CN114419341B - Convolutional neural network image recognition method based on transfer learning improvement - Google Patents
Convolutional neural network image recognition method based on transfer learning improvement Download PDFInfo
- Publication number
- CN114419341B CN114419341B CN202210067935.6A CN202210067935A CN114419341B CN 114419341 B CN114419341 B CN 114419341B CN 202210067935 A CN202210067935 A CN 202210067935A CN 114419341 B CN114419341 B CN 114419341B
- Authority
- CN
- China
- Prior art keywords
- image
- neural network
- convolutional neural
- image set
- salient region
- 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.)
- Active
Links
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 42
- 238000000034 method Methods 0.000 title claims abstract description 32
- 238000013526 transfer learning Methods 0.000 title claims abstract description 24
- 230000006872 improvement Effects 0.000 title abstract description 5
- 238000012549 training Methods 0.000 claims abstract description 37
- 238000004364 calculation method Methods 0.000 claims abstract description 15
- 238000001514 detection method Methods 0.000 claims abstract description 7
- 238000001228 spectrum Methods 0.000 claims description 20
- 238000013528 artificial neural network Methods 0.000 claims description 19
- 239000013598 vector Substances 0.000 claims description 16
- 238000011176 pooling Methods 0.000 claims description 11
- 230000004913 activation Effects 0.000 claims description 7
- 230000006870 function Effects 0.000 claims description 7
- 239000011159 matrix material Substances 0.000 claims description 6
- 238000010276 construction Methods 0.000 claims description 4
- 238000009499 grossing Methods 0.000 claims description 4
- 230000011218 segmentation Effects 0.000 claims description 4
- 230000003595 spectral effect Effects 0.000 claims description 3
- 230000005012 migration Effects 0.000 abstract description 6
- 238000013508 migration Methods 0.000 abstract description 6
- 230000007812 deficiency Effects 0.000 abstract description 2
- 230000008569 process Effects 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 238000001914 filtration Methods 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 3
- 239000000284 extract Substances 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 210000002569 neuron Anatomy 0.000 description 2
- 238000012546 transfer Methods 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 230000003213 activating effect Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000005315 distribution function Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000006467 substitution reaction Methods 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
- G06N3/04—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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- 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)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a convolutional neural network image recognition method based on transfer learning improvement, which comprises the steps of constructing an original image set, extracting the original image set by using a saliency detection method, and obtaining a salient region image set; constructing a filter set using the salient region image set; constructing a convolutional neural network training model by using a transfer learning method, and training the convolutional neural network; and identifying a significant region image set by using the trained convolutional neural network and the filter set, and obtaining the image depth characteristic. The invention effectively removes the influence of the picture background through the saliency detection, suppresses the information of the interference area, and uses the migration learning method, only the parameter of the original training model is required to be finely tuned, and the invention can be applied to the image recognition of the specific area, thereby not only reducing the data volume required by training, but also eliminating the deficiency of the convolutional neural network in the aspect of parameter calculation, saving the training time and improving the image recognition performance of the algorithm.
Description
Technical Field
The invention relates to the technical field of image recognition, in particular to an improved convolutional neural network image recognition method based on transfer learning.
Background
Deep learning based computer vision techniques are one of the important topics in the field of artificial intelligence that can enable intelligent devices to analyze, process and understand the content of images in order to make the correct decisions at the correct time. The convolutional neural network is one of the most widely applied methods in the deep learning technology, and has strong feature extraction capability and nonlinear data fitting capability.
The existing image recognition method based on deep learning is difficult to reduce the influence of the picture background on target recognition, and brings certain deviation to the accuracy of target recognition; meanwhile, the traditional convolutional neural network involves a large number of learnable parameters, and the parameter calculation is complex, so that the traditional convolutional neural network is extremely time-consuming in calculation.
Disclosure of Invention
The invention provides an improved convolutional neural network image recognition method based on transfer learning, which aims to solve the technical problems that a large number of parameters are required to be set in the traditional image recognition method.
In order to achieve the above object, the technical scheme of the present invention is as follows:
The improved convolutional neural network image recognition method based on transfer learning comprises the following steps:
step 1, constructing an original image set, and extracting the original image set by using a saliency detection method to obtain a salient region image set;
step 2, constructing a filter set by using the salient region image set;
Step 3, constructing a convolutional neural network training model by using a transfer learning method, and training the convolutional neural network;
And 4, identifying a significant region image set by using the trained convolutional neural network and the filter set, and obtaining the image depth characteristic.
Further, the obtaining the salient region image set in the step 1 specifically includes:
step 1.1, changing an original image in an original image set from a space domain to a frequency domain to obtain a frequency spectrum of the original image;
step 1.2, performing smoothing treatment on the spectrum of the original image to obtain a salient region image set S (x), wherein the calculation formula for obtaining the salient region image set S (x) is as follows:
S(x)＝g(x)*F-1[exp(T(f)·+P(f))]2
wherein g (x) is a gaussian filter; t (f) is the spectral residual of the original image; p (f) is the phase spectrum of the original image.
Further, the constructing the filter set in step 2 specifically includes: the significant region image set is segmented and sent into a neural network to generate feature vectors, at least one feature vector is selected randomly to be combined to obtain a filter set J 'of a convolution layer, and the calculation formula of the filter set J' is as follows:
J”＝Jmin|J'-klE|
Wherein E is a salient region image set; kl is the size of the segmented image; j is the image matrix after segmentation; j' is the feature vector.
Further, in the step 3, the construction of the convolutional neural network training model by using the transfer learning method specifically comprises the following steps: and (3) invoking a training model of a source domain of the transfer learning method to initialize the data characteristics of a target domain of the convolutional neural network, and converting the data characteristics of the target domain of the convolutional neural network into the data characteristics of the source domain, thereby establishing a neural network training model.
Further, the image depth feature obtained in step 4 is specifically: the salient region image set is used as input of a convolutional neural network, and sequentially passes through a convolutional layer, a pooling layer, an activation layer and a full connection layer of a filter set to learn to obtain image feature information, and depth features are extracted from the image feature information to obtain image depth features F.
Further, a specific calculation formula for extracting depth features from the image feature information to obtain the image depth feature F is as follows:
F＝C[f1、f2、……fn]T
Where f n is the image feature of the image feature information, T is the transposed symbol, and C is the nonlinear function of the image feature information.
The beneficial effects are that: the invention effectively removes the influence of the picture background through the saliency detection, suppresses the information of the interference area, and uses the migration learning method, only the parameter of the original training model is required to be finely tuned, and the invention can be applied to the image recognition of the specific area, thereby not only reducing the data volume required by training, but also eliminating the deficiency of the convolutional neural network in the aspect of parameter calculation, saving the training time and improving the image recognition performance of the algorithm.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions of the prior art, the drawings that are needed in the embodiments or the description of the prior art will be briefly described below, it will be obvious that the drawings in the following description are some embodiments of the present invention, and that other drawings can be obtained according to these drawings without inventive effort to a person skilled in the art.
FIG. 1 is a flow chart of a convolutional neural network image recognition method based on the improvement of transfer learning;
FIG. 2 is a block diagram of convolutional neural network feature extraction;
FIG. 3 is a graph of a convolutional neural network pre-training model based on transfer learning;
Fig. 4 is a schematic diagram of transfer learning.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments of the present invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The embodiment provides a convolutional neural network image recognition method based on transfer learning improvement, as shown in fig. 1-4, comprising the following steps:
step 1, constructing an original image set, and extracting the original image set by using a saliency detection method to obtain a salient region image set;
step 2, constructing a filter set by using the salient region image set;
Step 3, constructing a convolutional neural network training model by using a transfer learning method, and training the convolutional neural network;
And 4, identifying a significant region image set by using the trained convolutional neural network and the filter set, and obtaining the image depth characteristic.
In a specific embodiment, the obtaining the salient region image set in step1 is specifically:
step 1.1, changing an original image in an original image set from a space domain to a frequency domain to obtain a frequency spectrum of the original image;
step 1.2, performing smoothing treatment on the spectrum of the original image to obtain a salient region image set S (x), wherein the calculation formula for obtaining the salient region image set S (x) is as follows:
S(x)＝g(x)*F-1[exp(T(f)+P(f))]2
wherein g (x) is a gaussian filter; t (f) is the spectral residual of the original image; p (f) is the phase spectrum of the original image.
The significance detection based on the spectrum residual method is to remove a statistical residual part of an image to obtain a significant region of the image, and specifically comprises the following steps:
Firstly, carrying out two-dimensional Fourier transform on an original image to enable the image to be transformed from a space domain to a frequency domain, simultaneously obtaining an amplitude spectrum, and then taking logarithms of the amplitude spectrum to obtain a logarithm spectrum;
A(f)＝α(F(img))
P(f)＝φ(F(img))
L(f)＝log(A(f))
Wherein F is a two-dimensional Fourier transform; alpha and phi are amplitude and phase respectively; a (f) and L (f) are respectively amplitude spectrum and logarithmic amplitude spectrum.
Smoothing the log spectrum by using an average filter;
V(f)＝L(f)*hn(f)
wherein V (f) is a smoothed log spectrum; h n (f) is an average filter;
then, differentiating the log spectrum and the smooth log spectrum to obtain a spectrum residual error T (f);
R(f)＝L(f)-V(f)
Taking T (f) as a real part and P (f) as an imaginary part, and performing two-dimensional Fourier inverse transformation. Squaring the result, and finally obtaining a salient region S (x) through Gaussian blur filtering g (x);
S(x)＝g(x)*F-1[exp(T(f)+P(f))]2
In a specific embodiment, the constructing the filter set in step 2 is specifically: the significant region image set is segmented and sent into a neural network to generate feature vectors, at least one feature vector is selected randomly to be combined to obtain a filter set J 'of a convolution layer, and the calculation formula of the filter set J' is as follows:
J″＝Jmin|J′-klE|
wherein E is a salient region image set; kl is the size of the segmented image; j is the image matrix after segmentation; j' is a feature vector;
Before the convolutional neural network starts to work, initializing network parameters, and laying a foundation for subsequent accurate calculation; inputting a significant region image set into a convolution layer to obtain a sample image set E= { E 1、e2……en };
All images in the sample image set are segmented according to the size of k×l, each segmented data block can be represented as a vector to form a data vector set D, and the corresponding vector matrix is J. I.e., j= [ d 1、d2……dn ];
Inputting the vector matrix after the segmentation processing into a neural network to generate a feature vector set J ' = [ d 1、d2……dkl ], and randomly selecting a filter set J ' =jmin|J ' -klE | of a plurality of feature vector combination convolution layers.
The filter set is a matrix in which each filter has different parameters. Two requirements need to be met in the selection of filters: (1) the neural network should adopt a layer-by-layer incremental method; (2) The depth of the convolution layer filter is multiplied by 2 every time the filter passes through the pooling layer so as to ensure the effective operation of the filter group;
in a specific embodiment, the construction of the convolutional neural network training model by using the transfer learning method in the step 3 is specifically as follows: invoking a training model of a source domain of a transfer learning method to initialize data features of a target domain of the convolutional neural network, and converting the data features of the target domain of the convolutional neural network into the data features of the source domain, thereby establishing a neural network training model;
The basic idea of transfer learning is to transfer knowledge in existing models and data to a target to be learned using the correlation between the learning target and the existing knowledge. Therefore, the transfer learning can simplify model training with a large number of parameters to a certain extent, and improve the efficiency of the model. The migration learning includes a source domain and a target domain:
R(s)＝J”{u1,G(u1)}
R(t)＝J”{u2,G(u2)}
Wherein R(s) and R (t) are respectively a source domain and a target domain; u 1、u2 is a certain feature space in the source domain and the target domain, respectively; g (u 1)、G(u2) is a marginal probability distribution function corresponding to the feature space;
The neural network model performs pre-training on a source data set, the number of training samples contained in the data set is far greater than that of a target data set, and then model parameters are transferred to the target data set for fine-tuning training; during the training process, the weights of the network-specific layers are shifted in each model, and the model is adjusted by training and learning the raw data.
In a specific embodiment, the obtaining the image depth feature in step 4 is specifically: the salient region image set is used as input of a convolutional neural network, and sequentially passes through a convolutional layer, a pooling layer, an activating layer and a full-connection layer of a filter to learn to obtain image feature information, and depth features are extracted from the image feature information to obtain image depth features F;
The input layer is responsible for loading the image into the neural network; the convolution layer extracts target features through convolution operation, the convolution layer extracts bottom features, and then a higher-level convolution layer extracts more abstract and complex features; the pooling layer samples the image characteristics according to the local correlation principle and retains effective data information; the activation layer mainly uses an activation function to carry out nonlinear transformation on the image characteristics, so that the classification capability of the neural network is enhanced; and the full connection layer inputs the obtained result into a classifier to finish the output of the neural network.
In a specific embodiment, a specific calculation formula for extracting depth features from image feature information to obtain image depth features F is as follows:
F＝C[f1、f2、……fn]T
Where f n is an image feature that is image feature information, T is a transposed symbol, and C is a nonlinear function of the image.
Specifically, as shown in fig. 2, in the neural network, the image is first subjected to a convolution layer in the first set of filters, and is subjected to a filtering convolution process:
namely u= (R (t) -R (S)) · S (x, y) i×Kj
Wherein i= … … M is the number of convolution layers; j= … … N is the number of convolution layer filters; k is a filter; s (x, y) is a preprocessed image set; u is a feature image set;
the output of the convolution layer is used as the input of the pooling layer, the image enters the pooling layer for downsampling, and four pixels output by the convolution layer are synthesized into one pixel by using maximum pooling:
The output of the pooling layer is used as the input of the activation layer, and the sigmoid function is used for nonlinear processing:
s is a sigmoid function, and C is a nonlinear function of the image;
And taking the image output by the activation layer as a training set, performing convolution filtering by using a second group of filters to obtain an exponentially amplified image, and performing nonlinear processing on the activation layer.
The output of the second group of filters is used as a training set to carry out convolution filtering in a convolution layer 3, and finally, an image depth feature of a column vector is generated:
F＝C[f1、f2、……fn]T。
FIG. 3 is a block diagram of convolutional neural network feature extraction, a source domain dataset, i.e., an existing large number of datasets, and a target domain dataset, i.e., a relatively small number of datasets to learn. The training process is divided into forward propagation and error reverse propagation, wherein the forward propagation is that an image set of a source domain is sent into a neural network for training, and a training set is output through a convolution layer and a pooling layer; the error counter propagation starts from the output training set, and after the error of the neuron is obtained through calculation by the pooling layer and the convolution layer, the input weight of the neuron is updated, so that the aim of optimizing the neural network is fulfilled. After the neural network preprocessing, the target domain data set is sent into the preprocessed neural network for forward propagation.
Fig. 4 is a schematic diagram of a transfer learning, classical neural network learning requiring a large number of data sets. In the case of very small training data, the performance of the neural network will be greatly affected. Thus, the selection on image recognition of a small dataset reuses models that have been trained on the relevant data. Migration learning typically deals with the scenario where we transfer knowledge learned from source tasks in the source domain into target tasks in the target domain. The source domain image set is subjected to migration learning in the process of extracting image features by the neural network, and then the target domain image set is subjected to migration learning, so that the target image features are extracted.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and not for limiting the same; 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 or all of the technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit of the invention.
Claims (1)
1. The improved convolutional neural network image recognition method based on transfer learning is characterized by comprising the following steps of:
step 1, constructing an original image set, and extracting the original image set by using a saliency detection method to obtain a salient region image set;
step 2, constructing a filter set by using the salient region image set;
Step 3, constructing a convolutional neural network training model by using a transfer learning method, and training the convolutional neural network;
step 4, identifying a significant region image set by using the trained convolutional neural network and the filter set to obtain image depth characteristics;
The obtaining of the salient region image set in the step 1 specifically includes:
step 1.1, changing an original image in an original image set from a space domain to a frequency domain to obtain a frequency spectrum of the original image;
step 1.2, performing smoothing treatment on the spectrum of the original image to obtain a salient region image set S (x), wherein the calculation formula for obtaining the salient region image set S (x) is as follows:
S(x)＝g(x)*F－1[exp(T(f)+P(f))]2
Wherein g (x) is a gaussian filter; t (f) is the spectral residual of the original image; p (f) is the phase spectrum of the original image;
The construction filter set in the step 2 is specifically: the significant region image set is segmented and sent into a neural network to generate feature vectors, at least one feature vector is selected randomly to be combined to obtain a filter set J 'of a convolution layer, and the calculation formula of the filter set J' is as follows:
J″＝Jmin|J′-klE|
Wherein E is a salient region image set; kl is the size of the segmented image; j is the image matrix after segmentation; j' is a feature vector;
In the step 3, the construction of the convolutional neural network training model by using the transfer learning method specifically comprises the following steps: invoking a training model of a source domain of a transfer learning method to initialize data features of a target domain of the convolutional neural network, and converting the data features of the target domain of the convolutional neural network into the data features of the source domain, thereby establishing a neural network training model;
The step 4 of obtaining the image depth features specifically comprises the following steps: the salient region image set is used as input of a convolutional neural network, and sequentially passes through a convolutional layer, a pooling layer, an activation layer and a full connection layer of a filter set to learn so as to obtain image feature information, and depth features are extracted from the image feature information so as to obtain image depth features F;
the specific calculation formula for extracting depth features from the image feature information to obtain the image depth features F is as follows:
F＝C[f1、f2、……fn]T
Where f n is the image feature of the image feature information, T is the transposed symbol, and C is the nonlinear function of the image feature information.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210067935.6A CN114419341B (en) | 2022-01-20 | 2022-01-20 | Convolutional neural network image recognition method based on transfer learning improvement |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210067935.6A CN114419341B (en) | 2022-01-20 | 2022-01-20 | Convolutional neural network image recognition method based on transfer learning improvement |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114419341A CN114419341A (en) | 2022-04-29 |
| CN114419341B true CN114419341B (en) | 2024-04-26 |
Family
ID=81276193
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210067935.6A Active CN114419341B (en) | 2022-01-20 | 2022-01-20 | Convolutional neural network image recognition method based on transfer learning improvement |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114419341B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117095813A (en) * | 2023-08-23 | 2023-11-21 | 脉得智能科技(无锡)有限公司 | Method, system, equipment and medium for constructing sarcopenia automatic diagnosis model |
Citations (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102509290A (en) * | 2011-10-25 | 2012-06-20 | 西安电子科技大学 | Saliency-based synthetic aperture radar (SAR) image airfield runway edge detection method |
| CN103069420A (en) * | 2010-08-25 | 2013-04-24 | 伊斯曼柯达公司 | Detecting recurring events in consumer image collections |
| CN103327359A (en) * | 2013-06-14 | 2013-09-25 | 中国计量学院 | Video significance region searching method applied to video quality evaluation |
| CN107016409A (en) * | 2017-03-20 | 2017-08-04 | 华中科技大学 | A kind of image classification method and system based on salient region of image |
| CN107908876A (en) * | 2017-11-16 | 2018-04-13 | 宁波工程学院 | Motor vehicle driven by mixed power operating mode Forecasting Methodology based on multiple dimensioned convolutional neural networks |
| CN109614489A (en) * | 2018-12-13 | 2019-04-12 | 大连海事大学 | It is a kind of to report severity recognition methods based on transfer learning and the Bug of feature extraction |
| CN110245292A (en) * | 2019-05-28 | 2019-09-17 | 华东师范大学 | A kind of natural language Relation extraction method based on neural network filtering noise characteristic |
| CN110533580A (en) * | 2019-08-08 | 2019-12-03 | 西安交通大学 | A kind of image Style Transfer method generating neural network based on confrontation |
| CN111309907A (en) * | 2020-02-10 | 2020-06-19 | 大连海事大学 | Real-time Bug assignment method based on deep reinforcement learning |
| CN111897985A (en) * | 2020-06-23 | 2020-11-06 | 西安交通大学 | Image multi-label classification method, system, equipment and readable storage medium |
| CN112183765A (en) * | 2020-10-30 | 2021-01-05 | 浙江大学 | Multi-source multi-modal data preprocessing method and system for shared learning |
| AU2020103613A4 (en) * | 2020-11-23 | 2021-02-04 | Agricultural Information and Rural Economic Research Institute of Sichuan Academy of Agricultural Sciences | Cnn and transfer learning based disease intelligent identification method and system |
| CN112464816A (en) * | 2020-11-27 | 2021-03-09 | 南京特殊教育师范学院 | Local sign language identification method and device based on secondary transfer learning |
| CN113505799A (en) * | 2021-06-30 | 2021-10-15 | 深圳市慧鲤科技有限公司 | Significance detection method and training method, device, equipment and medium of model thereof |
-
2022
- 2022-01-20 CN CN202210067935.6A patent/CN114419341B/en active Active
Patent Citations (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103069420A (en) * | 2010-08-25 | 2013-04-24 | 伊斯曼柯达公司 | Detecting recurring events in consumer image collections |
| CN102509290A (en) * | 2011-10-25 | 2012-06-20 | 西安电子科技大学 | Saliency-based synthetic aperture radar (SAR) image airfield runway edge detection method |
| CN103327359A (en) * | 2013-06-14 | 2013-09-25 | 中国计量学院 | Video significance region searching method applied to video quality evaluation |
| CN107016409A (en) * | 2017-03-20 | 2017-08-04 | 华中科技大学 | A kind of image classification method and system based on salient region of image |
| CN107908876A (en) * | 2017-11-16 | 2018-04-13 | 宁波工程学院 | Motor vehicle driven by mixed power operating mode Forecasting Methodology based on multiple dimensioned convolutional neural networks |
| CN109614489A (en) * | 2018-12-13 | 2019-04-12 | 大连海事大学 | It is a kind of to report severity recognition methods based on transfer learning and the Bug of feature extraction |
| CN110245292A (en) * | 2019-05-28 | 2019-09-17 | 华东师范大学 | A kind of natural language Relation extraction method based on neural network filtering noise characteristic |
| CN110533580A (en) * | 2019-08-08 | 2019-12-03 | 西安交通大学 | A kind of image Style Transfer method generating neural network based on confrontation |
| CN111309907A (en) * | 2020-02-10 | 2020-06-19 | 大连海事大学 | Real-time Bug assignment method based on deep reinforcement learning |
| CN111897985A (en) * | 2020-06-23 | 2020-11-06 | 西安交通大学 | Image multi-label classification method, system, equipment and readable storage medium |
| CN112183765A (en) * | 2020-10-30 | 2021-01-05 | 浙江大学 | Multi-source multi-modal data preprocessing method and system for shared learning |
| AU2020103613A4 (en) * | 2020-11-23 | 2021-02-04 | Agricultural Information and Rural Economic Research Institute of Sichuan Academy of Agricultural Sciences | Cnn and transfer learning based disease intelligent identification method and system |
| CN112464816A (en) * | 2020-11-27 | 2021-03-09 | 南京特殊教育师范学院 | Local sign language identification method and device based on secondary transfer learning |
| CN113505799A (en) * | 2021-06-30 | 2021-10-15 | 深圳市慧鲤科技有限公司 | Significance detection method and training method, device, equipment and medium of model thereof |
Non-Patent Citations (9)
| Title |
|---|
| Improving Graph Neural Network by Filter Preprocessing;Rui Zhang and Xin Li;《CSSE》;20211024;全文 * |
| Local and Global Feature Aggregation-Aware Network for Salient Object Detection;Zikai Da等;《Electronics》;20220112;全文 * |
| Phase-preserving Filtering Method for Interference Phase Diagram Based on Contourlet Transform;Jiayu Zhao等;《IEEE》;20200330;全文 * |
| 基于位置感知的图像显著性算法研究;向丹;《信息科技》;20170115;全文 * |
| 基于卷积神经网络的人脸检测和性别识别研究;汪济民;《信息科技》;20160115;全文 * |
| 基于深度神经网络的显著目标检测算法研究;王亚楠;《信息科技》;20210615;全文 * |
| 基于迁移学习的卷积神经网络玉米病害图像识别;许景辉;邵明烨;王一琛;韩文霆;;农业机械学报;20200225(第02期);全文 * |
| 基于迁移学习的花卉图像分类模型与算法研究;刘晓波;《信息科技》;20210915;全文 * |
| 多视角图像的特征点提取与立体匹配研究;王莎莎;《信息科技》;20190515;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114419341A (en) | 2022-04-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Tian et al. | Deep learning on image denoising: An overview | |
| Wang et al. | Multi-scale dilated convolution of convolutional neural network for image denoising | |
| CN109949255B (en) | Image reconstruction method and device | |
| CN109345508B (en) | Bone age evaluation method based on two-stage neural network | |
| Cai et al. | Blind inpainting using the fully convolutional neural network | |
| CN109035149B (en) | License plate image motion blur removing method based on deep learning | |
| CN109658344A (en) | Image de-noising method, device, equipment and storage medium based on deep learning | |
| CN106548159A (en) | Reticulate pattern facial image recognition method and device based on full convolutional neural networks | |
| CN112070158B (en) | Facial flaw detection method based on convolutional neural network and bilateral filtering | |
| CN115019173A (en) | Garbage identification and classification method based on ResNet50 | |
| Zhuo et al. | Ridnet: Recursive information distillation network for color image denoising | |
| CN114419341B (en) | Convolutional neural network image recognition method based on transfer learning improvement | |
| CN110287759B (en) | Eye fatigue detection method based on simplified input convolutional neural network O-CNN | |
| CN114882278A (en) | Tire pattern classification method and device based on attention mechanism and transfer learning | |
| CN110135435B (en) | Saliency detection method and device based on breadth learning system | |
| CN114581789A (en) | Hyperspectral image classification method and system | |
| CN113205102B (en) | Vehicle mark identification method based on memristor neural network | |
| CN112508966A (en) | Interactive image segmentation method and system | |
| CN116416212B (en) | Training method of road surface damage detection neural network and road surface damage detection neural network | |
| Ulu et al. | MLFAN: Multilevel Feature Attention Network With Texture Prior for Image Denoising | |
| CN112529081B (en) | Real-time semantic segmentation method based on efficient attention calibration | |
| Zhu et al. | MRDA-Net: multiscale residual dense attention network for image denoising | |
| Yapici et al. | A review of image denoising with deep learning | |
| Wu et al. | RDS-denoiser: a detail-preserving convolutional neural network for image denoising | |
| CN113627468B (en) | Non-local neural network image processing method and system based on unsupervised learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |