CN114693926A - Image semantic segmentation method based on deep learning - Google Patents

Image semantic segmentation method based on deep learning Download PDF

Info

Publication number
CN114693926A
CN114693926A CN202210292438.6A CN202210292438A CN114693926A CN 114693926 A CN114693926 A CN 114693926A CN 202210292438 A CN202210292438 A CN 202210292438A CN 114693926 A CN114693926 A CN 114693926A
Authority
CN
China
Prior art keywords
image data
semantic segmentation
training
deep learning
image
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
CN202210292438.6A
Other languages
Chinese (zh)
Inventor
匡芳君
张思扬
邢军
沈良忠
郝慧君
李博然
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.)
Wenzhou Business College
Original Assignee
Wenzhou Business College
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 Wenzhou Business College filed Critical Wenzhou Business College
Priority to CN202210292438.6A priority Critical patent/CN114693926A/en
Publication of CN114693926A publication Critical patent/CN114693926A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformation in the plane of the image
    • G06T3/40—Scaling the whole image or part thereof
    • G06T3/4007—Interpolation-based scaling, e.g. bilinear interpolation

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)
  • Bioinformatics & Computational Biology (AREA)
  • Computational Linguistics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Evolutionary Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Image Analysis (AREA)

Abstract

The invention provides an image semantic segmentation method based on deep learning, which comprises four steps of data acquisition, model construction, model training and semantic segmentation, wherein the original image data and the image data for training are pre-processed to reduce the resolution of the original image data and the image data for training and avoid the influence of overhigh resolution on the operation speed of image semantic segmentation, a residual connecting module and a dense connecting module are introduced into the image semantic segmentation network model based on deep learning, the residual modules are used for carrying out feature fusion, so that the information quantity for describing the image features and details is increased, the dense connecting module is used for enhancing the transmission among the features, thereby avoiding the loss of feature information in the feature extraction process, improving the feature extraction effect, and testing the trained model by introducing a test data set, and optimizing the trained model continuously according to the test result.

Description

Image semantic segmentation method based on deep learning
Technical Field
The invention relates to the technical field of image semantic segmentation, in particular to an image semantic segmentation method based on deep learning.
Background
The digital image processing technology is a interdisciplinary field, with the continuous development of computer science and technology, image processing and analysis gradually form a scientific system, and new processing methods are developed endlessly, and although the development history is not long, the method attracts the wide attention of people in all aspects. Firstly, vision is the most important perception means of human beings, and images are the basis of vision, so that digital images become effective tools for researchers in various fields such as psychology, physiology, computer science and the like to research visual perception;
in the field of computer vision, the application of the current neural network mainly comprises image recognition, target positioning and detection and semantic segmentation, wherein the image recognition is used for telling you what the image is, the target positioning and detection are used for telling you where the target is in the image, and the semantic segmentation is used for answering the two questions from the pixel level;
image semantic segmentation (semantic segmentation), which is understood literally, a computer is segmented according to image semantics, and a scene understanding task in the computer vision field is mainly realized by an image segmentation technology, but traditional image segmentation methods including a region segmentation method, an edge detection segmentation method and an image segmentation method cannot really recognize semantics of objects in a picture, have less information describing image features and details, and cannot achieve the purpose of understanding scene information, so that the invention provides an image semantic segmentation method based on deep learning to solve the problems in the prior art.
Disclosure of Invention
In view of the above problems, the present invention aims to provide an image semantic segmentation method based on deep learning, which solves the problem of less information describing image features and details in the prior art.
In order to realize the purpose of the invention, the invention is realized by the following technical scheme: an image semantic segmentation method based on deep learning comprises the following steps:
step one, data acquisition
The method comprises the steps of obtaining original image data to be processed and image data for training, and preprocessing the original image data and the image data for training to obtain preprocessed original image data and preprocessed image data for training;
step two: model construction
Constructing an image semantic segmentation network model based on deep learning, wherein the image semantic segmentation network model based on deep learning comprises a down-sampling encoder layer, an up-sampling decoder layer, a residual error connection module and a dense connection module;
step three: model training
Inputting the training image data obtained in the first step into the deep learning-based image semantic segmentation network model constructed in the second step, and performing model training to obtain a trained image semantic segmentation network model;
step four: semantic segmentation
And (4) inputting the preprocessed original image data obtained in the first step into the trained image semantic segmentation network model in the third step, and outputting a result.
The further improvement lies in that: in the first step, the preprocessing is to perform downsampling on the original image data and the training image data, reduce the resolution of the original image data and the training image data, and avoid the influence of too high image resolution on the operation speed of image semantic segmentation.
The further improvement lies in that: in the first step, a plurality of groups of image data for training are required to be obtained to form an image data set for model training.
The further improvement lies in that: in the first step, the resolution of the preprocessed original image data is consistent with that of the image data for training.
The further improvement lies in that: in the second step, a plurality of groups of encoder layers for down sampling and decoder layers for up sampling are provided.
The further improvement lies in that: in the second step, the residual connecting module is used for connecting with the encoder layers and introducing a residual item in the model training process, and the residual connecting module is not connected with all the encoder layers.
The further improvement lies in that: in the second step, the image semantic segmentation network model is an improvement on the existing full convolution network model.
The further improvement is that: and in the third step, a test image data set is obtained and input into the trained model for testing.
The invention has the beneficial effects that: according to the image semantic segmentation method based on the deep learning, the residual connection module and the dense connection module are introduced into the image semantic segmentation network model based on the deep learning, the residual module is used for carrying out feature fusion, so that the information quantity for describing image features and details is increased, the dense connection module is used for enhancing the transmission among the features, therefore, the loss of feature information in the feature extraction process is avoided, the feature extraction effect is improved, the original image data and the image data for training can be preprocessed, the resolution of the original image data and the image data for training are reduced, and the problem that the running speed of image semantic segmentation is influenced due to the overhigh resolution of the images is avoided.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to these drawings without creative efforts.
FIG. 1 is a flowchart illustrating a first step of the present invention.
FIG. 2 is a schematic diagram of an encoder layer connection structure according to a second embodiment of 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.
Example one
According to fig. 1, the embodiment proposes an image semantic segmentation method based on deep learning, which includes the following steps:
step one, data acquisition
The method comprises the steps of obtaining original image data to be processed and image data for training, and preprocessing the original image data and the image data for training to obtain preprocessed original image data and preprocessed image data for training, wherein in the first step, preprocessing is to perform down-sampling on the original image data and the image data for training, the resolution ratios of the original image data and the image data for training are reduced, and the problem that the operating speed of semantic segmentation of images is affected due to overhigh resolution ratios of the images is avoided;
step two: model construction
Constructing an image semantic segmentation network model based on deep learning, wherein the image semantic segmentation network model based on deep learning comprises a down-sampling encoder layer, an up-sampling decoder layer, a residual error connection module and a dense connection module, in the second step, a plurality of groups of encoder layers for down sampling and decoder layers for up sampling are arranged, in the second step, the residual error connecting module is used for connecting with the encoder layers, and residual error items are introduced in the process of model training, and the residual error connecting module is not connected with all the encoder layers, in the second step, the image semantic segmentation network model is an improvement of the existing full convolution network model, a plurality of groups of encoder layers form an encoder, the front part of the coder layers in the coder are connected by adopting a residual error connecting module, the rear part of the coder layers are connected by adopting a dense connecting module, and a plurality of groups of decoder layers form a decoder;
step three: model training
Inputting the training image data obtained in the first step into the deep learning-based image semantic segmentation network model constructed in the second step, and performing model training to obtain a trained image semantic segmentation network model;
step four: semantic segmentation
And (4) inputting the preprocessed original image data obtained in the first step into the trained image semantic segmentation network model in the third step, and outputting a result.
Example two
The embodiment provides an image semantic segmentation method based on deep learning, which includes the following steps:
step one, data acquisition
The method comprises the steps of obtaining original image data to be processed and image data for training, and preprocessing the original image data and the image data for training to obtain preprocessed original image data and preprocessed image data for training, wherein in the first step, preprocessing is to perform down-sampling on the original image data and the image data for training, the resolution ratios of the original image data and the image data for training are reduced, and the problem that the operating speed of semantic segmentation of images is affected due to overhigh resolution ratios of the images is avoided;
step two: model construction
Constructing an image semantic segmentation network model based on deep learning, wherein the image semantic segmentation network model based on deep learning comprises a down-sampling encoder layer, an up-sampling decoder layer, a residual error connection module and an intensive connection module, in the second step, a plurality of groups of encoder layers and up-sampling decoder layers are arranged, in the second step, the residual error connection module is used for being connected with the encoder layers and introducing a residual error item in the model training process, the residual error connection module is not connected with all the encoder layers, in the second step, the image semantic segmentation network model is formed by improving the existing full convolution network model, a plurality of groups of encoder layers form an encoder, and a plurality of groups of decoder layers form a decoder;
wherein, the residual connecting module mainly forms residual connection between the encoder layers, and in the constructed image semantic segmentation network model, as shown in FIG. 2, is a distribution schematic diagram of the encoder layers, through residual connection, the condition that the parameters of the identity mapping of the layer are learned in the model construction process can be avoided, thereby simplifying the difficulty of model learning, avoiding the influence of a redundant layer on the network effect, effectively relieving the degradation phenomenon caused by the increase of the network depth, meanwhile, the residual error connection module also adopts an element addition method for fusion, so that the information quantity describing the image characteristics and details is increased after the subsequent characteristic fusion is carried out, the encoder layer also comprises a convolution of 3 multiplied by 3, when the picture is input, the feature extraction is carried out through the convolution of 3 multiplied by 3, and the generated feature graph enters a subsequent network model on one hand and is directly fused with a new feature graph on the other hand;
the dense connection module leads the output of the previous encoder layer into the following encoder layer, namely, all layers are directly connected on the premise of ensuring the maximum information transmission between the layers in the network, and the dense connection module is fused in a channel number merging mode through an element splicing method, the feature graph fused by the method not only contains the features after convolution extraction, but also directly receives the enhancement of the initial features, and the information contained in different feature graphs is more effectively utilized
Step three: model training
Inputting the training image data obtained in the first step into the deep learning-based image semantic segmentation network model constructed in the second step, and performing model training to obtain a trained image semantic segmentation network model;
step four: semantic segmentation
Inputting the preprocessed original image data obtained in the first step into a trained image semantic segmentation network model in the third step, outputting a result, wherein the result output is output through a decoder layer, the decoder layer performs up-sampling through deconvolution initialized by a bilinear interpolation filter, and simultaneously, a splicing fusion method is adopted to directly fuse a feature map output by each encoder layer with the feature map sampled on the decoder layer, so that the segmentation result is refined by using spatial information of different resolutions at different stages in an encoder, and accurate and detailed segmentation is generated, and the image semantic separation operation of the original image data is completed.
EXAMPLE III
The embodiment provides an image semantic segmentation method based on deep learning, which includes the following steps:
step one, data acquisition
The method comprises the steps of acquiring original image data to be processed and image data for training, and preprocessing the original image data and the image data for training to obtain preprocessed original image data and preprocessed image data for training, wherein in the first step, preprocessing is to downsample the original image data and the image data for training, the resolutions of the original image data and the image data for training are reduced, and the operating speed of semantic segmentation of an image is prevented from being influenced by overhigh image resolution;
step two: model construction
Constructing an image semantic segmentation network model based on deep learning, wherein the image semantic segmentation network model based on deep learning comprises a down-sampling encoder layer, an up-sampling decoder layer, a residual error connection module and an intensive connection module, in the second step, a plurality of groups of encoder layers and up-sampling decoder layers are arranged, in the second step, the residual error connection module is used for being connected with the encoder layers and introducing a residual error item in the model training process, the residual error connection module is not connected with all the encoder layers, in the second step, the image semantic segmentation network model is formed by improving the existing full convolution network model, a plurality of groups of encoder layers form an encoder, and a plurality of groups of decoder layers form a decoder;
step three: model training
Inputting the training image data obtained in the first step into the deep learning-based image semantic segmentation network model constructed in the second step, performing model training to obtain a trained image semantic segmentation network model, acquiring a test image data set in the third step, inputting the test image data set into the trained model for testing, and testing according to a test result;
step four: semantic segmentation
And (4) inputting the preprocessed original image data obtained in the first step into the trained image semantic segmentation network model in the third step, and outputting a result.
The difference between this embodiment and the first and second embodiments is that, in the trained model, a test image data set is introduced to perform corresponding tests on the trained model, and parameters of the image semantic segmentation network model are continuously adjusted according to a test result, that is, the image semantic segmentation network model is continuously optimized according to the test result, which is beneficial to improving the image semantic segmentation effect.
According to the method, the residual connecting module and the dense connecting module are introduced into the construction of the image semantic segmentation network model based on deep learning, the residual module is used for carrying out feature fusion, so that the information quantity for describing image features and details is increased, the dense connecting module is used for enhancing the transmission among the features, the loss of feature information in the feature extraction process is avoided, the feature extraction effect is improved, the original image data and the image data for training can be preprocessed, the resolution of the original image data and the image data for training are reduced, and the problem that the operating speed of image semantic segmentation is influenced due to overhigh image resolution is avoided.
The foregoing illustrates and describes the principles, general features, and advantages of the present invention. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, which are described in the specification and illustrated only to illustrate the principle of the present invention, but that various changes and modifications may be made therein without departing from the spirit and scope of the present invention, which fall within the scope of the invention as claimed. The scope of the invention is defined by the appended claims and equivalents thereof.

Claims (8)

1. An image semantic segmentation method based on deep learning is characterized in that: the method comprises the following steps:
step one, data acquisition
The method comprises the steps of obtaining original image data to be processed and image data for training, and preprocessing the original image data and the image data for training to obtain preprocessed original image data and preprocessed image data for training;
step two: model construction
Constructing an image semantic segmentation network model based on deep learning, wherein the image semantic segmentation network model based on deep learning comprises a down-sampling encoder layer, an up-sampling decoder layer, a residual error connection module and a dense connection module;
step three: model training
Inputting the training image data obtained in the first step into the deep learning-based image semantic segmentation network model constructed in the second step, and performing model training to obtain a trained image semantic segmentation network model;
step four: semantic segmentation
And (4) inputting the preprocessed original image data obtained in the first step into the trained image semantic segmentation network model in the third step, and outputting a result.
2. The image semantic segmentation method based on deep learning according to claim 1, characterized in that: in the first step, the preprocessing is to perform downsampling on the original image data and the training image data, reduce the resolution of the original image data and the training image data, and avoid the influence of too high image resolution on the operation speed of image semantic segmentation.
3. The image semantic segmentation method based on deep learning according to claim 1, characterized in that: in the first step, a plurality of groups of image data for training are required to be obtained to form an image data set for model training.
4. The image semantic segmentation method based on deep learning according to claim 1, characterized in that: in the first step, the resolution of the preprocessed original image data is consistent with that of the preprocessed image data for training.
5. The image semantic segmentation method based on deep learning according to claim 1, characterized in that: in the second step, a plurality of groups of encoder layers for down sampling and decoder layers for up sampling are provided.
6. The image semantic segmentation method based on deep learning according to claim 1, characterized in that: in the second step, the residual connecting module is used for connecting with the encoder layers and introducing a residual item in the model training process, and the residual connecting module is not connected with all the encoder layers.
7. The image semantic segmentation method based on deep learning according to claim 1, characterized in that: in the second step, the image semantic segmentation network model is an improvement on the existing full convolution network model.
8. The image semantic segmentation method based on deep learning according to claim 1, characterized in that: and in the third step, a test image data set is obtained and input into the trained model for testing.
CN202210292438.6A 2022-03-23 2022-03-23 Image semantic segmentation method based on deep learning Pending CN114693926A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210292438.6A CN114693926A (en) 2022-03-23 2022-03-23 Image semantic segmentation method based on deep learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210292438.6A CN114693926A (en) 2022-03-23 2022-03-23 Image semantic segmentation method based on deep learning

Publications (1)

Publication Number Publication Date
CN114693926A true CN114693926A (en) 2022-07-01

Family

ID=82138633

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210292438.6A Pending CN114693926A (en) 2022-03-23 2022-03-23 Image semantic segmentation method based on deep learning

Country Status (1)

Country Link
CN (1) CN114693926A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116912257A (en) * 2023-09-14 2023-10-20 东莞理工学院 Concrete pavement crack identification method based on deep learning and storage medium

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116912257A (en) * 2023-09-14 2023-10-20 东莞理工学院 Concrete pavement crack identification method based on deep learning and storage medium
CN116912257B (en) * 2023-09-14 2023-12-29 东莞理工学院 Concrete pavement crack identification method based on deep learning and storage medium

Similar Documents

Publication Publication Date Title
CN111325751B (en) CT image segmentation system based on attention convolution neural network
CN108491836B (en) Method for integrally identifying Chinese text in natural scene image
CN112329780B (en) Depth image semantic segmentation method based on deep learning
CN111652081A (en) Video semantic segmentation method based on optical flow feature fusion
CN111402138A (en) Image super-resolution reconstruction method of supervised convolutional neural network based on multi-scale feature extraction fusion
CN111914654A (en) Text layout analysis method, device, equipment and medium
CN114724155A (en) Scene text detection method, system and equipment based on deep convolutional neural network
CN114820579A (en) Semantic segmentation based image composite defect detection method and system
CN112149526B (en) Lane line detection method and system based on long-distance information fusion
CN113160058A (en) Thermal imaging super-resolution reconstruction method combining segmentation map and storage medium
CN110852199A (en) Foreground extraction method based on double-frame coding and decoding model
CN115082928B (en) Method for asymmetric double-branch real-time semantic segmentation network facing complex scene
CN114359297A (en) Attention pyramid-based multi-resolution semantic segmentation method and device
CN114693926A (en) Image semantic segmentation method based on deep learning
CN116486080A (en) Lightweight image semantic segmentation method based on deep learning
CN117197624A (en) Infrared-visible light image fusion method based on attention mechanism
CN116229106A (en) Video significance prediction method based on double-U structure
CN114119694A (en) Improved U-Net based self-supervision monocular depth estimation algorithm
CN114049652A (en) Human body posture migration method and system based on action driving
CN112419159A (en) Character image super-resolution reconstruction system and method
CN116934613A (en) Branch convolution channel attention module for character repair
CN111612803A (en) Vehicle image semantic segmentation method based on image definition
CN116310375A (en) Blind image quality assessment method based on visual attention mechanism
CN112488115B (en) Semantic segmentation method based on two-stream architecture
CN114693712A (en) Dark vision and low-illumination image edge detection method based on deep learning

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