CN116562358A - Construction method of image processing Gabor kernel convolutional neural network - Google Patents
Construction method of image processing Gabor kernel convolutional neural network Download PDFInfo
- Publication number
- CN116562358A CN116562358A CN202310270483.6A CN202310270483A CN116562358A CN 116562358 A CN116562358 A CN 116562358A CN 202310270483 A CN202310270483 A CN 202310270483A CN 116562358 A CN116562358 A CN 116562358A
- Authority
- CN
- China
- Prior art keywords
- neural network
- gabor
- convolutional neural
- parameters
- improved
- 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
Links
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 73
- 238000012545 processing Methods 0.000 title claims abstract description 19
- 238000010276 construction Methods 0.000 title claims abstract description 16
- 238000012549 training Methods 0.000 claims abstract description 39
- 230000006870 function Effects 0.000 claims abstract description 36
- 238000000034 method Methods 0.000 claims abstract description 19
- 238000013528 artificial neural network Methods 0.000 claims abstract description 12
- 230000002068 genetic effect Effects 0.000 claims abstract description 8
- 238000011156 evaluation Methods 0.000 claims description 13
- 230000014509 gene expression Effects 0.000 claims description 11
- 239000011159 matrix material Substances 0.000 claims description 5
- 238000005070 sampling Methods 0.000 claims description 5
- 108090000623 proteins and genes Proteins 0.000 claims description 4
- 238000012216 screening Methods 0.000 claims description 4
- 230000008569 process Effects 0.000 abstract description 5
- 238000000605 extraction Methods 0.000 description 10
- 210000004027 cell Anatomy 0.000 description 8
- 238000004364 calculation method Methods 0.000 description 6
- 238000001228 spectrum Methods 0.000 description 4
- 238000005457 optimization Methods 0.000 description 3
- 210000001525 retina Anatomy 0.000 description 3
- 230000000007 visual effect Effects 0.000 description 3
- 208000003098 Ganglion Cysts Diseases 0.000 description 2
- 208000005400 Synovial Cyst Diseases 0.000 description 2
- 239000005557 antagonist Substances 0.000 description 2
- 238000009395 breeding Methods 0.000 description 2
- 230000001488 breeding effect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 210000002287 horizontal cell Anatomy 0.000 description 2
- 230000035772 mutation Effects 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 210000000608 photoreceptor cell Anatomy 0.000 description 2
- 241000124008 Mammalia Species 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 210000000411 amacrine cell Anatomy 0.000 description 1
- 230000003042 antagnostic effect Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000019771 cognition Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 230000005284 excitation Effects 0.000 description 1
- 230000004438 eyesight Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 210000000056 organ Anatomy 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 210000000977 primary visual cortex Anatomy 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 238000002945 steepest descent method Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000004382 visual function Effects 0.000 description 1
Classifications
-
- 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
- G06N3/086—Learning methods using evolutionary algorithms, e.g. genetic algorithms or genetic programming
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- General Physics & Mathematics (AREA)
- Computing Systems (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Multimedia (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Medical Informatics (AREA)
- Databases & Information Systems (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Physiology (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to the technical field of image classification processing, and provides a construction method of an image processing Gabor kernel convolutional neural network. The method comprises the following steps: based on the Gaussian envelope functions of the center convex and the center concave and the sinusoidal grating functions, obtaining an optimized and improved Gabor filter and a shallow layer of the Gabor kernel convolutional neural network through an initial training sample set and a plurality of group genetic algorithms; based on an optimized and improved Gabor filter, training Gabor kernel convolutional neural network layers except for a shallow layer of the Gabor kernel convolutional neural network through a back propagation algorithm to obtain a target Gabor kernel convolutional neural network, inputting a training sample set into the target Gabor kernel convolutional neural network, extracting depth features, and classifying. The method effectively reduces the time consumption and the memory requirement of the neural network in the training process, and prevents the network classification accuracy from being reduced. The application range is wide, and the robustness is strong.
Description
Technical Field
The invention relates to the technical field of image classification processing, in particular to a construction method of an image processing Gabor kernel convolutional neural network.
Background
The deep learning algorithm represented by the convolutional neural network is widely focused due to strong modeling capability and deep feature extraction capability, and is successfully applied to the fields of natural language processing, image recognition and the like. The convolutional neural network is an artificial neural network which is established by simulating the action mechanism of human neurons under external excitation, a black box model of an image-result does not need to manually extract features, and the convolutional neural network has stronger intelligence, and particularly has obvious advantages when the information is complex, the background information is unclear, the classification rule is difficult to express, and the target appearance features are distorted and deformed.
The multi-layer feature extraction structure of the convolutional neural network is widely applied to various cognition and identification problems, and the high-efficiency expression capability of the multi-layer feature extraction structure on the target depth features makes the multi-layer feature extraction structure the most common target feature extraction method. However, feature extraction of conventional convolutional neural networks is a pure data-driven approach, i.e., learning robust expressions from data, but usually at the cost of higher computation and time. The deep architecture and complex training of conventional convolutional neural networks requires intensive computational effort on modern computing platforms. Trainable weight kernel learning requires enormous computation and time consumption, and complex computational requirements have hampered its practical application.
Disclosure of Invention
In view of the above, the invention provides a construction method of an image processing Gabor kernel convolutional neural network, which aims to solve the technical problems of large calculation amount, high algorithm complexity and high time consumption in the prior art.
The invention provides a construction method of an image processing Gabor kernel convolutional neural network, which comprises the following steps:
s0. randomly sampling the constructed training sample set to construct an initial training sample set;
s1, obtaining a Gabor filter and a shallow layer of a Gabor kernel convolutional neural network which are optimized and improved through the initial training sample set and a plurality of group genetic algorithms based on a Gaussian envelope function and a sinusoidal grating function of a central convex part and a central concave part;
s2, training Gabor kernel convolutional neural network layers except for a shallow layer of the Gabor kernel convolutional neural network through a back propagation algorithm based on the optimized and improved Gabor filter to obtain a target Gabor kernel convolutional neural network;
s3, inputting the training sample set into the target Gabor kernel convolutional neural network, extracting depth features and classifying, wherein the depth features comprise textures and edges.
Further, the step S1 includes:
s11, constructing an improved Gabor filter based on a Gaussian envelope function and a sinusoidal grating function of the center convex and the center concave;
s12, randomly generating an initial population containing individuals with set scale, wherein the base factor of each individual in the initial population is 4 times of the number of the improved Gabor filters, and the genes of each individual represent standard deviation parameters and aspect ratio parameters of Gaussian envelope functions of the improved Gabor filters, and wavelength parameters and phase difference parameters of sinusoidal gratings;
s13, calculating global errors of all individuals in multiple groups by adopting the initial training sample set, taking the global errors as an evaluation function, optimizing standard deviation parameters, length-width ratio parameters and wavelength parameters and phase difference parameters of a sinusoidal grating of the Gaussian envelope function of the improved Gabor filter through propagation, intersection and variation, screening out target individuals with the global errors reaching a preset value, and storing the target individuals to a next generation group, wherein the target individuals comprise individuals with the previous generation loss value reaching the preset value;
s14, repeating the step S13 until all standard deviation parameters and length-width ratio parameters of the Gaussian envelope function, wavelength parameters and phase difference parameters of the sinusoidal grating are optimized or preset reproduction algebra is achieved, returning all target individuals to the improved Gabor filter to obtain the optimized and improved Gabor filter, and taking the optimized and improved Gabor filter as a convolution kernel to form a shallow layer of the Gabor kernel convolution neural network.
Further, in S13, the obtaining of the global error includes: the two norms of the difference between all sample predictions and standard values obtained by Gabor kernel convolutional neural network.
Further, in S13, the evaluation function includes the following expression:
wherein y is m ' is the predicted value of the mth sample through Gabor CNN, y m Is the standard value of the mth sample, and n represents the number of samples.
Further, the target individual is screened by the following expression:
wherein M is d ＝{(σ 1 ,γ 1 ,λ 1 ,ψ 1 ),(σ 2 ,γ 2 ,λ 2 ,ψ 2 ),...,(σ m ,γ m ,λ m ,ψ m )} d M=1, 2,..n represents the population set in the d-th iteration.
Further, the step S2 includes:
s21, taking the optimized and improved Gabor filter as a convolution kernel to form a shallow layer of a Gabor kernel convolution neural network;
s22, training Gabor kernel convolutional neural network layers except the shallow layer of the Gabor kernel convolutional neural network through a back propagation algorithm to obtain a target Gabor kernel convolutional neural network.
Further, in S22, the training, by the back propagation algorithm, the Gabor kernel convolutional neural network layer other than the shallow layer of the Gabor kernel convolutional neural network further includes: and updating parameter weight kernels except standard deviation parameters and length-width ratio parameters of the Gaussian envelope function, wavelength parameters and phase difference parameters of the sinusoidal grating by adopting the back propagation algorithm, and updating weights by a bp algorithm.
Further, the update weight is performed as follows:
wherein DeltaW is l Representing the gradient of the l-layer weight matrix, ΔB l Represents the gradient of the bias vector of the layer, eta is the learning rate, W l For the l-layer weight matrix, B l Is the layer bias vector.
Compared with the prior art, the invention has the beneficial effects that:
1. the Gabor kernel convolutional neural network constructed by the invention eliminates most gradient calculation and weight updating operations, and greatly reduces the calculation amount and the complexity of calculation requirements.
2. The method can effectively reduce the time consumption and the memory requirement of the neural network in the training process on the premise of ensuring the network capacity.
3. The construction method adopted by the invention realizes high-efficiency expression of the characteristics through inherent fault tolerance of the network and improvement of the Gabor filter, and prevents the reduction of the classification precision of the network.
4. The improved Gabor filter and the optimization method are suitable for most neural network structures, can be applied to database image classification tasks such as MNIST and the like, and are wide in application range and strong in robustness.
Drawings
In order to more clearly illustrate the technical solutions of the present invention, the following description will briefly explain the drawings used in the embodiments or the description of the prior art, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings can be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flowchart of a method for constructing an image processing Gabor kernel convolutional neural network according to an embodiment of the present invention;
fig. 2 is a schematic flow chart of a method for constructing an image processing Gabor kernel convolutional neural network according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of an embodiment of a Gabor convolutional neural network of LeNet structure provided by the present invention;
fig. 4 is a schematic diagram of an improved gaussian envelope and Gabor filter provided by an embodiment of the present invention.
Detailed Description
In the following description, for purposes of explanation and not limitation, specific details are set forth such as the particular system architecture, techniques, etc., in order to provide a thorough understanding of the embodiments of the present invention. It will be apparent, however, to one skilled in the art that the present invention may be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present invention with unnecessary detail.
The following describes in detail an image processing Gabor kernel convolutional neural network construction method according to the present invention with reference to the accompanying drawings.
As shown in fig. 1, the Gabor kernel convolutional neural network construction method includes:
s0. randomly sampling the constructed training sample set to construct an initial training sample set;
s1, obtaining an optimized and improved Gabor filter and a shallow layer of the Gabor kernel convolutional neural network through the initial training sample set and a plurality of group genetic algorithms based on a Gaussian envelope function and a sinusoidal grating function of a central convex part and a central concave part;
the S1 comprises the following steps:
s11, constructing an improved Gabor filter based on a Gaussian envelope function and a sinusoidal grating function of the center convex and the center concave;
the improved Gabor filter simulates the center-periphery concentric circle antagonistic receptive field characteristics of single cells in a mammal biological vision system, retains the receptive field center characteristic extraction capability of the Gabor filter, and increases the attention of the filter to the receptive field periphery characteristics.
Wherein the modified Gabor filter comprises a modified two-dimensional Gabor filter. For example, a Gabor convolutional neural network is constructed with a modified two-dimensional Gabor filter as a convolutional kernel introduced into an initial convolutional neural network shallow convolutional layer generated with a small number of samples and training times.
In particular, mammalian retina is an important organ for sensing external environment and is also a high-efficiency feature extraction structure. It is the core of visual function, which encodes the optical signals of the visual world into a sequence of digital pulse signals and delivers them to the brain. The retina contains mainly five types of neuronal cells: photoreceptor cells (rods and cones), bipolar cells, ganglion cells, horizontal cells, and amacrine cells. The horizontal cells send negative feedback signals to the photoreceptor cells and bipolar cells, so that bipolar cells and ganglion cells at the rear of the bipolar cells form a central-peripheral concentric circle antagonist type visual receptive field. One of the significant features of most primary visual cortex cells is the strong azimuthal selectivity, i.e., they react to the azimuth (optimal azimuth) of the respective specific fringes or stubs in the visual field, while being very insensitive or unresponsive to other azimuth.
The invention adopts the Gaussian envelope of the improved two-dimensional Gabor filter, and provides a center-periphery concentric circle antagonist type two-dimensional Gabor filter to simulate the characteristics of two types of retina receptive fields. In the modified Gabor filter, there are two gaussian envelopes to simulate the center on and center off receptive fields, and the modified gaussian envelope spectrum can be expressed as:
wherein, the spectrum is a central convex Gaussian envelope spectrum and is used for simulating a central on-type receptive field; is a central concave Gaussian envelope spectrum and is used for simulating a central off receptive field. In the improved two-dimensional Gabor filter, the standard deviation sigma of the Gaussian envelope controls the size of the Gabor filter receptive field, the aspect ratio gamma of the Gaussian envelope controls the external shape of the Gabor filter, the wavelength lambda and the direction theta of the sinusoidal grating respectively control the wavelength and the direction of the Gabor filter, and the phase difference phi controls the distance between the sinusoidal grating and the receptive field center.
S12, randomly generating an initial population containing individuals with set scale, wherein the base factor of each individual in the initial population is 4 times of the number of the improved Gabor filters, and the genes of each individual represent standard deviation parameters and aspect ratio parameters of Gaussian envelope functions of the improved Gabor filters, and wavelength parameters and phase difference parameters of sinusoidal gratings;
s13, calculating global errors of all individuals in multiple groups by adopting the initial training sample set, taking the global errors as an evaluation function, optimizing standard deviation parameters, length-width ratio parameters and wavelength parameters and phase difference parameters of a sinusoidal grating of the Gaussian envelope function of the improved Gabor filter through propagation, intersection and variation, screening out target individuals with the global errors reaching a preset value, and storing the target individuals to a next generation group, wherein the target individuals comprise individuals with the previous generation loss value reaching the preset value;
and randomly extracting a small number of samples from all samples, inputting the samples into an initially constructed Gabor convolutional neural network, calculating global errors between network output and sample labels, and adopting a plurality of group genetic algorithms as evaluation indexes to optimize standard deviation parameters and aspect ratio parameters of Gaussian envelopes, wavelength parameters and phase difference parameters of sinusoidal gratings for controlling the shape of the improved Gabor filter in a set interval, thereby realizing the optimized training of the improved Gabor filter.
Wherein the random extraction includes proportionally random extraction.
One suitable evaluation function is the key of optimizing various swarm genetic algorithms, and the evaluation function is used for controlling the value direction of variables in the optimization of various swarm genetic algorithms, including the solution of the problem to be optimized.
In the step S13, the obtaining of the global error includes: the two norms of the difference between all sample predictions and standard values obtained by Gabor kernel convolutional neural network.
Wherein global error is an important indicator of the accuracy of the reaction network. In the present invention, global error is employed as an evaluation function.
In S13, the evaluation function includes the following expression:
wherein y is m ' is the predicted value of the mth sample through Gabor CNN, y m Is its standard value, n represents the number of samples.
The screening of the target individuals comprises the following expression:
wherein M is d ＝{(σ 1 ,γ 1 ,λ 1 ,ψ 1 ),(σ 2 ,γ 2 ,λ 2 ,ψ 2 ),...,(σ m ,γ m ,λ m ,ψ m )} d M=1, 2,..n represents the population set in the d-th iteration.
S14, repeating the step S13 until all standard deviation parameters and length-width ratio parameters of the Gaussian envelope function, wavelength parameters and phase difference parameters of the sinusoidal grating are optimized or preset reproduction algebra is achieved, returning all target individuals to the improved Gabor filter to obtain the optimized and improved Gabor filter, and taking the optimized and improved Gabor filter as a convolution kernel to form a shallow layer of the Gabor kernel convolution neural network.
S2, training the Gabor kernel convolutional neural network layers except the shallow layer of the Gabor kernel convolutional neural network through a back propagation algorithm based on the optimized and improved Gabor filter to obtain a target Gabor kernel convolutional neural network.
S21, taking the optimized and improved Gabor filter as a convolution kernel to form a shallow layer of a Gabor kernel convolution neural network;
s22, training Gabor kernel convolutional neural network layers except the shallow layer of the Gabor kernel convolutional neural network through a back propagation algorithm to obtain a target Gabor kernel convolutional neural network.
In S22, the training, by the back propagation algorithm, the Gabor kernel convolutional neural network layer other than the shallow layer of the Gabor kernel convolutional neural network further includes: and updating parameter weight kernels except standard deviation parameters and length-width ratio parameters of the Gaussian envelope function, wavelength parameters and phase difference parameters of the sinusoidal grating by adopting the back propagation algorithm, and updating weights by a bp algorithm.
The learning rule of the method is that the steepest descent method is used, and the weight and the threshold value of the network are continuously adjusted through Back Propagation, so that the square sum of errors of the network is minimum, and the training speed is high.
The weights include the following expression:
wherein eta is learning rate, W l For the l-layer weight matrix, B l Is the layer bias vector.
S3, inputting the training sample set into the target Gabor kernel convolutional neural network, extracting depth features and classifying, wherein the depth features comprise textures and edges.
The invention adopts a few samples and a plurality of group genetic algorithms, so that most gradient calculation and weight updating operations are eliminated, and a large amount of calculation amount is greatly reduced; on the premise of ensuring the network capability, the time consumption and the memory requirement of the neural network in the training process can be effectively reduced; the high-efficiency expression of the characteristics is realized through the inherent fault-tolerant capability of the network and the improved Gabor filter, and the degradation of the classification precision of the network is prevented; the improved Gabor filter and the optimization method are suitable for most of neural network structures, and have wide application range and strong robustness.
Any combination of the above optional solutions may be adopted to form an optional embodiment of the present application, which is not described herein in detail.
Example 1
As shown in fig. 3, the structure is as follows: a total of 5 layers, including 2 convolutional layers, each followed by a max pooling layer, with a full join layer to produce the final result. The first convolution layer has 6 modified Gabor filters as convolution kernels, 5 x 5 in size. The second convolution layer contains 72 conventional random kernels that extract 12 features for each input, 3 x 3 in size. The 10-dimensional column vector finally output by the full connection layer is used as a classification result.
Taking classification tasks of MNIST database as an example, the training method of the Gabor kernel convolutional neural network of the structure comprises the following steps:
(1) Random sampling is carried out in an MNIST database at a sampling rate of 0.1, and a traditional convolutional neural network of a LeNet structure is trained for 10 to 20 times to serve as an initial evaluation structure. And randomly selecting standard deviation parameters and length-width ratio parameters of the Gaussian envelope of the improved two-dimensional Gabor filter, wavelength parameters and phase difference parameters of the sinusoidal grating in the [0.1,10] interval. The initial scale of random generation is a population containing 20 individuals, the base factor of each individual in the population is the number 24 of Gabor filters, and the control parameters of 6 two-dimensional Gabor filters are represented. The number of the center on type Gabor filters is equal to that of the center off type Gabor filters, and the direction parameter theta bisects the direction space.
(2) Calculating global errors of each individual in the population by using the local samples and the evaluation structures randomly sampled in the step (1), generating a next generation population by breeding, crossing and mutation, wherein the breeding rate is 1, the crossing rate is set to be 0.8, and the mutation rate is set to be 0.6. And in each generation, sequencing global errors of individuals, selecting the individuals with the smallest global errors, namely reaching a preset value, as target individuals, and storing the target individuals into a next generation population.
(3) And (3) repeating the step (2) until the global error of the optimal individual is not reduced or the number of reproduction algebra reaches 25, and constructing a filter to replace a random convolution kernel in a 1 st convolution layer in the evaluation structure by utilizing the two-dimensional Gabor filter parameters represented by the most individual genes of the last generation.
(4) And training the rest parameters except the Gabor filter in the Gabor kernel convolutional neural network by adopting an error back-propagation algorithm and all samples of the MNIST database, wherein the learning rate is set to be 1, the grouping size is set to be 50, and the maximum training times are set to be 500 times. When the network is not converged or reaches the maximum training times, completing the construction and training of the Gabor convolutional neural network, and obtaining a target Gabor kernel convolutional neural network; the target Gabor kernel convolutional neural network obtained by the method can be applied to MNIST database image classification tasks.
It should be understood that the sequence number of each step in the foregoing embodiment does not mean that the execution sequence of each process should be determined by the function and the internal logic, and should not limit the implementation process of the embodiment of the present invention.
The above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention, and are intended to be included in the scope of the present invention.
Claims (8)
1. The construction method of the image processing Gabor kernel convolutional neural network is characterized by comprising the following steps of:
s0. randomly sampling the constructed training sample set to construct an initial training sample set;
s1, obtaining a Gabor filter and a shallow layer of a Gabor kernel convolutional neural network which are optimized and improved through the initial training sample set and a plurality of group genetic algorithms based on a Gaussian envelope function and a sinusoidal grating function of a central convex part and a central concave part;
s2, training Gabor kernel convolutional neural network layers except for a shallow layer of the Gabor kernel convolutional neural network through a back propagation algorithm based on the optimized and improved Gabor filter to obtain a target Gabor kernel convolutional neural network;
s3, inputting the training sample set into the target Gabor kernel convolutional neural network, extracting depth features and classifying, wherein the depth features comprise textures and edges.
2. The image processing Gabor kernel convolutional neural network construction method of claim 1, wherein S1 comprises:
s11, constructing an improved Gabor filter based on a Gaussian envelope function and a sinusoidal grating function of the center convex and the center concave;
s12, randomly generating an initial population containing individuals with set scale, wherein the base factor of each individual in the initial population is 4 times of the number of the improved Gabor filters, and the genes of each individual represent standard deviation parameters and aspect ratio parameters of Gaussian envelope functions of the improved Gabor filters, and wavelength parameters and phase difference parameters of sinusoidal gratings;
s13, calculating global errors of all individuals in multiple groups by adopting the initial training sample set, taking the global errors as an evaluation function, optimizing standard deviation parameters, length-width ratio parameters and wavelength parameters and phase difference parameters of a sinusoidal grating of the Gaussian envelope function of the improved Gabor filter through propagation, intersection and variation, screening out target individuals with the global errors reaching a preset value, and storing the target individuals to a next generation group, wherein the target individuals comprise individuals with the previous generation loss value reaching the preset value;
s14, repeating the step S13 until all standard deviation parameters and length-width ratio parameters of the Gaussian envelope function, wavelength parameters and phase difference parameters of the sinusoidal grating are optimized or preset reproduction algebra is achieved, returning all target individuals to the improved Gabor filter to obtain the optimized and improved Gabor filter, and taking the optimized and improved Gabor filter as a convolution kernel to form a shallow layer of the Gabor kernel convolution neural network.
3. The method for constructing an image processing Gabor kernel convolutional neural network according to claim 2, wherein in S13, the obtaining of the global error comprises: the two norms of the difference between all sample predictions and standard values obtained by Gabor kernel convolutional neural network.
4. The image processing Gabor kernel convolutional neural network construction method of claim 2, wherein in S13, the evaluation function comprises the following expression:
wherein y is m ' is the predicted value of the mth sample through Gabor CNN, y m Is the standard value of the mth sample, and n represents the number of samples.
5. The image processing Gabor kernel convolutional neural network construction method of claim 2, wherein the target individual is screened by the following expression:
wherein M is d ＝{(σ 1 ,γ 1 ,λ 1 ,ψ 1 ),(σ 2 ,γ 2 ,λ 2 ,ψ 2 ),...,(σ m ,γ m ,λ m ,ψ m )} d M=1, 2,..n represents the population set in the d-th iteration.
6. The image processing Gabor kernel convolutional neural network construction method of claim 1, wherein S2 comprises:
s21, taking the optimized and improved Gabor filter as a convolution kernel to form a shallow layer of a Gabor kernel convolution neural network;
s22, training Gabor kernel convolutional neural network layers except the shallow layer of the Gabor kernel convolutional neural network through a back propagation algorithm to obtain a target Gabor kernel convolutional neural network.
7. The method for constructing an image processing Gabor core convolutional neural network according to claim 6, wherein in S22, the training of Gabor core convolutional neural network layers other than the shallow layer of the Gabor core convolutional neural network by a back propagation algorithm further comprises: and updating parameter weight kernels except standard deviation parameters and length-width ratio parameters of the Gaussian envelope function, wavelength parameters and phase difference parameters of the sinusoidal grating by adopting the back propagation algorithm, and updating weights by a bp algorithm.
8. The image processing Gabor kernel convolutional neural network construction method of claim 7, wherein the update weights are performed as follows:
wherein DeltaW is l Representation ofGradient of l-layer weight matrix, delta B l Represents the gradient of the bias vector of the layer, eta is the learning rate, W l For the l-layer weight matrix, B l Is the layer bias vector.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310270483.6A CN116562358B (en) | 2023-03-16 | 2023-03-16 | Construction method of image processing Gabor kernel convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310270483.6A CN116562358B (en) | 2023-03-16 | 2023-03-16 | Construction method of image processing Gabor kernel convolutional neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116562358A true CN116562358A (en) | 2023-08-08 |
| CN116562358B CN116562358B (en) | 2024-01-09 |
Family
ID=87486857
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310270483.6A Active CN116562358B (en) | 2023-03-16 | 2023-03-16 | Construction method of image processing Gabor kernel convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116562358B (en) |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101286195A (en) * | 2008-06-03 | 2008-10-15 | 西南交通大学 | High precision palm print recognition method based on two-dimensional quadrature Log-Gabor filtering |
| CN101308542A (en) * | 2008-06-26 | 2008-11-19 | 西南交通大学 | Highly precise safe palm recognition method encrypted based on phase characteristic of Log-Gabor mixed filtering |
| CN103198299A (en) * | 2013-03-27 | 2013-07-10 | 西安电子科技大学 | Face recognition method based on combination of multi-direction dimensions and Gabor phase projection characteristics |
| CN103544705A (en) * | 2013-10-25 | 2014-01-29 | 华南理工大学 | Image quality testing method based on deep convolutional neural network |
| CN108171318A (en) * | 2017-11-30 | 2018-06-15 | 河南大学 | One kind is based on the convolutional neural networks integrated approach of simulated annealing-Gaussian function |
| CN109272500A (en) * | 2018-09-27 | 2019-01-25 | 西安电子科技大学 | Fabric classification method based on adaptive convolutional neural networks |
| US20200134821A1 (en) * | 2018-10-26 | 2020-04-30 | National Cheng Kung University | Convolutional neural network and associated method for identifying basal cell carcinoma |
| CN111401156A (en) * | 2020-03-02 | 2020-07-10 | 东南大学 | Image identification method based on Gabor convolution neural network |
| CN112036541A (en) * | 2020-10-16 | 2020-12-04 | 西安工程大学 | Fabric defect detection method based on genetic algorithm optimization neural network |
| CN112215034A (en) * | 2019-07-10 | 2021-01-12 | 重庆邮电大学 | Occlusion face recognition method based on robust representation and classification of matrix decomposition and Gabor features |
| CN114005046A (en) * | 2021-11-04 | 2022-02-01 | 长安大学 | Remote sensing scene classification method based on Gabor filter and covariance pooling |
| CN115546044A (en) * | 2022-05-26 | 2022-12-30 | 南京邮电大学 | Deep learning medical image denoising method based on Gabor filtering |
-
2023
- 2023-03-16 CN CN202310270483.6A patent/CN116562358B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101286195A (en) * | 2008-06-03 | 2008-10-15 | 西南交通大学 | High precision palm print recognition method based on two-dimensional quadrature Log-Gabor filtering |
| CN101308542A (en) * | 2008-06-26 | 2008-11-19 | 西南交通大学 | Highly precise safe palm recognition method encrypted based on phase characteristic of Log-Gabor mixed filtering |
| CN103198299A (en) * | 2013-03-27 | 2013-07-10 | 西安电子科技大学 | Face recognition method based on combination of multi-direction dimensions and Gabor phase projection characteristics |
| CN103544705A (en) * | 2013-10-25 | 2014-01-29 | 华南理工大学 | Image quality testing method based on deep convolutional neural network |
| CN108171318A (en) * | 2017-11-30 | 2018-06-15 | 河南大学 | One kind is based on the convolutional neural networks integrated approach of simulated annealing-Gaussian function |
| CN109272500A (en) * | 2018-09-27 | 2019-01-25 | 西安电子科技大学 | Fabric classification method based on adaptive convolutional neural networks |
| US20200134821A1 (en) * | 2018-10-26 | 2020-04-30 | National Cheng Kung University | Convolutional neural network and associated method for identifying basal cell carcinoma |
| CN112215034A (en) * | 2019-07-10 | 2021-01-12 | 重庆邮电大学 | Occlusion face recognition method based on robust representation and classification of matrix decomposition and Gabor features |
| CN111401156A (en) * | 2020-03-02 | 2020-07-10 | 东南大学 | Image identification method based on Gabor convolution neural network |
| CN112036541A (en) * | 2020-10-16 | 2020-12-04 | 西安工程大学 | Fabric defect detection method based on genetic algorithm optimization neural network |
| CN114005046A (en) * | 2021-11-04 | 2022-02-01 | 长安大学 | Remote sensing scene classification method based on Gabor filter and covariance pooling |
| CN115546044A (en) * | 2022-05-26 | 2022-12-30 | 南京邮电大学 | Deep learning medical image denoising method based on Gabor filtering |
Non-Patent Citations (3)
| Title |
|---|
| LI XIAO-DONG 等: "Selection algorithm of Gabor Kernel for Face Recognition", 《PROCEEDINGS OF THE 31ST CHINESE CONTROL CONFERENCE》, pages 3839 - 3843 * |
| 孟凡杰 等: "基于时频图像纹理特征的雷达辐射源信号识别", 《弹箭与制导学报》, vol. 37, no. 3, pages 152 - 156 * |
| 梁华刚 等: "自适应Gabor卷积核编码网络的表情识别方法", 《计算机工程与应用》, vol. 56, no. 10, pages 149 - 156 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116562358B (en) | 2024-01-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Kumar et al. | Image based tomato leaf disease detection | |
| Alaloul et al. | Data processing using artificial neural networks | |
| CN104751842B (en) | The optimization method and system of deep neural network | |
| CN110188358A (en) | The training method and device of Natural Language Processing Models | |
| CN105488528B (en) | Neural network image classification method based on improving expert inquiry method | |
| US5245696A (en) | Evolution and learning in neural networks: the number and distribution of learning trials affect the rate of evolution | |
| JPH07502357A (en) | Fast converging projective neural network | |
| CN106709482A (en) | Method for identifying genetic relationship of figures based on self-encoder | |
| SG182933A1 (en) | A data structure and a method for using the data structure | |
| CN110674774A (en) | Improved deep learning facial expression recognition method and system | |
| CN110298434A (en) | A kind of integrated deepness belief network based on fuzzy division and FUZZY WEIGHTED | |
| CN111382840B (en) | HTM design method based on cyclic learning unit and oriented to natural language processing | |
| Katz et al. | [29] Artificial neural networks | |
| CN110188621A (en) | A kind of three-dimensional face expression recognition methods based on SSF-IL-CNN | |
| Zheng et al. | Learning associative memories by error backpropagation | |
| CN116562358B (en) | Construction method of image processing Gabor kernel convolutional neural network | |
| Skorpil et al. | Back-propagation and k-means algorithms comparison | |
| CN109034279A (en) | Handwriting model training method, hand-written character recognizing method, device, equipment and medium | |
| CN117034060A (en) | AE-RCNN-based flood classification intelligent forecasting method | |
| CN113628615B (en) | Voice recognition method and device, electronic equipment and storage medium | |
| Lu et al. | Dynamic evolution analysis of desertification images based on BP neural network | |
| Zhao et al. | Efficient learning of NN-MLP based on individual evolutionary algorithm | |
| CN113344202A (en) | Novel deep multi-core learning network model training method, system and medium | |
| Hung et al. | Foundation of Deep Machine Learning in Neural Networks | |
| Deng et al. | Deep self-organizing reservoir computing model for visual object recognition |
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 |