CN112906706A - Improved image semantic segmentation method based on coder-decoder - Google Patents

Improved image semantic segmentation method based on coder-decoder Download PDF

Info

Publication number
CN112906706A
CN112906706A CN202110344753.4A CN202110344753A CN112906706A CN 112906706 A CN112906706 A CN 112906706A CN 202110344753 A CN202110344753 A CN 202110344753A CN 112906706 A CN112906706 A CN 112906706A
Authority
CN
China
Prior art keywords
network
boundary
decoder
image
map
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202110344753.4A
Other languages
Chinese (zh)
Inventor
张红英
李鑫
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.)
Southwest University of Science and Technology
Original Assignee
Southwest University of Science and Technology
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 Southwest University of Science and Technology filed Critical Southwest University of Science and Technology
Priority to CN202110344753.4A priority Critical patent/CN112906706A/en
Publication of CN112906706A publication Critical patent/CN112906706A/en
Pending legal-status Critical Current

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
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • 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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/253—Fusion techniques of extracted features
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Landscapes

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

Abstract

The invention provides an improved image semantic segmentation method based on a coder and a decoder. Firstly, a cavity space pyramid pooling module in an improved encoder extracts image multi-scale features, and then a decoder performs feature map cross-layer fusion on extracted high-level and low-level semantic information; then, a visual activation function is used for improving the spatial context modeling capability of the codec network; and finally, introducing an optimization branch, generating an offset map containing different offset information of each pixel in each direction by using a boundary branch and a direction branch, and refining the coarse prediction result map generated by the decoder through coordinate mapping to generate a final semantic segmentation result map. The invention extracts and fuses the multi-scale features of the image by utilizing the coder and the decoder, refines the class boundary by using the offset map information, has excellent semantic segmentation performance and has wide applicability.

Description

Improved image semantic segmentation method based on coder-decoder
Technical Field
The invention relates to an image processing technology, in particular to an image semantic segmentation method for extracting and fusing multi-scale features and class boundary optimization by adopting a codec structure.
Background
The image semantic segmentation is used as a foundation technology different from target detection and image classification in a computer vision task, and a predefined label representing the semantic category of each pixel in an image is allocated to achieve a pixel-level classification task. Specifically, image semantic segmentation means that what a target object in an image is and where the target object is distinguished from the pixel level, that is, the target in the image is detected, then the outline between each individual and a scene is drawn, and finally the individual and the scene are classified and a color is given to the object belonging to the same class for representation. In recent years, with the development of deep learning technology in computer vision, image semantic segmentation is widely applied to aspects such as automatic driving and intelligent medical treatment. The inherent invariance of the deep neural network can learn the characteristics of dense abstraction, and the system performance is much better than that of a system designed according to the traditional sample characteristics.
The codec network aims to perform cross-layer fusion on feature map information of different scales acquired by a coder through a decoder, so that high-level semantic information and low-level spatial information are effectively fused, but a single codec structure easily causes the problems of loss of small-scale targets and boundary blurring in image segmentation, so that the acquisition of rich spatial context feature information and the optimization research of segmentation boundaries become the key points of image semantic segmentation.
Disclosure of Invention
The invention aims to solve the problem of image semantic segmentation, each pixel of an image is accurately classified through a deep learning network, and image semantic information can be segmented through the method.
In order to achieve the above object, the present invention provides an image semantic segmentation method using codec to extract and fuse multi-scale features and class boundary optimization, wherein the method mainly comprises five parts, the first part is to preprocess a data set; the second part is to perform feature extraction and cross-layer fusion on the input image; the third part is to carry out semantic rough segmentation on the image; the fourth part is the boundary optimization of the rough prediction graph; and the fifth part is network training and testing, and a final segmentation result graph is predicted.
The first part comprises two steps:
step 1, downloading a semantic segmentation public data set, and selecting images with complex scenes, various details and complete categories as training samples;
step 2, randomly zooming the training images in the range of [0.5, 2], then training images by random cutting, enhancing the randomness of training samples, preventing the problem of over-training fitting, and forming a final training set;
the second part comprises two steps:
and 3, inputting the training samples in the step 2 into a codec network for multi-scale feature extraction and cross-layer fusion to obtain a fused feature map. The specific implementation is as follows:
(1) the encoder network is used for feature extraction and multi-scale feature fusion and comprises a downsampling operation and an improved Spatial Pyramid Pooling (ASPP) module. And taking a residual error network as a backbone network, performing 1/4 down-sampling on the input samples to generate a low-level spatial feature map, transmitting the low-level spatial feature map into a decoder for standby, and taking a feature map with the size of 1/16 generated by continuous down-sampling as the input of an improved ASPP module to acquire high-level semantic information. Improved ASPP module in encoder
Figure 501034DEST_PATH_IMAGE002
Convolutional layer, four
Figure 214912DEST_PATH_IMAGE004
The expansion convolution layer (the expansion rates are respectively 4, 8, 12 and 24) and the average pooling layer, multi-scale feature extraction is carried out on the input feature map, nonlinear activation is carried out by using a FRELU activation function, and finally, Concatenate fusion is carried out;
(2) the decoder network is used for carrying out cross-layer fusion on different level features in the encoder;
step 4, inputting the training samples in the step 2 into a boundary optimization network, and extracting a high-resolution feature map as the input of boundary branches and direction branches through a parallel network HRNet;
the third part comprises two steps:
step 5, adjusting the channel number of the standby characteristic diagram of the decoder in the step 3, and performing concatemate fusion on the standby characteristic diagram and the improved ASPP module output characteristic diagram after deconvolution up-sampling operation;
step 6, mapping the feature map subjected to cross-layer fusion in the step 5 to an RGB space through convolution, and recovering to the resolution of the input image through deconvolution operation;
the fourth section comprises two steps:
and 7, taking the feature graph extracted in the step 4 as the input of the boundary branch and the direction branch, generating an offset graph with different offset information in each direction, and optimizing a rough result. The specific implementation is as follows:
(1) with boundary branches supervised by a binary cross-entropy function
Figure 73278DEST_PATH_IMAGE002
Convolution, BN normalization and ReLU activation function sum
Figure 61962DEST_PATH_IMAGE002
The linear classifier formed by convolution is formed, boundary division is carried out through a preset threshold value, all offsets are rescaled by artificial scaling factors, and false pixel prediction is reduced;
(2) the directional branches being supervised by a standard class cross entropy function
Figure 79597DEST_PATH_IMAGE002
Convolution, BN normalization and ReLU activation function sum
Figure 902059DEST_PATH_IMAGE002
A linear classifier formed by convolution is formed, and a real scene graph is divided by discrete partitions;
(3) masking the (2) output discrete directional diagram and the (1) output boundary diagram to generate an offset diagram with different offset information in each direction;
(4) mapping the output offset map space in the step (3) to the output rough segmentation map in the step 6 for boundary optimization;
the fifth part comprises two steps:
step 8, debugging the network structure hyper-parameters from step 3 to step 7, setting network model parameters, wherein the initial learning rate is set to 0.01, 1/10 initial learning rate is used in the backbone network, a poly learning rate adjustment strategy is used, Epochs is set to 80, Bach size is set to 8, and a final training model is obtained;
and 9, inputting the test set in the step 1 into the training model in the step 8, and segmenting image semantics.
The invention provides a codec image semantic segmentation method integrating multi-scale features and boundary optimization. Firstly, an ASPP module in an improved encoder extracts multi-scale features of an image and fuses the multi-scale features, then channel number adjustment is carried out on feature maps of different levels in a decoder, cross-layer fusion is carried out on high-level semantic information and low-level spatial information of which the channel number is adjusted, a rough semantic prediction result map is generated by mapping to an RGB space through convolution, and the efficiency of capturing spatial context is improved in a video activation function FRELU in the codec; and finally, performing boundary optimization on the rough prediction result by using a pixel offset map generated by a target boundary map and a discrete directional diagram mask in the optimization branch to generate a fine semantic segmentation result map. The invention utilizes the improved codec to fuse the multi-scale features of the image and optimizes the class boundary, thereby realizing excellent semantic segmentation performance, high accuracy and good robustness.
Drawings
FIG. 1 is a network overall framework diagram of the present invention;
FIG. 2 is a block diagram of an improved spatial pyramid pooling of the present invention;
FIG. 3 is an optimization schematic of the present invention;
FIG. 4 is an original input image;
FIG. 5 is a semantic segmentation image of FIG. 4 predicted using the present invention.
Detailed Description
For better understanding of the present invention, the codec image semantic segmentation method with multi-scale feature and boundary optimization according to the present invention is described in more detail below with reference to specific embodiments. In the following description, detailed descriptions of the current prior art, which will be omitted herein, may obscure the subject matter of the present invention.
Step 1, downloading a semantic segmentation public data set, and selecting a training sample with complex scene, various details and complete categories;
step 2, randomly zooming the training image in the range of [0.5, 2], then training the image by random cutting, enhancing the randomness of the training sample, preventing the problem of over-training fitting, and forming a final training set sample 101;
fig. 1 is a network model diagram of a codec image semantic segmentation method based on multi-scale feature and boundary optimization, which is performed according to the following steps in the present embodiment:
and 3, inputting the training samples in the step 2 into a codec network for multi-scale feature extraction and cross-layer fusion to obtain a fused feature map. The specific implementation is as follows:
(1) the encoder network is used for feature extraction and multi-scale feature fusion and consists of a downsampling operation and an improved ASPP module. And taking a residual error network as a backbone network, performing 1/4 down-sampling on the input samples to generate a low-level spatial feature map 102, transmitting the low-level spatial feature map into a decoder for standby, and taking a feature map 103 with the size of 1/16 generated by continuous down-sampling as the input of an improved ASPP module to acquire high-level semantic information. The improved ASPP module in the encoder is shown in fig. 2, and performs multi-scale feature extraction on an input feature map 201, which is obtained by
Figure 244792DEST_PATH_IMAGE002
Convolutional layer, four
Figure 37167DEST_PATH_IMAGE004
The expansion convolution layer (the expansion rates are respectively 4, 8, 12 and 24) and the average pooling layer form 202, the FRELU activation function is used for nonlinear activation, and finally the Concatenate fusion 203 is carried out;
(2) the decoder network is used for carrying out cross-layer fusion on different level features in the encoder;
step 4, inputting the training samples in the step 2 into a boundary optimization network, and extracting a high-resolution feature map as the input of boundary branches and direction branches through a parallel network HRNet;
the third part comprises two steps:
step 5, adjusting 104 the channel number of the standby characteristic diagram of the decoder in the step 3, and performing concatemate fusion 106 on the standby characteristic diagram of the decoder and an ASPP module output characteristic diagram 105 after deconvolution up-sampling operation;
step 6, mapping the feature map subjected to cross-layer fusion in the step 5 to an RGB space through convolution, and recovering to the resolution 107 of the input image through deconvolution operation;
the fourth section comprises two steps:
step 7, as shown in fig. 3, the optimization branch takes the feature map 301 extracted in step 4 as input of the boundary branch 302 and the direction branch 303, generates an offset map 305 having different offset information in each direction, and optimizes a rough result. The specific implementation is as follows:
(1) with boundary branches supervised by a binary cross-entropy function
Figure 909308DEST_PATH_IMAGE002
Convolution, BN normalization and ReLU activation function sum
Figure 902672DEST_PATH_IMAGE002
A linear classifier formed by convolution is formed 302, a preset threshold N =5 is used for dividing the boundary, and an artificial scaling factor a =2 is set for rescaling all offsets, so that the prediction of false pixels is reduced;
(2) the directional branches being supervised by a standard class cross entropy function
Figure 1209DEST_PATH_IMAGE002
Convolution, BN normalization and ReLU activation function sum
Figure 66117DEST_PATH_IMAGE002
A linear classifier component 303 formed by convolution, which divides the real scene graph by using discrete partition m = 8;
(3) masking (2) the output discrete pattern and (1) the output boundary map 304 to generate an offset map 305 having different offset amount information in each direction;
(4) mapping (3) the output offset map space to the output rough segmentation map 107 in the step 6 for boundary optimization 306;
Figure 58344DEST_PATH_IMAGE006
wherein the content of the first and second substances,
Figure 488188DEST_PATH_IMAGE008
is to refine the post-labelIn the figure, the figure shows that,
Figure 808442DEST_PATH_IMAGE010
representing the position of the boundary pixel i,
Figure 552407DEST_PATH_IMAGE012
an offset vector representing the generated intra pixels;
Figure 523774DEST_PATH_IMAGE014
representing the location of the identified internal pixel;
the fifth part comprises two steps:
step 8, debugging the network structure hyper-parameters from step 3 to step 7, setting network model parameters, wherein the initial learning rate is set to 0.01, 1/10 initial learning rate is used in the backbone network, a poly learning rate adjustment strategy is used, Epochs is set to 80, Bach size is set to 8, and a final training model is obtained;
step 9, inputting the test image into the pre-trained model, and predicting the semantic segmentation image 108 shown in fig. 5.
The invention provides a method for segmenting image semantics of a coder/decoder, which integrates multi-scale features and boundary optimization according to the structural characteristics of the coder/decoder and an image semantics segmenting method based on deep learning, and the method is characterized in that the cross-layer integration characteristics of the coder/decoder are more fully utilized, a space pyramid pooling module of a coder is improved to obtain the multi-scale image features, space insensitive information is activated through a visual activation function, high-level semantic information and low-level spatial information are integrated through the decoder, and then the resolution of a predicted image is restored through deconvolution; and finally, performing boundary pixel optimization on the generated rough prediction image by using the optimization branch to generate a final semantic segmentation prediction result image. The method has the advantages of simple algorithm, strong operability and wide applicability.
While the invention has been described with respect to the illustrative embodiments thereof, it is to be understood that the invention is not limited thereto but is intended to cover various changes and modifications which are obvious to those skilled in the art, and which are intended to be included within the spirit and scope of the invention as defined and defined in the appended claims.

Claims (4)

1. An improved image semantic segmentation method based on a coder-decoder is characterized in that a coder-decoder structure is adopted to extract and fuse multi-scale features and optimize class boundaries, and the method comprises five parts, namely data set preprocessing, feature extraction and cross-layer fusion of input images, semantic rough segmentation, boundary optimization, network training and testing;
the first part comprises two steps:
step 1, downloading a semantic segmentation public data set, and selecting images with complex scenes, various details and complete categories as training samples;
step 2, randomly zooming the training images in the range of [0.5, 2], then training images by random cutting, enhancing the randomness of training samples, preventing the problem of over-training fitting, and forming a final training set;
the second part comprises two steps:
and 3, inputting the training samples in the step 2 into a codec network for multi-scale feature extraction and cross-layer fusion to obtain a fused feature map. The specific implementation is as follows:
(1) the encoder network is used for feature extraction and multi-scale feature fusion and consists of a downsampling operation and an improved ASPP module; using a residual error network as a backbone network, performing 1/4 down-sampling on an input sample to generate a low-level spatial feature map, transmitting the low-level spatial feature map into a decoder for standby, and using a feature map with the size of 1/16 generated by continuous down-sampling as the input of an improved ASPP module to acquire high-level semantic information; improved ASPP module in encoder
Figure DEST_PATH_IMAGE001
Convolutional layer, four
Figure 405508DEST_PATH_IMAGE002
Expansion convolutional layer (expansion ratio of 4, 8, 12, 24, respectively) and globalThe method comprises the steps of average pooling layer composition, multi-scale feature extraction is conducted on an input feature graph, nonlinear activation is conducted through an FRELU activation function, and finally Concatenate fusion is conducted;
(2) the decoder network is used for carrying out cross-layer fusion on different level features in the encoder;
step 4, inputting the training samples in the step 2 into a boundary optimization network, and extracting a high-resolution feature map as the input of boundary branches and direction branches through a parallel network HRNet;
the third part comprises two steps:
step 5, adjusting the channel number of the standby characteristic diagram of the decoder in the step 3, and performing concatemate fusion on the characteristic diagram output by the improved ASPP module subjected to deconvolution up-sampling operation;
step 6, mapping the feature map subjected to cross-layer fusion in the step 5 to an RGB space through convolution, and recovering the feature map into the resolution of the input image through deconvolution operation;
the fourth section comprises two steps:
step 7, taking the characteristic graph extracted in the step 4 as the input of a boundary branch and a direction branch, generating an offset graph with different offset information in each direction, and optimizing a rough result; the specific implementation is as follows:
(1) with boundary branches supervised by a binary cross-entropy function
Figure 289150DEST_PATH_IMAGE001
Convolution, BN normalization and ReLU activation function sum
Figure 28436DEST_PATH_IMAGE001
The linear classifier formed by convolution is formed, boundary division is carried out through a preset threshold value, all offsets are rescaled by artificial scaling factors, and false pixel prediction is reduced;
(2) the directional branches being supervised by a standard class cross entropy function
Figure 466370DEST_PATH_IMAGE001
Convolution, BN normalization and ReLU excitationThe sum of the living functions
Figure 683725DEST_PATH_IMAGE001
A linear classifier formed by convolution is formed, and a real scene graph is divided by discrete partitions;
(3) masking the (2) output discrete directional diagram and the (1) output boundary diagram to generate an offset diagram with different offset information in each direction;
(4) mapping the output offset map space in the step (3) to the output rough segmentation map in the step 6 for boundary optimization;
the fifth part comprises two steps:
step 8, debugging the network structure hyper-parameters from step 3 to step 7, setting network model parameters, wherein the initial learning rate is set to 0.01, 1/10 initial learning rate is used in the backbone network, a poly learning rate adjustment strategy is used, Epochs is set to 80, Bach size is set to 8, and a final training model is obtained;
and 9, inputting the test set in the step 1 into the training model in the step 8, and segmenting image semantics.
2. The improved image semantic segmentation method based on codec as claimed in claim 1, characterized by using the ASPP block improved in step 3 (1), wherein the expansion ratio of the ASPP block is set to 4, 8, 12, 24; in the step 3, a visual activation function FRELU is used for acquiring space insensitive information; and 3, the codec structure is used for multi-scale extraction and cross-layer fusion of image features.
3. The improved codec-based image semantic segmentation method according to claim 1, wherein, by using step 7 (1), an input image boundary map is extracted, wherein a threshold is set to 5 and an artificial scaling factor is 2; extracting a discrete directional diagram by using the step 7 (2), wherein the discrete partition is set to be 8; in step 7 (3), the offset map is used for image segmentation optimization.
4. The method of claim 1, wherein in step 9, Epochs is set to 80, the backbone network learning rate is 1/10 of the initial learning rate, a poly learning rate adjustment strategy is used, and Bach size is set to 8.
CN202110344753.4A 2021-03-31 2021-03-31 Improved image semantic segmentation method based on coder-decoder Pending CN112906706A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110344753.4A CN112906706A (en) 2021-03-31 2021-03-31 Improved image semantic segmentation method based on coder-decoder

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110344753.4A CN112906706A (en) 2021-03-31 2021-03-31 Improved image semantic segmentation method based on coder-decoder

Publications (1)

Publication Number Publication Date
CN112906706A true CN112906706A (en) 2021-06-04

Family

ID=76109601

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110344753.4A Pending CN112906706A (en) 2021-03-31 2021-03-31 Improved image semantic segmentation method based on coder-decoder

Country Status (1)

Country Link
CN (1) CN112906706A (en)

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113516640A (en) * 2021-07-05 2021-10-19 首都师范大学 CT image fine crack segmentation device and method based on classification branches
CN113705463A (en) * 2021-08-30 2021-11-26 安徽大学 Factory footprint extraction method and system based on multi-scale gating dense connection
CN113971660A (en) * 2021-09-30 2022-01-25 哈尔滨工业大学 Computer vision method for bridge health diagnosis and intelligent camera system
CN114037674A (en) * 2021-11-04 2022-02-11 天津大学 Industrial defect image segmentation detection method and device based on semantic context
CN114419036A (en) * 2022-03-28 2022-04-29 北京矩视智能科技有限公司 Surface defect region segmentation method and device based on boundary information fusion
CN114648668A (en) * 2022-05-18 2022-06-21 浙江大华技术股份有限公司 Method and apparatus for classifying attributes of target object, and computer-readable storage medium
CN116661530A (en) * 2023-07-31 2023-08-29 山西聚源生物科技有限公司 Intelligent control system and method in edible fungus industrial cultivation
CN117237644A (en) * 2023-11-10 2023-12-15 广东工业大学 Forest residual fire detection method and system based on infrared small target detection
CN117409329A (en) * 2023-12-15 2024-01-16 深圳安德空间技术有限公司 Method and system for reducing false alarm rate of underground cavity detection by three-dimensional ground penetrating radar

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107644426A (en) * 2017-10-12 2018-01-30 中国科学技术大学 Image, semantic dividing method based on pyramid pond encoding and decoding structure
CN112183635A (en) * 2020-09-29 2021-01-05 南京农业大学 Method for realizing segmentation and identification of plant leaf lesions by multi-scale deconvolution network

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107644426A (en) * 2017-10-12 2018-01-30 中国科学技术大学 Image, semantic dividing method based on pyramid pond encoding and decoding structure
CN112183635A (en) * 2020-09-29 2021-01-05 南京农业大学 Method for realizing segmentation and identification of plant leaf lesions by multi-scale deconvolution network

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
YUHUI YUAN等: ""SegFix: Model-Agnostic Boundary Refinement for Segmentation"", 《ARXIV》 *
YUKUN ZHU等: ""Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation"", 《ARXIV》 *

Cited By (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113516640A (en) * 2021-07-05 2021-10-19 首都师范大学 CT image fine crack segmentation device and method based on classification branches
CN113705463A (en) * 2021-08-30 2021-11-26 安徽大学 Factory footprint extraction method and system based on multi-scale gating dense connection
CN113705463B (en) * 2021-08-30 2024-02-20 安徽大学 Factory footprint extraction method and system based on multi-scale gate control intensive connection
CN113971660A (en) * 2021-09-30 2022-01-25 哈尔滨工业大学 Computer vision method for bridge health diagnosis and intelligent camera system
CN114037674A (en) * 2021-11-04 2022-02-11 天津大学 Industrial defect image segmentation detection method and device based on semantic context
CN114037674B (en) * 2021-11-04 2024-04-26 天津大学 Industrial defect image segmentation detection method and device based on semantic context
CN114419036B (en) * 2022-03-28 2022-06-24 北京矩视智能科技有限公司 Surface defect region segmentation method and device based on boundary information fusion
CN114419036A (en) * 2022-03-28 2022-04-29 北京矩视智能科技有限公司 Surface defect region segmentation method and device based on boundary information fusion
CN114648668A (en) * 2022-05-18 2022-06-21 浙江大华技术股份有限公司 Method and apparatus for classifying attributes of target object, and computer-readable storage medium
CN116661530A (en) * 2023-07-31 2023-08-29 山西聚源生物科技有限公司 Intelligent control system and method in edible fungus industrial cultivation
CN116661530B (en) * 2023-07-31 2023-09-29 山西聚源生物科技有限公司 Intelligent control system and method in edible fungus industrial cultivation
CN117237644A (en) * 2023-11-10 2023-12-15 广东工业大学 Forest residual fire detection method and system based on infrared small target detection
CN117237644B (en) * 2023-11-10 2024-02-13 广东工业大学 Forest residual fire detection method and system based on infrared small target detection
CN117409329A (en) * 2023-12-15 2024-01-16 深圳安德空间技术有限公司 Method and system for reducing false alarm rate of underground cavity detection by three-dimensional ground penetrating radar
CN117409329B (en) * 2023-12-15 2024-04-05 深圳安德空间技术有限公司 Method and system for reducing false alarm rate of underground cavity detection by three-dimensional ground penetrating radar

Similar Documents

Publication Publication Date Title
CN112906706A (en) Improved image semantic segmentation method based on coder-decoder
CN111047551B (en) Remote sensing image change detection method and system based on U-net improved algorithm
CN111932553B (en) Remote sensing image semantic segmentation method based on area description self-attention mechanism
CN111931684B (en) Weak and small target detection method based on video satellite data identification features
CN111612807B (en) Small target image segmentation method based on scale and edge information
CN112396607B (en) Deformable convolution fusion enhanced street view image semantic segmentation method
CN109753913B (en) Multi-mode video semantic segmentation method with high calculation efficiency
Chen et al. Vehicle detection in high-resolution aerial images based on fast sparse representation classification and multiorder feature
CN112232349A (en) Model training method, image segmentation method and device
Kim et al. Multi-task convolutional neural network system for license plate recognition
CN111046880A (en) Infrared target image segmentation method and system, electronic device and storage medium
CN110929593A (en) Real-time significance pedestrian detection method based on detail distinguishing and distinguishing
CN110263786A (en) A kind of road multi-targets recognition system and method based on characteristic dimension fusion
CN112950477A (en) High-resolution saliency target detection method based on dual-path processing
CN114037640A (en) Image generation method and device
Zang et al. Traffic lane detection using fully convolutional neural network
CN116052016A (en) Fine segmentation detection method for remote sensing image cloud and cloud shadow based on deep learning
CN114913498A (en) Parallel multi-scale feature aggregation lane line detection method based on key point estimation
CN114155371A (en) Semantic segmentation method based on channel attention and pyramid convolution fusion
Han et al. A method based on multi-convolution layers joint and generative adversarial networks for vehicle detection
CN115861756A (en) Earth background small target identification method based on cascade combination network
Sulehria et al. Vehicle number plate recognition using mathematical morphology and neural networks
CN116012395A (en) Multi-scale fusion smoke segmentation method based on depth separable convolution
CN112149526A (en) Lane line detection method and system based on long-distance information fusion
CN117079163A (en) Aerial image small target detection method based on improved YOLOX-S

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
WD01 Invention patent application deemed withdrawn after publication
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20210604