CN111539463A - Method and system for realizing image classification by simulating neuron dendritic branches - Google Patents

Method and system for realizing image classification by simulating neuron dendritic branches Download PDF

Info

Publication number
CN111539463A
CN111539463A CN202010294167.9A CN202010294167A CN111539463A CN 111539463 A CN111539463 A CN 111539463A CN 202010294167 A CN202010294167 A CN 202010294167A CN 111539463 A CN111539463 A CN 111539463A
Authority
CN
China
Prior art keywords
image
branch
layer
network model
weight
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202010294167.9A
Other languages
Chinese (zh)
Other versions
CN111539463B (en
Inventor
吴亚
魏守卫
顾嘉奇
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.)
Suzhou Wangao Computer Technology Co ltd
Original Assignee
Suzhou Wangao Computer 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 Suzhou Wangao Computer Technology Co ltd filed Critical Suzhou Wangao Computer Technology Co ltd
Priority to CN202010294167.9A priority Critical patent/CN111539463B/en
Publication of CN111539463A publication Critical patent/CN111539463A/en
Application granted granted Critical
Publication of CN111539463B publication Critical patent/CN111539463B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

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

Abstract

The invention provides a method and a system for realizing image classification by simulating a neuron dendritic branch, which can reduce the training time consumption and improve the image classification precision, and the following steps are carried out on a convolutional layer of a CNN network model: step 1: setting a near branch layer and a far branch layer by simulating a neuron dendritic branch, defining learning weight values corresponding to the near branch layer and the far branch layer respectively, then obtaining a new convolution kernel through multiplication coupling, and performing convolution operation on the new convolution kernel and an input image to extract characteristics to obtain an output image; step 2: the input image is cut in the dimension of the number of image channels, the numerical values of elements of the cut image in the same dimension are compared bit by bit, the maximum value is taken, the image with the maximum value is subjected to convolution operation with a convolution kernel to extract features, and the output image is obtained.

Description

Method and system for realizing image classification by simulating neuron dendritic branches
Technical Field
The invention relates to the field of computer vision, in particular to a Convolutional Neural Network (CNN) in deep learning, and specifically relates to a method and a system for realizing image classification by simulating a neuron dendritic branch.
Background
Currently, the field of computer vision has been greatly developed by means of deep learning technology, and supervised deep learning is connection-oriented learning based on a neural network. Convolutional neural networks CNN are deep neural network models that are widely used. With the addition of big data and GPU computing power, deep learning has made breakthrough progress in many areas, such as visual image understanding, attribute recognition, object detection, natural language processing, and even automatic driving.
However, the defects of CNN still limit the development of the computer vision field. One is that there is an interpretability problem; secondly, the training of the network model is long in time consumption and large in occupied space of parameters. And thirdly, the recognition accuracy is not obviously improved along with the continuous deepening of the neural network depth.
Disclosure of Invention
Aiming at the problems, the invention provides a method and a system for realizing image classification by simulating the dendritic branches of neurons, which can reduce the time consumption of training and improve the image classification precision.
The technical scheme is as follows: a method for achieving image classification by simulating a neuronal dendritic branching, comprising:
preprocessing an image to be classified; establishing a CNN network model, wherein the CNN network model at least comprises a convolution layer and a pooling layer; training the CNN network model to obtain a trained CNN network model; carrying out image classification through the trained CNN network model to obtain a classification result, and is characterized in that the following steps are carried out on the convolution layer of the CNN network model:
step 1: setting a near branch layer and a far branch layer by simulating a neuron dendritic branch, defining learning weight values corresponding to the near branch layer and the far branch layer respectively, then obtaining a new convolution kernel through multiplication coupling, and performing convolution operation on the new convolution kernel and an input image to extract characteristics to obtain an output image;
step 2: the input image is cut in the dimension of the number of image channels, the numerical values of elements of the cut image in the same dimension are compared bit by bit, the maximum value is taken, the image with the maximum value is subjected to convolution operation with a convolution kernel to extract features, and the output image is obtained.
Further, the execution order of step 1 and step 2 can be interchanged.
Further, step 1 comprises the following steps:
step 101: setting a near branch layer and a far branch layer on a simulated neuron dendritic branch, setting a convolution kernel on a convolution layer of a CNN network model, respectively representing the learning weight of the convolution kernel as a near branch weight W and a far branch weight K, initializing the near branch weight W and the far branch weight K, representing the shape of the near branch weight W as W (filter _ height, t filter _ width,1, out _ channels), and representing the shape of the far branch weight K as K (1,1, in _ channels, out _ channels), wherein four dimensions of the near branch weight W and the far branch weight K are respectively represented as convolution kernel height, convolution kernel width, input channel number and output channel number;
step 102: expanding the near branch weight W and the far branch weight K, and converting the near branch weight W and the far branch weight K into the same shape, so that the near branch weight W and the far branch weight K can be multiplicatively coupled;
step 103: carrying out multiplication operation on the expanded near branch weight W and far branch weight K to obtain a new convolution kernel R;
step 104: performing convolution operation on an input image and a convolution kernel R to extract characteristics, outputting an image X, wherein the shape of the image X is represented as X (pitch, height, width, channels), the 4 dimensions of the image X are represented as the number of samples, the height of the image, the width of the image and the number of image channels, and processing the output image through a nonlinear activation function.
Further, in step 102, when the near branch weight W and the far branch weight K are expanded, the near branch weight W is copied in the dimension of the input channel number according to the dimension 3, the copy multiple is [1,1, in _ channels,1], and the expanded tensor is W _ tilled; and copying the far branch weight K in the dimension of the input channel number according to the dimensions 1 and 2, wherein the copying multiple is multiplys [ filter _ height, filter _ width,1,1], and the expanded tensor is K _ TILED.
Further, step 2 comprises the following steps:
step 201: the input image isThe image channels are divided in dimension, the shape of the image is represented as X (pitch, height, width, channels), and the shapes of the image X1 and the image X2 generated after the division are represented as X (pitch, height, width, channels)
Figure BDA0002451542950000021
Wherein, the number of image channels is even;
step 202: comparing the values of the elements in the same dimension of X1 and X2 bit by bit, and taking the maximum value Xmax＝max(X1,X2)，XmaxIs expressed as
Figure BDA0002451542950000022
Step 203: initializing a near branch weight W, and obtaining the image X obtained in the step 202maxConvolution operation is carried out on the near branch weight W to extract features, images are output, and the output images are processed through a nonlinear activation function.
Further, the nonlinear activation function adopts a ReLU or Leaky-ReLU nonlinear activation function.
Further, the CNN network model employs a ResNet or inclusion neural network.
A system for image classification by simulation of neuronal dendritic branching comprising: the neural network unit is used for preprocessing the image to be classified; establishing a CNN network model, wherein the CNN network model at least comprises a convolution layer and a pooling layer; training the CNN network model to obtain a trained CNN network model; carrying out image classification through the trained CNN network model to obtain a classification result;
the method is characterized in that: the neural network unit includes:
the branch module is used for setting a near branch layer and a far branch layer in a convolutional layer of a CNN network model by simulating a neuron dendritic branch, defining learning weight values corresponding to the near branch layer and the far branch layer respectively, then obtaining a new convolutional kernel through multiplication coupling, and extracting characteristics through carrying out convolutional operation on the new convolutional kernel and an input image to obtain an output image;
and the maximum value extracting module is used for cutting the input image in the dimension of the image channel number in the convolution layer of the CNN network model, comparing the numerical values of elements of the cut image in the same dimension bit by bit and extracting the maximum value, and then performing convolution operation on the image with the maximum value and a convolution kernel to extract the characteristics to obtain the output image.
An apparatus for implementing image classification by simulating neuron dendritic branches, comprising a memory, a processor and a program stored in the memory and executable on the processor, wherein the processor implements the method for implementing image classification by simulating neuron dendritic branches.
A computer-readable storage medium on which a program is stored, characterized in that: the program, when executed by a processor, implements a method for image classification by simulating neuronal dendritic branching as described above.
The method for realizing image classification by simulating the neuron dendritic branches comprises the steps of arranging a near branch layer and a far branch layer by the simulated neuron dendritic branches, wherein the near branch is closer to a cell body and extends to a farther area of a space and belongs to coarse distribution of the space, the far branch is branched on the near branch, a small-range space is covered, the density is higher, the influence of input on the cell body can be better simulated, and the stability of similar characteristics is realized. When the image is input, the far-end similar features or slightly different features are constrained to the near branch input, so that stable input is provided for the cell body, in other words, the slight difference of the far-end input features is eliminated, and the anti-interference performance is enhanced; for example, when the stripe line segment images at different positions are input into the far-end branch extended from the same near branch, the same input signal is generated on the near branch, that is, a stable line segment feature detector with a constant position is formed; from a gradient descent learning method commonly adopted in deep learning, a trunk path of a near branch can be learned once every time a far branch is learned, so that the learning frequency of a near branch weight is far more than that of a far branch, and a better support is provided for the stability of output characteristics;
in addition, the input image is cut in the dimension of the number of image channels, the numerical values of elements in the same dimension of the cut image are compared bit by bit, the maximum value is taken, the image and a convolution kernel are subjected to convolution operation to extract features, the output image is obtained, similar features can be stabilized on the output features, the stability of feature characterization is improved, the maximum value of the two features is taken as a stable feature and is transmitted to the next layer, and half of parameter quantity is reduced; the classification precision is improved because the maximum value is taken, and other interference characteristics are removed;
when the obtained neural network is used for image classification, the weight parameter quantity can be reduced, the time consumption of training is reduced, the feature stability is improved, and the image classification precision is improved.
Drawings
FIG. 1 is a schematic diagram of a branched structure of a neuron dendrite;
FIG. 2 is a flow chart of the main part of the method for realizing image classification by simulating the branch of neuron dendrites according to the present invention;
FIG. 3 is a schematic diagram illustrating the implementation of step 1 in the method for image classification by simulating neuron dendritic branches according to the present invention;
FIG. 4 is a schematic diagram illustrating the implementation of step 2 in the method for image classification by simulating neuron dendritic branches according to the present invention.
Detailed Description
The following detailed description of the preferred embodiments of the present invention, taken in conjunction with the accompanying drawings, will make the advantages and features of the invention easier to understand by those skilled in the art, and thus will clearly and clearly define the scope of the invention.
Referring to fig. 1, fig. 2, fig. 3 and fig. 4, the method for realizing image classification by simulating the dendritic branches of neurons of the invention comprises the following steps:
preprocessing an image to be classified; establishing a CNN network model, wherein the CNN network model at least comprises a convolution layer and a pooling layer; training the CNN network model to obtain a trained CNN network model; carrying out image classification through the trained CNN network model to obtain a classification result, and executing the following steps on the convolution layer of the CNN network model:
step 1: the simulated neuron dendritic branches are provided with a near branch layer and a far branch layer, referring to a neuron structure diagram of figure 1, the dendrites of neuron cell bodies are provided with multiple layers of branches, and the branches close to the cell bodies are near branches and are thicker; repeatedly branching the near branches far away from the cell bodies, gradually thinning the branches to form far branches, defining learning weight values respectively corresponding to the near branch layer and the far branch layer, then obtaining a new convolution kernel through multiplication coupling, and performing convolution operation on the new convolution kernel and an input image to extract characteristics to obtain an output image;
step 2: the input image is cut in the dimension of the number of image channels, the numerical values of elements of the cut image in the same dimension are compared bit by bit, the maximum value is taken, the image with the maximum value is subjected to convolution operation with a convolution kernel to extract features, and the output image is obtained.
Specifically, the step 1 comprises the following steps:
step 101: setting a near branch layer and a far branch layer on a simulated neuron dendritic branch, setting a convolution kernel on a convolution layer of a CNN network model, respectively representing the learning weight of the convolution kernel as a near branch weight W and a far branch weight K, initializing the near branch weight W and the far branch weight K, representing the shape of the near branch weight W as W (filter _ height, t filter _ width,1, out _ channels), and representing the shape of the far branch weight K as K (1,1, in _ channels, out _ channels), wherein four dimensions of the near branch weight W and the far branch weight K are respectively represented as convolution kernel height, convolution kernel width, input channel number and output channel number;
step 102: expanding the near branch weight W and the far branch weight K, converting the near branch weight W and the far branch weight K into the same shape, enabling the near branch weight W and the far branch weight K to be capable of being multiply coupled, when expanding the near branch weight W and the far branch weight K, copying the near branch weight W on the dimension of the input channel number according to the dimension 3, wherein the copy multiple is multiplys ═ 1,1, in _ channels,1], and the expanded tensor is W _ TILED; copying the far branch weight K in the dimension of the input channel number according to the dimensions 1 and 2, wherein the copying multiple is multiplys [ filter _ height, filter _ width,1,1], and the expanded tensor is K _ TILED;
step 103: performing multiplication operation on the expanded near branch weight W and far branch weight K, and performing multiplication coupling on W _ TILED and K _ TILED to obtain a new convolution kernel R, wherein the shape of the convolution kernel R is (filter _ height, filter _ width, in _ channels, out _ channels);
step 104: performing convolution operation on an input image and a convolution kernel R to extract characteristics, outputting an image X, wherein the shape of the image X is represented as X (pitch, height, width, channels), the 4 dimensions of the image X are respectively represented as the number of samples, the height of the image, the width of the image and the number of image channels, and processing the output image through a ReLU or a Leaky-ReLU nonlinear activation function.
Specifically, the step 2 comprises the following steps:
step 201: dividing the input image in the dimension of the image channel number, wherein the shape of the image is represented as X (batch, height, width, channels), and the shapes of the image X1 and the image X2 generated after division are represented as X1 and X2
Figure BDA0002451542950000051
Wherein, the number of image channels is even;
step 202: comparing the values of the elements in the same dimension of X1 and X2 bit by bit, and taking the maximum value Xmax＝max(X1,X2)，XmaxIs expressed as
Figure BDA0002451542950000052
Step 203: initializing a near branch weight W, and obtaining the image X obtained in the step 202maxConvolution operation is carried out on the near branch weight W to extract features, images are output, and the output images are processed through a ReLU or Leaky-ReLU nonlinear activation function.
In addition, the execution sequence of the step 1 and the step 2 in the method for realizing image classification by simulating the neuron dendritic branches can be interchanged, namely, the step 1 can be executed first, and then the step 2 can be executed; step 2 may be performed first, and then step 1 may be performed.
In addition, in the method for realizing image classification by simulating the neuron dendritic branches, the CNN network model can adopt a ResNet or inclusion neural network, the content executed in the steps 1 and 2 is the operation in each convolution layer of the CNN network model, and the operation in other layers of the CNN network model can be executed by referring to the specific structure of the network model, and the scheme does not change in the arrangement of other layers, so that the description is omitted.
The following provides a specific application of the method for realizing image classification by simulating the dendritic branches of neurons, and the specific application is as follows:
preprocessing an image to be classified; establishing a CNN network model, wherein the CNN network model at least comprises a convolution layer and a pooling layer; training the CNN network model to obtain a trained CNN network model; carrying out image classification through the trained CNN network model to obtain a classification result;
setting input images X (batch, height, width, in _ channels), out _ channels, stripes and padding, wherein batch is the number of pictures, width is the picture width, height is the picture height, in _ channel is the number of picture channels, out _ channels is the number of output channels, stripes is the step size of each dimension of the image during convolution, the step size is a one-dimensional vector, default [ batch, height, width, channels ] depends on input data _ format, and generally setting [1, stripes, 1], and the first bit and the last bit are 1; padding is string type, and has values of "SAME" and "VALID", indicating the form of convolution, whether the boundary is considered. "SAME" is considered boundary, insufficient time to fill the surrounding with 0, and "VALID" is not considered;
specific parameters in one of the convolutional layers of the CNN network model are "batch" 128, height "32, width" 32, in _ channels "64, out _ channels" 128, threads ═ 1,1,1,1 ", padding" SAME ".
The input image is processed in the convolution layer of the CNN network model and the following steps are executed:
step 101: initializing a near branch weight W and a far branch weight K. The near branch weight W is initialized, the shape is (3,3, 1, 128), the random value is initialized, and the standard deviation is
Figure BDA0002451542950000061
Initializing a far branch weight value K in the shape ofK (1,1,64,128), random value initialization, and standard deviation of
Figure BDA0002451542950000062
Step 102: expanding the near branch weight W and the far branch weight K, expanding the near branch weight W according to multiplys ═ 1,1,64,1, and the expanded W _ TILED is (3,3,64, 128); the distant branch weight K is expanded according to multiplys ═ 3,3,1, and the shape of the expanded K _ TILED is (3,3,64, 128);
step 103: multiplying the elements at the same positions of the W _ TILED and the K _ TILED to obtain a new convolution kernel R, wherein the shape of the new convolution kernel R is (3,3,64, 128);
step 104: performing an operation f (X) on the input image and the convolution kernel R, wherein f is ∑ Rx + b, extracting features, outputting an image X, wherein the shape of the image X is represented as X (128,32,32,64), and outputting the image with ReLU or improvement thereof: Leaky-ReLU nonlinear activation function processing;
step 201, the image X obtained in step 104 is divided, X (128,32,32,64) is divided into two parts in the channels dimension, wherein the dimension is even number and is selected from
Figure BDA0002451542950000071
Processing the X generated after the division1And X2Are all
Figure BDA0002451542950000072
Step 202, calculating X1And X2Is measured. Comparing X bit by bit1And X2Elements in the same position, taking the maximum value Xmax＝max(X1,X2),XmaxIs in the shape of
Figure BDA0002451542950000073
Transmitting the data to the next layer by taking the maximum value of the two characteristics as a stable characteristic; the classification precision is improved because the maximum value is taken, and other interference characteristics are removed, so that the classification precision is improved;
step 203, the image X obtained in step 202 is processedmaxConvolution operation is carried out on the near branch weight W to extract features, and the near branch weight is initialized
Figure BDA0002451542950000074
And calculating f (x) ∑ Wxmax+ b, the output of the cell body is processed by ReLU or its modified Leaky-ReLU nonlinear activation function, the shape of the output of the cell body is (128,32, 128), and it can be seen that the input channel of the near branch weight W is halved, so the weight parameter number is reduced, and the training time is reduced.
The method for realizing image classification by simulating the neuron dendritic branches comprises the steps of arranging a near branch layer and a far branch layer by the simulated neuron dendritic branches, wherein the near branch is closer to a cell body and extends to a farther area of a space and belongs to coarse distribution of the space, the far branch is branched on the near branch, a small-range space is covered, the density is higher, the influence of input on the cell body can be better simulated, and the stability of similar characteristics is realized. When the image is input, the far-end similar features or slightly different features are constrained to the near branch input, so that stable input is provided for the cell body, in other words, the slight difference of the far-end input features is eliminated, and the anti-interference performance is enhanced; for example, when the stripe line segment images at different positions are input into the far-end branch extended from the same near branch, the same input signal is generated on the near branch, that is, a stable line segment feature detector with a constant position is formed; from a gradient descent learning method commonly adopted in deep learning, a trunk path of a near branch can be learned once every time a far branch is learned, so that the learning frequency of a near branch weight is far more than that of a far branch, and a better support is provided for the stability of output characteristics;
in addition, the input image is cut in the dimension of the image channel number, the numerical values of elements in the same dimension of the cut image are compared bit by bit, the maximum value is taken, the image with the maximum value is subjected to convolution operation with the convolution kernel to extract the features, the output image is obtained, the similar features can be stabilized on the output features, the stability of feature characterization is improved, the weight parameter number can be reduced, and the training time is reduced;
when the obtained neural network is used for image classification, the weight parameter quantity can be reduced, the time consumption of training is reduced, the feature stability is improved, and the image classification precision is improved.
In an embodiment of the present invention, there is also provided a system for implementing image classification by simulating a neuron dendritic branch, including: the neural network unit is used for preprocessing the image to be classified; establishing a CNN network model, wherein the CNN network model at least comprises a convolution layer and a pooling layer; training the CNN network model to obtain a trained CNN network model; carrying out image classification through the trained CNN network model to obtain a classification result;
the neural network unit includes:
the branch module is used for setting a near branch layer and a far branch layer in a convolutional layer of a CNN network model by simulating a neuron dendritic branch, defining learning weight values corresponding to the near branch layer and the far branch layer respectively, then obtaining a new convolutional kernel through multiplication coupling, and extracting characteristics through carrying out convolutional operation on the new convolutional kernel and an input image to obtain an output image;
and the maximum value extracting module is used for cutting the input image in the dimension of the image channel number in the convolution layer of the CNN network model, comparing the numerical values of elements of the cut image in the same dimension bit by bit and extracting the maximum value, and then performing convolution operation on the image with the maximum value and a convolution kernel to extract the characteristics to obtain the output image.
In an embodiment of the present invention, an apparatus for implementing image classification by simulating neuron dendritic branches is further provided, and the apparatus includes a memory, a processor, and a program stored in the memory and executable on the processor, where the processor implements the above-mentioned method for implementing image classification by simulating neuron dendritic branches when executing the program.
In the above-mentioned implementation of the apparatus for implementing image classification by simulating the dendrite branches of neurons, the memory and the processor are electrically connected directly or indirectly to implement data transmission or interaction. For example, the components may be electrically connected to each other via one or more communication buses or signal lines, such as a bus. The memory stores computer-executable instructions for implementing the data access control method, and includes at least one software functional module which can be stored in the memory in the form of software or firmware, and the processor executes various functional applications and data processing by running the software programs and modules stored in the memory.
The Memory may be, but is not limited to, a Random Access Memory (RAM), a Read Only Memory (ROM), a Programmable Read-Only Memory (PROM), an Erasable Read-Only Memory (EPROM), an electrically Erasable Read-Only Memory (EEPROM), and the like. The memory is used for storing programs, and the processor executes the programs after receiving the execution instructions.
The processor may be an integrated circuit chip having signal processing capabilities. The processor may be a general-purpose processor, and includes a Central Processing Unit (CPU), a Network Processor (NP), and the like. The various methods, steps, and logic blocks disclosed in the embodiments of the present application may be implemented or performed. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
In an embodiment of the present invention, a computer-readable storage medium is also presented, on which a program is stored, which when executed by a processor, implements the method for implementing image classification by simulating neuron dendritic branches as described above.
As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, apparatus, or computer program product. Accordingly, embodiments of the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, embodiments of the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
Embodiments of the present invention are described with reference to flowchart illustrations and/or block diagrams of methods, systems, apparatuses and computer program products according to embodiments of the invention. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing terminal to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal, create means for implementing the functions specified in the flowchart.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing terminal to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart.
These computer program instructions may also be loaded onto a computer or other programmable data processing terminal to cause a series of operational steps to be performed on the computer or other programmable terminal to produce a computer implemented process such that the instructions which execute on the computer or other programmable terminal provide steps for implementing the functions specified in the flowchart.
The method for realizing image classification by simulating the neuron dendritic branches, the system for realizing image classification by simulating the neuron dendritic branches, the device for realizing image classification by simulating the neuron dendritic branches and the application of the computer-readable storage medium provided by the invention are described in detail above, specific embodiments are applied in the method for explaining the principle and the implementation mode of the invention, and the description of the above embodiments is only used for helping to understand the method and the core idea of the invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, there may be variations in the specific embodiments and the application scope, and in summary, the content of the present specification should not be construed as a limitation to the present invention.

Claims (10)

1. A method for achieving image classification by simulating a neuronal dendritic branching, comprising:
preprocessing an image to be classified; establishing a CNN network model, wherein the CNN network model at least comprises a convolution layer and a pooling layer; training the CNN network model to obtain a trained CNN network model; carrying out image classification through the trained CNN network model to obtain a classification result, and is characterized in that the following steps are carried out on the convolution layer of the CNN network model:
step 1: setting a near branch layer and a far branch layer by simulating a neuron dendritic branch, defining learning weight values corresponding to the near branch layer and the far branch layer respectively, then obtaining a new convolution kernel through multiplication coupling, and performing convolution operation on the new convolution kernel and an input image to extract characteristics to obtain an output image;
step 2: the input image is cut in the dimension of the number of image channels, the numerical values of elements of the cut image in the same dimension are compared bit by bit, the maximum value is taken, the image with the maximum value is subjected to convolution operation with a convolution kernel to extract features, and the output image is obtained.
2. The method for image classification by simulation of neuronal dendritic branching according to claim 1, characterized in that: the order of execution of step 1 and step 2 can be interchanged.
3. The method for image classification by simulation of neuronal dendritic branching according to claim 1, characterized in that: the step 1 comprises the following steps:
step 101: setting a near branch layer and a far branch layer on a simulated neuron dendritic branch, setting a convolution kernel on a convolution layer of a CNN network model, respectively representing the learning weight of the convolution kernel as a near branch weight W and a far branch weight K, initializing the near branch weight W and the far branch weight K, representing the shape of the near branch weight W as W (filter _ height, filter _ width,1, out _ channels), and representing the shape of the far branch weight K as K (1,1, in _ channels, out _ channels), wherein four dimensions of the near branch weight W and the far branch weight K are respectively represented as the height of the convolution kernel, the width of the convolution kernel, the number of input channels and the number of output channels;
step 102: expanding the near branch weight W and the far branch weight K, and converting the near branch weight W and the far branch weight K into the same shape, so that the near branch weight W and the far branch weight K can be multiplicatively coupled;
step 103: carrying out multiplication operation on the expanded near branch weight W and far branch weight K to obtain a new convolution kernel R;
step 104: performing convolution operation on an input image and a convolution kernel R to extract characteristics, outputting an image X, wherein the shape of the image X is represented as X (pitch, height, width, channels), the 4 dimensions of the image X are represented as the number of samples, the height of the image, the width of the image and the number of image channels, and processing the output image through a nonlinear activation function.
4. The method for image classification by simulation of neuronal dendritic branching according to claim 3, characterized in that: in step 102, when the near branch weight W and the far branch weight K are expanded, the near branch weight W is copied in the dimension of the input channel number according to the dimension 3, the copy multiple is multiplys ═ 1,1, in _ channels,1], and the expanded tensor is W _ tilled; and copying the far branch weight K in the dimension of the input channel number according to the dimensions 1 and 2, wherein the copying multiple is multiplys [ filter _ height, filter _ width,1,1], and the expanded tensor is K _ TILED.
5. The method for image classification by simulation of neuronal dendritic branching according to claim 1, characterized in that: the step 2 comprises the following steps:
step 201: dividing the input image in the dimension of the image channel number, wherein the shape of the image is represented as X (batch, height, width, channels), and the shapes of the image X1 and the image X2 generated after division are represented as X1 and X2
Figure FDA0002451542940000021
Wherein, the number of image channels is even;
step 202: comparing the numerical value of the elements in the same dimension of X1 and X2 bit by bit, and takingMaximum value Xmax＝max(X1,X2)，XmaxIs expressed as
Figure FDA0002451542940000022
Step 203: initializing a near branch weight W, and obtaining the image X obtained in the step 202maxConvolution operation is carried out on the near branch weight W to extract features, images are output, and the output images are processed through a nonlinear activation function.
6. The method for image classification by simulation of neuronal dendritic branching according to claim 3 or claim 5, characterized in that: the nonlinear activation function adopts a ReLU or Leaky-ReLU nonlinear activation function.
7. The method for image classification by simulation of neuronal dendritic branching according to claim 1, characterized in that: the CNN network model employs a ResNet or inclusion neural network.
8. A system for image classification by simulation of neuronal dendritic branching comprising: the neural network unit is used for preprocessing the image to be classified; establishing a CNN network model, wherein the CNN network model at least comprises a convolution layer and a pooling layer; training the CNN network model to obtain a trained CNN network model; carrying out image classification through the trained CNN network model to obtain a classification result;
the method is characterized in that: the neural network unit includes:
the branch module is used for setting a near branch layer and a far branch layer in a convolutional layer of a CNN network model by simulating a neuron dendritic branch, defining learning weight values corresponding to the near branch layer and the far branch layer respectively, then obtaining a new convolutional kernel through multiplication coupling, and extracting characteristics through carrying out convolutional operation on the new convolutional kernel and an input image to obtain an output image;
and the maximum value extracting module is used for cutting the input image in the dimension of the image channel number in the convolution layer of the CNN network model, comparing the numerical values of elements of the cut image in the same dimension bit by bit and extracting the maximum value, and then performing convolution operation on the image with the maximum value and a convolution kernel to extract the characteristics to obtain the output image.
9. An apparatus for implementing image classification by simulating neuron dendritic branches, comprising a memory, a processor and a program stored on the memory and executable on the processor, wherein the processor implements the method for implementing image classification by simulating neuron dendritic branches according to claim 1 when executing the program.
10. A computer-readable storage medium on which a program is stored, characterized in that: the program, when executed by a processor, implements a method for image classification by simulating neuronal dendritic branching according to claim 1.
CN202010294167.9A 2020-04-15 2020-04-15 Method and system for realizing image classification by simulating neuron dendrite branching Active CN111539463B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010294167.9A CN111539463B (en) 2020-04-15 2020-04-15 Method and system for realizing image classification by simulating neuron dendrite branching

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010294167.9A CN111539463B (en) 2020-04-15 2020-04-15 Method and system for realizing image classification by simulating neuron dendrite branching

Publications (2)

Publication Number Publication Date
CN111539463A true CN111539463A (en) 2020-08-14
CN111539463B CN111539463B (en) 2023-09-15

Family

ID=71974927

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010294167.9A Active CN111539463B (en) 2020-04-15 2020-04-15 Method and system for realizing image classification by simulating neuron dendrite branching

Country Status (1)

Country Link
CN (1) CN111539463B (en)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107256550A (en) * 2017-06-06 2017-10-17 电子科技大学 A kind of retinal image segmentation method based on efficient CNN CRF networks
CN109344888A (en) * 2018-09-19 2019-02-15 广东工业大学 A kind of image-recognizing method based on convolutional neural networks, device and equipment
CN109376859A (en) * 2018-09-27 2019-02-22 东南大学 A kind of neural networks pruning method based on diamond shape convolution
CN109472801A (en) * 2018-11-22 2019-03-15 廖祥 It is a kind of for multiple dimensioned neuromorphic detection and dividing method

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107256550A (en) * 2017-06-06 2017-10-17 电子科技大学 A kind of retinal image segmentation method based on efficient CNN CRF networks
CN109344888A (en) * 2018-09-19 2019-02-15 广东工业大学 A kind of image-recognizing method based on convolutional neural networks, device and equipment
CN109376859A (en) * 2018-09-27 2019-02-22 东南大学 A kind of neural networks pruning method based on diamond shape convolution
CN109472801A (en) * 2018-11-22 2019-03-15 廖祥 It is a kind of for multiple dimensioned neuromorphic detection and dividing method

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
孙文磊: ""Spiking神经网络及其在图像分割中的应用"" *

Also Published As

Publication number Publication date
CN111539463B (en) 2023-09-15

Similar Documents

Publication Publication Date Title
EP4036803A1 (en) Neural network model processing method and apparatus, computer device, and storage medium
US20230083518A1 (en) Image segmentation method, system, and device, and readable storage medium
EP3905194A1 (en) Pose estimation method and apparatus
CN109005398B (en) Stereo image parallax matching method based on convolutional neural network
CN111339935B (en) Optical remote sensing picture classification method based on interpretable CNN image classification model
CN113065997B (en) Image processing method, neural network training method and related equipment
CN106778910A (en) Deep learning system and method based on local training
CN110070867A (en) Voice instruction recognition method, computer installation and computer readable storage medium
US20230153965A1 (en) Image processing method and related device
CN110689136A (en) Deep learning model obtaining method, device, equipment and storage medium
CN114266894A (en) Image segmentation method and device, electronic equipment and storage medium
CN114429208A (en) Model compression method, device, equipment and medium based on residual structure pruning
CN110807428A (en) Coal sample identification method and device, server and storage medium
CN110866609B (en) Method, device, server and storage medium for acquiring interpretation information
CN111967478A (en) Feature map reconstruction method and system based on weight inversion, storage medium and terminal
CN111539463B (en) Method and system for realizing image classification by simulating neuron dendrite branching
CN115294361A (en) Feature extraction method and device
CN115409159A (en) Object operation method and device, computer equipment and computer storage medium
CN114998172A (en) Image processing method and related system
CN114820755A (en) Depth map estimation method and system
CN113963428A (en) Model training method, occlusion detection method, system, electronic device, and medium
CN111539420B (en) Panoramic image saliency prediction method and system based on attention perception features
CN113902933A (en) Ground segmentation network model training method, device, equipment and medium
CN112541438A (en) Text recognition method and device
CN112561933A (en) Image segmentation method and device

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant