CN113192052A - Decomposition fusion subtraction automatic encoder algorithm for image feature extraction - Google Patents

Decomposition fusion subtraction automatic encoder algorithm for image feature extraction Download PDF

Info

Publication number
CN113192052A
CN113192052A CN202110550127.0A CN202110550127A CN113192052A CN 113192052 A CN113192052 A CN 113192052A CN 202110550127 A CN202110550127 A CN 202110550127A CN 113192052 A CN113192052 A CN 113192052A
Authority
CN
China
Prior art keywords
layer
fusion
decomposition
features
feature
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
CN202110550127.0A
Other languages
Chinese (zh)
Inventor
孙宇
魏本征
李翔
张魁星
丛金玉
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.)
Shandong University of Traditional Chinese Medicine
Original Assignee
Shandong University of Traditional Chinese Medicine
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 Shandong University of Traditional Chinese Medicine filed Critical Shandong University of Traditional Chinese Medicine
Priority to CN202110550127.0A priority Critical patent/CN113192052A/en
Publication of CN113192052A publication Critical patent/CN113192052A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0012—Biomedical image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • 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
    • G06V10/30—Noise filtering
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10072—Tomographic images
    • G06T2207/10081—Computed x-ray tomography [CT]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20212—Image combination
    • G06T2207/20221—Image fusion; Image merging
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30061—Lung

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Mining & Analysis (AREA)
  • Multimedia (AREA)
  • General Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Radiology & Medical Imaging (AREA)
  • Quality & Reliability (AREA)
  • Image Processing (AREA)

Abstract

The invention discloses a decomposition fusion subtraction automatic encoder algorithm for image feature extraction, which comprises a decomposition fusion subtraction network structure and a decomposition reconstruction loss function, wherein the decomposition fusion subtraction network structure is firstly divided into a main channel and an auxiliary channel, the main channel and the auxiliary channel are subjected to mutually independent feature extraction, features in the auxiliary channel and features in the main channel are subjected to feature fusion, local features of each layer at the front end and global features of each layer at the rear end are subjected to cross-layer fusion of features by utilizing jump connection in the main channel, the features are extracted by establishing the two channels, the features are fused in a feature layer between the front feature layer and the rear feature layer and a feature layer between the channels, and a fusion weight determination strategy is designed; the latest decomposition reconstruction loss function is designed, the reconstruction loss is calculated among the symmetrical characteristic layers, the loss and the loss among signals form a final reconstruction loss function, and a segmentation optimization strategy is adopted in the optimization process of the final reconstruction loss function, so that the optimization capability of the characteristics is enhanced.

Description

Decomposition fusion subtraction automatic encoder algorithm for image feature extraction
Technical Field
The invention relates to the technical field of image feature extraction, in particular to a decomposition fusion subtraction automatic encoder algorithm for image feature extraction.
Background
Although the existing automatic encoder algorithm obtains remarkable results when being used for extracting various image features, the problems that all detail features of partial images have potential values, imaging parts have high similarity, the influence of imaging equipment is large and the like make the extraction of high-quality features in the image feature extraction process especially necessary. The system analysis of the existing various automatic encoder algorithms shows that the existing various automatic encoder algorithms have the problems of large algorithm parameter quantity, low universality on different application scenes and the like on the whole, specifically, the existing various automatic encoder algorithms lack of feature fusion in the encoding process, the signal loss is serious in the decoding process, the optimization target in a loss function is single, and the quality of the extracted features is limited finally.
Based on this, the present invention designs a decomposition-fusion-subtraction automatic encoder algorithm for image feature extraction to solve the above mentioned problems.
Disclosure of Invention
The invention aims to provide a decomposition fusion subtraction automatic encoder algorithm for image feature extraction, which solves the problems of insufficient feature extraction, low quality and the like by establishing a new feature extraction strategy in an encoder; the problems of low characteristic decoding efficiency, large characteristic loss and the like are solved in a decoder. In the reconstruction loss function, the method mainly solves the problems that the intermediate characteristic layers cannot be directly optimized, the reconstruction loss function is easy to fall into precocity integrally to influence the target characteristic extraction quality and the like.
In order to achieve the purpose, the invention provides the following technical scheme: a decomposition-fusion-subtraction automatic encoder algorithm for image feature extraction comprises establishing a decomposition-fusion-subtraction network structure and a decomposition reconstruction loss function,
the decomposition, fusion and subtraction network structure is firstly divided into a main channel and an auxiliary channel, mutually independent feature extraction is carried out on the main channel and the auxiliary channel, feature fusion is carried out on features in the auxiliary channel and features in the main channel, meanwhile, cross-layer fusion of features is carried out on local features of each layer of the front end and global features of each layer of the rear end in the main channel by utilizing jump connection, and the fusion process among feature layers can be expressed as follows:
Q＝αβ(Wa*Ea+ba+1)+χβ(Wd*Ed+bd+1) (1)
wherein Q is a fused feature layer; alpha and chi are weight coefficients of the fusion layer and the fused layer; beta is an activation function; waIs the node weight of the fusion layer; wdNode weight of the fused layer; eaAn input for a fusion layer; edIs the input of the fused layer; ba+1Is the fusion layer bias vector; bd+1Is biased to the fused layer; is a convolution operation;
the decomposition reconstruction loss function may be specifically expressed as:
Figure RE-GDA0003130979060000021
wherein L isDMRAE(W, b) represents a decomposition reconstruction loss; n denotes the number of samples, xiAnd yiRepresenting a reconstruction loss between the input signal and the recovered signal for the ith sample; e.g. of the typeiAnd diRepresenting the reconstruction loss between symmetric feature layers in the ith sample feature extraction.
Preferably, each layer of local features of the front end is transmitted only once, and each layer of global features of the back end is also fused with one layer of local features.
Compared with the prior art, the invention has the beneficial effects that: the invention establishes and uses the dual channel to extract the characteristic in the coder, and carries on the characteristic fusion in the characteristic layer before and after and the characteristic layer between the channels, and has designed the fusion weight to confirm the tactics; the latest decomposition reconstruction loss function is designed, the function calculates the reconstruction loss between symmetrical characteristic layers, the loss and the loss between signals form a final reconstruction loss function, and a segmentation optimization strategy is adopted in the optimization process of the final reconstruction loss function to enhance the optimization capability of the characteristics.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a block diagram of a network architecture of an automatic encoder for decomposition and fusion reduction according to the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Referring to fig. 1, the present invention provides a technical solution: a decomposition-fusion-subtraction automatic encoder algorithm for image feature extraction comprises establishing a decomposition-fusion-subtraction network structure and a decomposition reconstruction loss function,
the method comprises the steps of providing a decomposition and fusion network structure for enhancing feature extraction capacity, dividing the decomposition and fusion network structure into a main channel and an auxiliary channel, performing feature extraction on the main channel and the auxiliary channel independently, performing feature fusion on features in the auxiliary channel and features in the main channel, performing cross-layer fusion on local features of each layer at the front end and global features of each layer at the rear end by using jump connection in the main channel, and in the feature fusion process, in order to avoid feature redundancy of a feature rear end feature layer and increase of model parameters, transmitting local features of each layer at the front end only once, and fusing global features of each layer at the rear end with local features of only one layer. The fusion process between feature layers can be expressed as:
Q＝αβ(Wa*Ea+ba+1)+χβ(Wd*Ed+bd+1) (1)
wherein Q is a fused feature layer; alpha and chi are weight coefficients of the fusion layer and the fused layer; beta is an activation function; waIs the node weight of the fusion layer; wdNode weight of the fused layer; eaAn input for a fusion layer; edIs the input of the fused layer; ba+1Is the fusion layer bias vector; bd+1Is biased to the fused layer; is a convolution operation;
a decomposition reconstruction loss function is provided in the optimization process of the features extracted by the algorithm, and the optimization capability of the loss function is improved by establishing a fusion strategy for establishing a relation between feature layers, optimizing the loss between features and the loss between signals, and adopting strategies such as segmentation optimization and the like for the loss between signals.
The decomposition reconstruction loss function may be specifically expressed as:
Figure RE-GDA0003130979060000041
wherein L isDMRAE(W, b) represents a decomposition reconstruction loss; n denotes the number of samples, xiAnd yiRepresenting a reconstruction loss between the input signal and the recovered signal for the ith sample; e.g. of the typeiAnd diRepresenting the reconstruction loss between symmetrical feature layers in the ith sample feature extraction; the loss functions of the two processes are squared differences.
Compared with the prior similar products or methods, the technical scheme of the invention has the advantages or can achieve the beneficial technical effects that:
1. the decomposition and fusion network structure in the decomposition and fusion automatic encoder improves the richness of extracted features through two-channel feature extraction and feature fusion, reduces the feature loss and improves the decoding efficiency through simplifying a decoding feature layer; meanwhile, the decomposition reconstruction loss function of the algorithm can also strengthen the relation between the corresponding layers of the coding and decoding stages and avoid premature model.
2. As shown in table 1, the decomposition fusion subtraction automatic encoder realizes smaller network scale compared with similar algorithms, and the parameter amount is reduced by 98.84%, 77.92%, 57.86%, 43.68% and 35.54% compared with automatic encoder algorithms such as dual noise reduction automatic encoder, sparse automatic encoder, noise reduction automatic encoder, convolution noise reduction automatic encoder and convolution automatic encoder.
3. Compared with the similar algorithm, the decomposition fusion subtraction automatic encoder has better feature extraction capability, and the extracted features have stronger expression capability. The feature extraction experiment on breast cancer axillary lymph node metastasis ultrasonic images shows that the classification accuracy of the features extracted by the decomposition and fusion reduction automatic encoder is improved by 12.50%, 16.00%, 22.25%, 16.50% and 10.50% respectively compared with the convolution noise reduction automatic encoder, the sparse automatic encoder, the noise reduction automatic encoder and the double noise reduction automatic encoder, and the decomposition and fusion reduction automatic encoder has better feature extraction capability.
4. Compared with the similar algorithm, the decomposition and fusion reduction automatic encoder has better robustness on different data sets, and the classification accuracy of the extracted features on the lung CT image is improved by 13.87%, 22.69%, 40.19%, 19.42% and 10.49% respectively compared with the convolution automatic encoder, the convolution noise reduction automatic encoder, the sparse automatic encoder, the noise reduction automatic encoder and the double noise reduction automatic encoder, which shows that the decomposition and fusion reduction automatic encoder has good robustness on different data sets.
5. The decomposition and fusion reduction network structure and the decomposition and reconstruction loss function have obvious effect on the improvement of the feature extraction capability. The method selects a convolution automatic encoder with better comprehensive performance as a basis, improves the convolution automatic encoder by respectively using a decomposition and fusion reduction network structure and decomposition and reconstruction loss, respectively extracts features on breast cancer axillary lymph node metastasis ultrasonic images and lung CT images by using the improved model, and classifies samples based on the extracted features. The results show that in breast cancer axillary lymph node metastasis ultrasound images, the classification accuracy of the features extracted by the improved convolution automatic encoder using the decomposition-fusion subtraction network structure and the decomposition-reconstruction loss function is respectively improved by 6.00% and 7.50% compared with the convolution automatic encoder, and is respectively improved by 7.12% and 8.68% on a lung CT image dataset.
Table 1: decomposition-fusion-subtraction autoencoder network parameter table:
Figure BDA0003075115640000051
in the description herein, references to the description of "one embodiment," "an example," "a specific example" or the like are intended to mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, a schematic representation of the above terms does not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
The preferred embodiments of the invention disclosed above are intended to be illustrative only. The preferred embodiments are not intended to be exhaustive or to limit the invention to the precise embodiments disclosed. Obviously, many modifications and variations are possible in light of the above teaching. The embodiments were chosen and described in order to best explain the principles of the invention and the practical application, to thereby enable others skilled in the art to best utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims (2)

1. A decomposition-fusion-subtraction autoencoder algorithm for image feature extraction, characterized by: including establishing a decomposition-fusion-subtraction network structure and a decomposition-reconstruction-loss function,
the decomposition, fusion and subtraction network structure is firstly divided into a main channel and an auxiliary channel, mutually independent feature extraction is carried out on the main channel and the auxiliary channel, feature fusion is carried out on features in the auxiliary channel and features in the main channel, meanwhile, cross-layer fusion of features is carried out on local features of each layer of the front end and global features of each layer of the rear end in the main channel by utilizing jump connection, and the fusion process among feature layers can be expressed as follows:
Q＝αβ(Wa*Ea+ba+1)+χβ(Wd*Ed+bd+1) (1)
wherein Q is a fused feature layer; alpha and chi are weight coefficients of the fusion layer and the fused layer; beta is an activation function; waIs the node weight of the fusion layer; wdNode weight of the fused layer; eaAn input for a fusion layer; edIs the input of the fused layer; ba+1Is the fusion layer bias vector; bd+1Is the fused layer offset vector; is a convolution operation;
the decomposition reconstruction loss function may be specifically expressed as:
Figure RE-FDA0003130979050000011
wherein L isDMRAE(W, b) represents a decomposition reconstruction loss; n denotes the number of samples, xiAnd yiRepresenting a reconstruction loss between the input signal and the recovered signal for the ith sample; e.g. of the typeiAnd diRepresenting the reconstruction loss between symmetric feature layers in the ith sample feature extraction.
2. A decomposition-fusion-subtraction auto-encoder algorithm for image feature extraction as claimed in claim 1, wherein: each layer of local features of the front end are transmitted only once, and each layer of global features of the rear end are only fused with one layer of local features.
CN202110550127.0A 2021-05-20 2021-05-20 Decomposition fusion subtraction automatic encoder algorithm for image feature extraction Pending CN113192052A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110550127.0A CN113192052A (en) 2021-05-20 2021-05-20 Decomposition fusion subtraction automatic encoder algorithm for image feature extraction

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110550127.0A CN113192052A (en) 2021-05-20 2021-05-20 Decomposition fusion subtraction automatic encoder algorithm for image feature extraction

Publications (1)

Publication Number Publication Date
CN113192052A true CN113192052A (en) 2021-07-30

Family

ID=76982919

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110550127.0A Pending CN113192052A (en) 2021-05-20 2021-05-20 Decomposition fusion subtraction automatic encoder algorithm for image feature extraction

Country Status (1)

Country Link
CN (1) CN113192052A (en)

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112183561A (en) * 2020-11-09 2021-01-05 山东中医药大学 Joint fusion subtraction automatic encoder algorithm for image feature extraction
CN112446415A (en) * 2020-10-09 2021-03-05 山东中医药大学 Fusion subtraction automatic encoder algorithm for image feature extraction

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112446415A (en) * 2020-10-09 2021-03-05 山东中医药大学 Fusion subtraction automatic encoder algorithm for image feature extraction
CN112183561A (en) * 2020-11-09 2021-01-05 山东中医药大学 Joint fusion subtraction automatic encoder algorithm for image feature extraction

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
YU SUN等: "a co-melting reduction auto-encoder", 2020 7TH INTERNATIONAL CONFERENCE ON INFORMATION SCIENCE AND CONTROL ENGINEERING, pages 1530 - 1533 *

Similar Documents

Publication Publication Date Title
US10887569B2 (en) Virtual viewpoint synthesis method based on local image segmentation
CN112348766B (en) Progressive feature stream depth fusion network for surveillance video enhancement
US5786855A (en) Method and apparatus for coding segmented regions in video sequences for content-based scalability
CN101193261B (en) Video communication system and method
CN110751612A (en) Single image rain removing method of multi-channel multi-scale convolution neural network
CN113392960B (en) Target detection network and method based on mixed hole convolution pyramid
CN101662677B (en) Code stream conversion system, code stream conversion method, code stream identification unit and scheme determination unit
CN110111288B (en) Image enhancement and blind image quality evaluation network system based on deep assisted learning
CN110689599A (en) 3D visual saliency prediction method for generating countermeasure network based on non-local enhancement
CN101986716A (en) Quick depth video coding method
CN110351548B (en) Stereo image quality evaluation method guided by deep learning and disparity map weighting
CN101621683A (en) Fast stereo video coding method based on AVS
CN108830812A (en) A kind of high frame per second of video based on network deep learning remakes method
CN113034563A (en) Self-supervision type monocular depth estimation method based on feature sharing
Liu et al. Fast depth intra coding based on depth edge classification network in 3D-HEVC
CN115546046A (en) Single image defogging method fusing frequency and content characteristics
CN113192052A (en) Decomposition fusion subtraction automatic encoder algorithm for image feature extraction
CN114119694A (en) Improved U-Net based self-supervision monocular depth estimation algorithm
CN115442613A (en) Interframe information-based noise removal method using GAN
CN111652922B (en) Binocular vision-based monocular video depth estimation method
CN104732508B (en) Based on the image interfusion method that distributed compression is perceived
CN116630850A (en) Twin target tracking method based on multi-attention task fusion and bounding box coding
CN107509074B (en) Self-adaptive 3D video compression coding and decoding method based on compressed sensing
CN113096176B (en) Semantic segmentation-assisted binocular vision unsupervised depth estimation method
CN110958417B (en) Method for removing compression noise of video call video based on voice clue

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