CN117115442A - Semantic segmentation method based on visible light-infrared photoelectric reconnaissance image fusion - Google Patents

Semantic segmentation method based on visible light-infrared photoelectric reconnaissance image fusion Download PDF

Info

Publication number
CN117115442A
CN117115442A CN202311037515.4A CN202311037515A CN117115442A CN 117115442 A CN117115442 A CN 117115442A CN 202311037515 A CN202311037515 A CN 202311037515A CN 117115442 A CN117115442 A CN 117115442A
Authority
CN
China
Prior art keywords
convolution
reconnaissance image
infrared photoelectric
feature
semantic 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.)
Pending
Application number
CN202311037515.4A
Other languages
Chinese (zh)
Inventor
程文明
陈国强
魏振兴
张国财
麻斌鑫
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.)
Zhejiang Aerospace Runbo Measurement And Control Technology Co ltd
Original Assignee
Zhejiang Aerospace Runbo Measurement And Control Technology Co ltd
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 Zhejiang Aerospace Runbo Measurement And Control Technology Co ltd filed Critical Zhejiang Aerospace Runbo Measurement And Control Technology Co ltd
Priority to CN202311037515.4A priority Critical patent/CN117115442A/en
Publication of CN117115442A publication Critical patent/CN117115442A/en
Pending legal-status Critical Current

Links

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
    • 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
    • G06N3/0455—Auto-encoder networks; Encoder-decoder 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • 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
    • G06N3/048—Activation functions
    • 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
    • G06V10/42—Global feature extraction by analysis of the whole pattern, e.g. using frequency domain transformations or autocorrelation
    • 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
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
    • G06V10/449—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters
    • G06V10/451—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters with interaction between the filter responses, e.g. cortical complex cells
    • G06V10/454—Integrating the filters into a hierarchical structure, e.g. convolutional neural networks [CNN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
    • G06V10/806—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of extracted 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
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks

Landscapes

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

Abstract

The invention discloses a semantic segmentation method based on visible light-infrared photoelectric reconnaissance image fusion, which comprises the following steps of: s1, acquiring an infrared photoelectric pod reconnaissance image and a visible photoelectric pod reconnaissance image; s2, respectively extracting features in an infrared photoelectric pod reconnaissance image and a visible photoelectric pod reconnaissance image by using a convNeXT feature extractor; s3, fusing the extracted features through a differential feature fusion module; and S4, performing up-sampling operation one by one through a decoder to obtain a semantic segmentation graph. The invention has the characteristics of effectively improving the segmentation accuracy and reducing the false detection and missing detection phenomena caused by interference factors.

Description

Semantic segmentation method based on visible light-infrared photoelectric reconnaissance image fusion
Technical Field
The invention relates to the field of remote sensing image processing, in particular to a semantic segmentation method based on visible light-infrared photoelectric reconnaissance image fusion.
Background
With the development of deep learning technology, many effective semantic segmentation networks have emerged in recent years. However, the mainstream semantic segmentation network mainly adopts a photoelectric scout image of visible light. When the illumination condition is not satisfied, the quality of the visible light image is liable to be degraded, and the segmentation performance is liable to be degraded. For example, most algorithms fail to segment objects correctly in near complete darkness. Therefore, there is a need to develop a semantic segmentation method based on visible light-infrared photoelectric reconnaissance image fusion to reduce false detection and missing detection phenomena caused by interference factors and improve segmentation accuracy.
Disclosure of Invention
The invention aims to provide a semantic segmentation method based on visible light-infrared photoelectric reconnaissance image fusion. The invention has the characteristics of effectively improving the segmentation accuracy and reducing the false detection and missing detection phenomena caused by interference factors.
The technical scheme of the invention is as follows: a semantic segmentation method based on visible light-infrared photoelectric reconnaissance image fusion comprises the following steps:
s1, acquiring an infrared photoelectric pod reconnaissance image and a visible photoelectric pod reconnaissance image;
s2, respectively extracting features in an infrared photoelectric pod reconnaissance image and a visible photoelectric pod reconnaissance image by using a convNeXT feature extractor;
s3, fusing the extracted features through a differential feature fusion module;
and S4, performing up-sampling operation one by one through a decoder to obtain a semantic segmentation graph.
In the semantic segmentation method based on visible light-infrared photoelectric reconnaissance image fusion, the convNeXT feature extractor processes the image by adopting four ConvNeXt convolution blocks in sequence to extract features with different scales;
the ConvNeXt convolution block connection mode adopts a reverse bottleneck structure, and the separable convolution of the ConvNeXt convolution block connection mode is moved up to a first layer;
the ConvNeXt convolution block is used in the following steps: firstly, carrying out convolution operation on an image by using 7 multiplied by 7 convolution check, and then, carrying out linear regularization treatment and twice 1 multiplied by 1 convolution operation to finish feature extraction; the gel activation function is used to operate between two 1 x 1 convolution operations.
In the semantic segmentation method based on visible light-infrared photoelectric reconnaissance image fusion, the fusion calculation formula of the differential feature fusion module is as follows:
wherein OUT is the fused output characteristic, M is the characteristic fusion weight, X is the characteristic vector obtained by extracting the characteristic of the infrared photoelectric pod reconnaissance image, and Y is the characteristic vector obtained by extracting the characteristic of the visible photoelectric pod reconnaissance image.
In the semantic segmentation method based on visible light-infrared photoelectric reconnaissance image fusion, the calculation process of the feature fusion weight M is as follows:
calculating the channel attention of the local feature, wherein the calculation formula is as follows:
L(X)＝B(PWConv2(δ(B(PWConv1(G(X))))))，
wherein PWConv represents point convolution, namely 1 multiplied by 1 convolution, B represents batch normalization layers, delta represents an activation function GELU, and X is a feature vector obtained by extracting features of an infrared photoelectric pod reconnaissance image;
the channel attention of the global feature is calculated, and the calculation formula is as follows:
L(X)＝B(PWConv2(δ(B(PWConv1(X)))))，
wherein G represents a global average pooling operation; PWConv represents a point convolution, i.e., a 1×1 convolution; b represents a batch normalization layer; delta represents the activation function GELU; y is a feature vector obtained by extracting features of a photoelectric pod reconnaissance image of visible light;
and finally, adding the two weights element by element to obtain a final feature fusion weight M.
In the semantic segmentation method based on visible light-infrared photoelectric reconnaissance image fusion, the decoder adopts a simple residual block plus a sampling function.
Compared with the prior art, the method and the device have the advantages that the advantages of the infrared photoelectric pod reconnaissance image are utilized, the convNEXT feature extractor is utilized to extract the features in the infrared photoelectric pod reconnaissance image and the visible light photoelectric pod reconnaissance image respectively, and the extracted features are fused through the differential feature fusion module, so that the visible light image and the infrared image can be fused effectively, and semantic segmentation with better robustness and higher accuracy is obtained.
In particular
The ConvNeXt convolution block adopted by the invention carries out convolution operation through a 7X 7 convolution check image, and then uses linear regularization treatment and twice 1X 1 convolution operation to finish the extraction of the features; operating using a GELU activation function between two 1 x 1 convolution operations; therefore, noise can be well restrained, and false detection and missing detection caused by interference factors such as illumination vegetation can be reduced. And the difference feature fusion module selectively reserves difference pixel points, so that the model can obtain richer global and local information.
In conclusion, the method has the characteristics of effectively improving the segmentation accuracy and reducing the false detection and missing detection phenomena caused by interference factors.
Experiments prove that on the visible light-infrared photoelectric reconnaissance image data set, the segmentation accuracy is improved by 2.1%, and the accuracy is improved by 2.3%.
Drawings
FIG. 1 is a general block diagram of the present invention;
FIG. 2 is a schematic diagram of a ConvNeXt convolution block structure;
fig. 3 is a schematic structural diagram of the differential feature fusion module.
Detailed Description
The invention is further illustrated by the following figures and examples, which are not intended to be limiting.
Examples. The semantic segmentation method based on visible light-infrared photoelectric reconnaissance image fusion is formed as shown in figures 1-3, and comprises the following steps:
s1, acquiring an infrared photoelectric pod reconnaissance image and a visible photoelectric pod reconnaissance image;
s2, respectively extracting features in an infrared photoelectric pod reconnaissance image and a visible photoelectric pod reconnaissance image by using a convNeXT feature extractor;
s3, fusing the extracted features through a differential feature fusion module;
and S4, performing up-sampling operation one by one through a decoder to obtain a semantic segmentation graph.
The convNeXT feature extractor processes the image by adopting four ConvNeXt convolution blocks in sequence, and extracts features with different scales;
the ConvNeXt convolution block connection mode adopts a reverse bottleneck structure, and the separable convolution of the ConvNeXt convolution block connection mode is moved up to a first layer;
the ConvNeXt convolution block is used in the following steps: firstly, carrying out convolution operation on an image by using 7 multiplied by 7 convolution check, and then, carrying out linear regularization treatment and twice 1 multiplied by 1 convolution operation to finish feature extraction; the gel activation function is used to operate between two 1 x 1 convolution operations.
The fusion calculation formula of the differential feature fusion module is as follows:
wherein OUT is the fused output characteristic, M is the characteristic fusion weight, X is the characteristic vector obtained by extracting the characteristic of the infrared photoelectric pod reconnaissance image, and Y is the characteristic vector obtained by extracting the characteristic of the visible photoelectric pod reconnaissance image.
The calculation process of the feature fusion weight M is as follows:
calculating the channel attention of the local feature, wherein the calculation formula is as follows:
L(X)＝B(PWConv2(δ(B(PWConv1(X)))))，
wherein PWConv represents point convolution, namely 1 multiplied by 1 convolution, B represents batch normalization layers, delta represents an activation function GELU, and X is a feature vector obtained by extracting features of an infrared photoelectric pod reconnaissance image;
the channel attention of the global feature is calculated, and the calculation formula is as follows:
L(X)＝B(PWConv2(δ(B(PWConv1(G(X))))))，
wherein G represents a global average pooling operation; PWConv represents a point convolution, i.e., a 1×1 convolution; b represents a batch normalization layer; delta represents the activation function GELU; y is a feature vector obtained by extracting features of a photoelectric pod reconnaissance image of visible light;
and finally, adding the two weights element by element to obtain a final feature fusion weight M.
The decoder uses a simple residual block, which is a method proposed by the ResNet network, plus a sampling function (upsampling by bilinear interpolation).
The semantic segmentation network proposed by the invention is shown in figure 1, and ConvNeXt is used as a backbone network of an encoder. The convNeXT feature extractor extracts features with different scales through four ConvNeXt convolution blocks, then carries out differential feature fusion on the features extracted by the infrared photoelectric pod reconnaissance image and the visible photoelectric pod reconnaissance image, finally carries out up-sampling operation one by one through a decoder, and finally obtains a segmentation map. The decoder uses a simple residual block plus a sampling function.
ConvNeXt convolution block as shown in FIG. 2, D7X17 represents a convolution operation with dimension 7, and C is the number of channels. The ConvNeXt connection uses a reverse bottleneck structure to shift its separable convolution up to the first layer. The typical convolution kernel size is 3 x 3, and the present invention uses a 7 x 7 convolution kernel, so the network employed by the present invention is also referred to as a large convolution kernel network. Using linear regularization after separable convolutions, a GELU activation function is used between the two 1 x 1 convolutions.
Feature fusion refers to a combination of features from different levels or branches, typically achieved by simple linear operations. The differential feature fusion structure provided by the invention is shown in fig. 3, and differential pixel points are selectively reserved, so that the network is more trainable. The specific process is as follows:
1) Firstly, subtracting absolute values of initial features X and Y element by element to obtain initial feature differences;
2) The feature fusion weight can be effectively obtained through the multi-scale channel attention unit;
the specific calculation formula can be expressed as:
wherein OUT is the fused output feature; m is a feature fusion weight; m (|X-Y|) and 1-M (|X-Y|) are composed of real numbers between 0 and 1, so that characteristic points between X and Y in the training process of the network are selected and removed, and finally the characteristic which is most contributed to the segmentation task is selected, so that the model can obtain richer global and local information.
By introducing a multi-scale channel attention unit, feature fusion weights can be effectively acquired. The unit consists of two separate parts: the attention weight containing global features is extracted by global average pooling; the other contains the attention weight of the local feature, directly using the point convolution operation.
The channel attention calculating method of the local features comprises the following steps:
L(X)＝B(PWConv2(δ(B(PWConv1(X)))))
wherein PWConv represents point convolution, namely 1X 1 convolution, reduces the number of input characteristic channels to 1/r originally, B represents a batch normalization layer, delta represents an activation function, and the invention refers to the structure of ConvNeXt and changes the activation function in two attention modules into GELU. The number of channels is restored to be the same as the number of the original input channels by convolution of 1×1, and r refers to the channel scaling ratio.
The channel attention calculation mode of the global features is different from the local channel attention in that the input features are subjected to a global average pooling operation. And adding the calculated two weights element by element to obtain a final feature fusion weight M.

Claims (5)

1. The semantic segmentation method based on visible light-infrared photoelectric reconnaissance image fusion is characterized by comprising the following steps of:
s1, acquiring an infrared photoelectric pod reconnaissance image and a visible photoelectric pod reconnaissance image;
s2, respectively extracting features in an infrared photoelectric pod reconnaissance image and a visible photoelectric pod reconnaissance image by using a convNeXT feature extractor;
s3, fusing the extracted features through a differential feature fusion module;
and S4, performing up-sampling operation one by one through a decoder to obtain a semantic segmentation graph.
2. The semantic segmentation method based on visible light-infrared photoelectric reconnaissance image fusion according to claim 1, wherein the semantic segmentation method is characterized in that: the convNeXT feature extractor processes the image by adopting four ConvNeXt convolution blocks in sequence, and extracts features with different scales;
the ConvNeXt convolution block connection mode adopts a reverse bottleneck structure, and the separable convolution of the ConvNeXt convolution block connection mode is moved up to a first layer;
the ConvNeXt convolution block is used in the following steps: firstly, carrying out convolution operation on an image by using 7 multiplied by 7 convolution check, and then, carrying out linear regularization treatment and twice 1 multiplied by 1 convolution operation to finish feature extraction; the gel activation function is used to operate between two 1 x 1 convolution operations.
3. The semantic segmentation method based on visible light-infrared photoelectric reconnaissance image fusion according to claim 2, wherein a fusion calculation formula of the differential feature fusion module is as follows:
wherein OUT is the fused output characteristic, M is the characteristic fusion weight, X is the characteristic vector obtained by extracting the characteristic of the infrared photoelectric pod reconnaissance image, and Y is the characteristic vector obtained by extracting the characteristic of the visible photoelectric pod reconnaissance image.
4. The semantic segmentation method based on visible light-infrared photoelectric reconnaissance image fusion according to claim 3, wherein the calculation process of the feature fusion weight M is as follows:
calculating the channel attention of the local feature, wherein the calculation formula is as follows:
L(X)＝B(PWConv2(δ(B(PWConv1(X))))，
wherein PWConv represents point convolution, namely 1 multiplied by 1 convolution, B represents batch normalization layers, delta represents an activation function GELU, and X is a feature vector obtained by extracting features of an infrared photoelectric pod reconnaissance image;
the channel attention of the global feature is calculated, and the calculation formula is as follows:
L(X)＝B(PWConv2(δ(B(PWConv1(G(X))))))，
wherein G represents a global average pooling operation; PWConv represents a point convolution, i.e., a 1×1 convolution; b represents a batch normalization layer; delta represents the activation function GELU; y is a feature vector obtained by extracting features of a photoelectric pod reconnaissance image of visible light;
and finally, adding the two weights element by element to obtain a final feature fusion weight M.
5. The semantic segmentation method based on visible-infrared photoelectric scout image fusion according to claim 1, wherein the decoder uses a residual block plus a sampling function.
CN202311037515.4A 2023-08-17 2023-08-17 Semantic segmentation method based on visible light-infrared photoelectric reconnaissance image fusion Pending CN117115442A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202311037515.4A CN117115442A (en) 2023-08-17 2023-08-17 Semantic segmentation method based on visible light-infrared photoelectric reconnaissance image fusion

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202311037515.4A CN117115442A (en) 2023-08-17 2023-08-17 Semantic segmentation method based on visible light-infrared photoelectric reconnaissance image fusion

Publications (1)

Publication Number Publication Date
CN117115442A true CN117115442A (en) 2023-11-24

Family

ID=88801398

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202311037515.4A Pending CN117115442A (en) 2023-08-17 2023-08-17 Semantic segmentation method based on visible light-infrared photoelectric reconnaissance image fusion

Country Status (1)

Country Link
CN (1) CN117115442A (en)

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112184604A (en) * 2020-09-15 2021-01-05 杭州电子科技大学 Color image enhancement method based on image fusion
CN113033630A (en) * 2021-03-09 2021-06-25 太原科技大学 Infrared and visible light image deep learning fusion method based on double non-local attention models
CN114612359A (en) * 2022-03-09 2022-06-10 南京理工大学 Visible light and infrared image fusion method based on feature extraction
CN114897883A (en) * 2022-06-11 2022-08-12 长春理工大学 Infrared and visible light image fusion method based on ResNet50 and double-pyramid
CN115063329A (en) * 2022-06-10 2022-09-16 中国人民解放军国防科技大学 Visible light and infrared image fusion enhancement method and system under low-illumination environment
CN115601723A (en) * 2022-10-24 2023-01-13 成都信息工程大学（Cn） Night thermal infrared image semantic segmentation enhancement method based on improved ResNet
CN115620010A (en) * 2022-09-20 2023-01-17 长春理工大学 Semantic segmentation method for RGB-T bimodal feature fusion
CN116580195A (en) * 2023-04-26 2023-08-11 齐鲁工业大学(山东省科学院) Remote sensing image semantic segmentation method and system based on ConvNeXt convolution

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112184604A (en) * 2020-09-15 2021-01-05 杭州电子科技大学 Color image enhancement method based on image fusion
CN113033630A (en) * 2021-03-09 2021-06-25 太原科技大学 Infrared and visible light image deep learning fusion method based on double non-local attention models
CN114612359A (en) * 2022-03-09 2022-06-10 南京理工大学 Visible light and infrared image fusion method based on feature extraction
CN115063329A (en) * 2022-06-10 2022-09-16 中国人民解放军国防科技大学 Visible light and infrared image fusion enhancement method and system under low-illumination environment
CN114897883A (en) * 2022-06-11 2022-08-12 长春理工大学 Infrared and visible light image fusion method based on ResNet50 and double-pyramid
CN115620010A (en) * 2022-09-20 2023-01-17 长春理工大学 Semantic segmentation method for RGB-T bimodal feature fusion
CN115601723A (en) * 2022-10-24 2023-01-13 成都信息工程大学（Cn） Night thermal infrared image semantic segmentation enhancement method based on improved ResNet
CN116580195A (en) * 2023-04-26 2023-08-11 齐鲁工业大学(山东省科学院) Remote sensing image semantic segmentation method and system based on ConvNeXt convolution

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
LEI WANG，ET AL.: "Infrared and Visible Image Fusion via Attention-Based Adaptive Feature Fusion", ENTROPY, pages 3 *
YIMIAN DAI，ET AL.: "Attentional Feature Fusion", 2021 IEEE WINTER CONFERENCE ON APPLICATIONS OF COMPUTER VISION (WACV) *
YU HUA，ET AL.: "A Residual ConvNeXt-based Network for Visible and Infrared Image Fusion", 《2023 4TH INTERNATIONAL CONFERENCE ON ELECTRONIC COMMUNICATION AND ARTIFICIAL INTELLIGENCE(ICECAI)》 *
ZHUANG LIU，ET AL.: "A ConvNet for the 2020s", ARXIV *
王钊, 等.: "基于 YOLO 的多模态特征差分注意融合行人检测", 计算机系统应用, pages 2 *

Similar Documents

Publication Publication Date Title
CN111462013B (en) Single-image rain removing method based on structured residual learning
CN113469094A (en) Multi-mode remote sensing data depth fusion-based earth surface coverage classification method
CN113033570B (en) Image semantic segmentation method for improving void convolution and multilevel characteristic information fusion
CN108491836B (en) Method for integrally identifying Chinese text in natural scene image
CN112381097A (en) Scene semantic segmentation method based on deep learning
CN114742799B (en) Industrial scene unknown type defect segmentation method based on self-supervision heterogeneous network
CN113763327B (en) Power plant pipeline high-pressure steam leakage detection method based on CBAM-Res_Unet
CN111666842A (en) Shadow detection method based on double-current-cavity convolution neural network
CN113076957A (en) RGB-D image saliency target detection method based on cross-modal feature fusion
CN113870124B (en) Weak supervision-based double-network mutual excitation learning shadow removing method
CN115908772A (en) Target detection method and system based on Transformer and fusion attention mechanism
CN115631513B (en) Transformer-based multi-scale pedestrian re-identification method
CN113066089A (en) Real-time image semantic segmentation network based on attention guide mechanism
CN116310916A (en) Semantic segmentation method and system for high-resolution remote sensing city image
CN114418987A (en) Retinal vessel segmentation method and system based on multi-stage feature fusion
CN112132867B (en) Remote sensing image change detection method and device
CN116778346B (en) Pipeline identification method and system based on improved self-attention mechanism
CN117115442A (en) Semantic segmentation method based on visible light-infrared photoelectric reconnaissance image fusion
CN115797782A (en) Remote sensing image small target detection method based on weighted extended feature pyramid
CN115641445A (en) Remote sensing image shadow detection method integrating asymmetric inner convolution and Transformer
CN113256528B (en) Low-illumination video enhancement method based on multi-scale cascade depth residual error network
CN113012071B (en) Image out-of-focus deblurring method based on depth perception network
CN112164078B (en) RGB-D multi-scale semantic segmentation method based on encoder-decoder
CN114549958A (en) Night and disguised target detection method based on context information perception mechanism
CN114022719A (en) Multi-feature fusion significance detection method

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