: the data preprocessing part can process the data by the following steps:
s1.1, preprocessing data, such as normalization, standardization, cutting and data enhancement, is beneficial to training of a deep network, accelerates a convergence process, avoids an overfitting problem and enhances the generalization capability of a model;
s1.2, preprocessing data and reading the data into a network: and reading the training samples and the labels in the data set into the convolutional neural network according to the size of the batch-size.
4. The method for semantic segmentation of the Unet network based on convolutional sparse coding as claimed in claim 1, wherein: the S2: the design process of the encoder of the CSC-Unet model can be composed of the following steps:
s2.1, designing a two-layer convolution sparse coding model network model to form an ML-CSC module: let the original signal X satisfy two-layer convolution sparse model, which can be expressed as X ═ D1Γ1,Γ1＝D2Γ2Wherein
Figure FDA0002824083000000021
Figure FDA0002824083000000022
S2.2, solving of ML-CSC problem: how to find gamma1And Γ2Can be seen as a Depth Coding Problem (DCP): in | | | Y-D1Γ1||2≤ε,
Figure FDA0002824083000000023
Γ1＝D2Γ2,
Figure FDA0002824083000000024
Solving under the condition
Figure FDA0002824083000000025
Where Y is the original signal X mixed with noise E, i.e. Y ═X+E(||E||2Epsilon is less than or equal to epsilon). Solving the depth coding problem with a layered basis pursuit algorithm (LBP) can yield:
Figure FDA0002824083000000026
wherein
Figure FDA0002824083000000027
S2.3, solving the LBP problem: an approximate solution to the LBP problem can be found using a multi-layered iterative soft threshold algorithm (ML-ISTA):
Figure FDA0002824083000000028
where T is the number of iterations, TλIs a threshold operator; if we further assume that the representation coefficients are non-negative, the approximate solution can be written as:
Figure FDA0002824083000000029
wherein
Figure FDA00028240830000000210
WkFor the convolution operation, when t is 0, i.e. there is no iteration number, the ML-CSC module is equivalent to two convolution operations, and the convolution coefficients are: w1And W2When the iteration time t is 1,
Figure FDA00028240830000000211
the number of learnable parameters is not increased in the iterative process;
s2.4, combining the ML-CSC module with a coding end of a traditional Unet segmentation network.
5. The method for semantic segmentation of the Unet network based on convolutional sparse coding as claimed in claim 1, wherein: the S3: the design process of the encoder of the CSC-Unet model and the jump connection can be composed of the following steps:
s3.1, combining the ML-CSC module with a decoding end of a traditional Unet segmentation network;
s3.2, using a jump structure to enable the global information to be combined with the position information, so as to generate accurate and fine segmentation;
and S3.3, selecting NLL _ LOSS as a LOSS function, activating a log-SoftMax function for input parameters, and selecting an Adam optimization function which is an optimization method of self-adaptive learning rate as an optimization function for optimization.
6. The method for semantic segmentation of the Unet network based on convolutional sparse coding as claimed in claim 1, wherein: in the S4, post-processing a result obtained by the CSC-Unet model to obtain a visualized accurate semantic segmentation map; the process can be composed of the following steps:
s4.1, performing precision operation on the result output by the model and the real label to obtain a confusion matrix so as to obtain corresponding index measurement parameters such as average intersection ratio (Miou), pixel precision, average pixel precision and the like, and further measure the network performance;
and S4.2, storing the prediction result of the model in a picture form, and visually feeling the accuracy of the segmentation.
CN202011445030.5A 2020-12-08 2020-12-08 Unet semantic segmentation method based on convolution sparse coding Active CN112580645B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011445030.5A CN112580645B (en) 2020-12-08 2020-12-08 Unet semantic segmentation method based on convolution sparse coding

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011445030.5A CN112580645B (en) 2020-12-08 2020-12-08 Unet semantic segmentation method based on convolution sparse coding

Publications (2)

Publication Number Publication Date
CN112580645A true CN112580645A (en) 2021-03-30
CN112580645B CN112580645B (en) 2024-05-03

Family

ID=75130881

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011445030.5A Active CN112580645B (en) 2020-12-08 2020-12-08 Unet semantic segmentation method based on convolution sparse coding

Country Status (1)

Country Link
CN (1) CN112580645B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113344811A (en) * 2021-05-31 2021-09-03 西南大学 Multilayer convolution sparse coding weighted recursive denoising deep neural network and method
CN113409322A (en) * 2021-06-18 2021-09-17 中国石油大学(华东) Deep learning training sample enhancement method for semantic segmentation of remote sensing image

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109670392A (en) * 2018-09-04 2019-04-23 中国人民解放军陆军工程大学 Road image semantic segmentation method based on hybrid automatic encoder
CN111210435A (en) * 2019-12-24 2020-05-29 重庆邮电大学 Image semantic segmentation method based on local and global feature enhancement module
WO2020215236A1 (en) * 2019-04-24 2020-10-29 哈尔滨工业大学（深圳） Image semantic segmentation method and system
CN111951249A (en) * 2020-08-13 2020-11-17 浙江理工大学 Mobile phone light guide plate defect visual detection method based on multitask learning network

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109670392A (en) * 2018-09-04 2019-04-23 中国人民解放军陆军工程大学 Road image semantic segmentation method based on hybrid automatic encoder
WO2020215236A1 (en) * 2019-04-24 2020-10-29 哈尔滨工业大学（深圳） Image semantic segmentation method and system
CN111210435A (en) * 2019-12-24 2020-05-29 重庆邮电大学 Image semantic segmentation method based on local and global feature enhancement module
CN111951249A (en) * 2020-08-13 2020-11-17 浙江理工大学 Mobile phone light guide plate defect visual detection method based on multitask learning network

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
郭爱心;殷保群;李运;: "基于深度卷积神经网络的小尺度行人检测", 信息技术与网络安全, no. 07 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113344811A (en) * 2021-05-31 2021-09-03 西南大学 Multilayer convolution sparse coding weighted recursive denoising deep neural network and method
CN113409322A (en) * 2021-06-18 2021-09-17 中国石油大学(华东) Deep learning training sample enhancement method for semantic segmentation of remote sensing image

Also Published As

Publication number Publication date
CN112580645B (en) 2024-05-03

Similar Documents

Publication Publication Date Title
CN113850825B (en) Remote sensing image road segmentation method based on context information and multi-scale feature fusion
CN109919204B (en) Noise image-oriented deep learning clustering method
CN111325664B (en) Style migration method and device, storage medium and electronic equipment
CN110689599A (en) 3D visual saliency prediction method for generating countermeasure network based on non-local enhancement
CN112634296A (en) RGB-D image semantic segmentation method and terminal for guiding edge information distillation through door mechanism
CN116721334B (en) Training method, device, equipment and storage medium of image generation model
CN112580645A (en) Unet semantic segmentation method based on convolutional sparse coding
CN113870335A (en) Monocular depth estimation method based on multi-scale feature fusion
CN112801047B (en) Defect detection method and device, electronic equipment and readable storage medium
CN115908772A (en) Target detection method and system based on Transformer and fusion attention mechanism
CN117218498B (en) Multi-modal large language model training method and system based on multi-modal encoder
CN110633706B (en) Semantic segmentation method based on pyramid network
CN114283352A (en) Video semantic segmentation device, training method and video semantic segmentation method
CN112053338A (en) Image decomposition method and related device and equipment
Sharma et al. Alchemist: Parametric control of material properties with diffusion models
CN116958712A (en) Image generation method, system, medium and device based on prior probability distribution
CN112418229A (en) Unmanned ship marine scene image real-time segmentation method based on deep learning
Di et al. FDNet: An end-to-end fusion decomposition network for infrared and visible images
CN117011219A (en) Method, apparatus, device, storage medium and program product for detecting quality of article
CN114565625A (en) Mineral image segmentation method and device based on global features
CN113936040A (en) Target tracking method based on capsule network and natural language query
CN117095136B (en) Multi-object and multi-attribute image reconstruction and editing method based on 3D GAN
Liu et al. Deep multi-scale network for single image dehazing with self-guided maps
CN117635941A (en) Remote sensing image semantic segmentation method based on multi-scale feature and global information modeling
CN117876858A (en) Underwater target detection method based on improved yolov7 algorithm

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