CN112215272A - Bezier curve-based image classification neural network attack method - Google Patents
Bezier curve-based image classification neural network attack method Download PDFInfo
- Publication number
- CN112215272A CN112215272A CN202011050109.8A CN202011050109A CN112215272A CN 112215272 A CN112215272 A CN 112215272A CN 202011050109 A CN202011050109 A CN 202011050109A CN 112215272 A CN112215272 A CN 112215272A
- Authority
- CN
- China
- Prior art keywords
- bezier curve
- image
- neural network
- bezier
- image classification
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 32
- 238000000034 method Methods 0.000 title claims abstract description 22
- 238000012549 training Methods 0.000 claims abstract description 8
- 238000005457 optimization Methods 0.000 claims description 6
- 238000013135 deep learning Methods 0.000 abstract description 4
- 238000013473 artificial intelligence Methods 0.000 abstract description 3
- 241000282414 Homo sapiens Species 0.000 description 11
- 238000013145 classification model Methods 0.000 description 7
- 238000010586 diagram Methods 0.000 description 5
- 230000000694 effects Effects 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 241000239290 Araneae Species 0.000 description 1
- 241000282472 Canis lupus familiaris Species 0.000 description 1
- 241000254173 Coleoptera Species 0.000 description 1
- 241000237858 Gastropoda Species 0.000 description 1
- 241000282412 Homo Species 0.000 description 1
- 240000008042 Zea mays Species 0.000 description 1
- 235000005824 Zea mays ssp. parviglumis Nutrition 0.000 description 1
- 235000002017 Zea mays subsp mays Nutrition 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 235000005822 corn Nutrition 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 210000003608 fece Anatomy 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06N3/12—Computing arrangements based on biological models using genetic models
- G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physiology (AREA)
- Genetics & Genomics (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to the field of artificial intelligence and deep learning, and discloses an image classification neural network attack method based on a Bezier curve. Firstly, building a plurality of image classification neural networks and loading pre-training parameters; then, selecting one thousand image samples meeting the same distribution from the classical ImageNet data set, and carrying out parameter fine adjustment on the constructed neural network by using the image samples to improve the accuracy; randomly generating a Bezier curve on the basis of an original image; then, searching an optimal curve in a large number of randomly generated Bezier curves by using a differential evolution algorithm; and finally, obtaining a confrontation image sample containing the optimal Bezier curve, and misleading the image classification neural network to carry out misclassification on the confrontation image sample.
Description
Technical Field
The invention belongs to the field of artificial intelligence and deep learning, and particularly relates to an image classification neural network attack method based on a Bezier curve.
Background
As is known, the performance of deep learning based on a convolutional neural network in the field of artificial intelligence exceeds that of the traditional machine learning method, the deep learning plays a leading role in the fields of image recognition, natural language processing, voice processing and the like, and particularly in the field of image classification, the performance of the most advanced model exceeds that of human beings.
However, szegdy et al found that the model became fragile when anti-turbulence was added to the image. At the same time, these perturbations are hardly distinguishable to humans. In the following years, a plurality of algorithms generated for the confrontation sample show that the confrontation image can obtain extremely high accuracy when attacking the normal model.
At present, most of attack algorithms aiming at deep learning models take an attack classification model under the condition of minimum disturbance as a target and limit l2Or l∞To ensure a range of variation for the perturbation addition. To achieve that the addition of the disturbance is not perceived by a human or does not affect the human's judgment of the image. However, few algorithms consider whether this perturbation-limiting approach is reasonable enough or effective enough.
In the Alitian pool game, by l∞To limit the criteria, a gradient-based algorithm is used to demonstrate the attack on the classification model. The maximum change value of each pixel point is 32 pixels. When a targeted classified attack is performed on images of snails, dung beetles and spiders, these images seem to become corn, electronic scales and dogs in human view.
Under the limitation of the 32-pixel change value, the original image may become recognized by human beings incorrectly after the image is modified, let alone by the classification model.
The main reason for this is that the magnitude of the modified pixel value does not represent the change of the modified image in human view, and the number of the disturbance pixel points has a larger influence on the human identification image than the value of the disturbance pixel points.
In conclusion, the method for the image classification neural network attack, which is effective and has small influence on human recognition, is found to have important research significance.
Disclosure of Invention
In order to overcome the defects of the method, the invention provides an image classification neural network attack method based on a Bezier curve, which can solve the problem that the modified image has a large influence on the human recognition effect and can effectively attack the image classification neural network. The specific method comprises the following steps:
step 1: building a plurality of image classification neural networks and loading pre-training parameters;
step 2: selecting one thousand image samples meeting the same distribution from the ImageNet data set, and carrying out parameter fine adjustment on the constructed neural network by using the image samples;
and step 3: randomly generating a Bezier curve on the basis of an original image;
and 4, step 4: searching an optimal curve in a large number of randomly generated Bezier curves by using a differential evolution algorithm;
and 5: adding the found optimal Bezier curve to the image sample, and misleading the image classification neural network to carry out misclassification on the image sample;
step 6: repeating the steps 3 to 5, and adding the optimal Bezier curve to all the images;
in the step 1, 3 classical image classification neural networks are respectively built: VGG16, ResNet50, densnet 201, and load the respective pre-training parameters.
In step 2, a thousand images satisfying the same distribution are selected from the ImageNet data set, so as to better verify the effectiveness of the proposed method. And the samples are utilized to carry out parameter fine adjustment on the neural network, so that the accurate identification rate of each neural network to the samples is ensured to be 100%.
In the step 3, pixel point values of the randomly generated bezier curve are all the same randomized initial values, and the position and the direction of the curve are also randomly generated. The theoretical basis of the bezier curve formula used is the bernstein polynomial:
wherein, betaiIs the Bernstein coefficient, bi，n(t) is:
wherein, i is 0, 1, …, n,
while Bernstein seriesWhen the number is a fixed sequence of points lying on a two-dimensional plane (e.g. P)0，P1，P2) The bernstein polynomial becomes the bezier curve equation. By defining n ═ 2, the quadratic bezier curve equation can be obtained:
B(t)＝(1-t)2P0+2t(1-t)P1+t2P2,t∈[0，1] (3)
a bezier curve can generate a complex smooth curve from only a few points. By means of a quadratic Bezier curve formula, a smooth Bezier curve can be drawn by giving three randomly generated coordinate points. And (4) randomly generating three key coordinate point values and RGB values of a large number of Bezier curves, and drawing different Bezier curves on a large number of copied images of the original image for the use of the step 4.
In the step 4, global optimization is performed on a large number of Bezier curves generated in the step 3 by using a differential evolution algorithm, and an optimal Bezier curve is searched. The basic idea of the differential evolution algorithm is as follows: starting from an initial population which is randomly generated, summing the vector difference of any two individuals in the population with the vector of a third individual to generate a new individual, then comparing the new individual with the corresponding individual in the current population, and if the adaptability of the new individual is stronger than that of the existing individual, replacing the old individual with the new individual in the next generation; through continuous evolution, good individuals are kept, bad individuals are eliminated, and the optimal solution is guided to be searched. The differential evolution algorithm does not require gradient information of the model, and therefore, it can be applied to a wider range of optimization problems than the gradient-based method. The differential evolution algorithm has the advantages of simple structure, rapid convergence and strong robustness, and is widely applied to the fields of data mining, pattern recognition, artificial neural networks and the like.
In the step 5, the found optimal Bezier curve is added to the image sample, and the neural network obtained in the step 2 and subjected to parameter fine tuning is used for classifying the image sample, and the classification result is recorded.
Compared with the prior art, the invention has the advantages that:
first, it is proposed not to2Or l∞The distance is used as a standard for measuring the size of the change, so that the situation that the changed image cannot be identified or even is identified wrongly when viewed by human beings is avoided.
Secondly, a differential evolution algorithm is adopted as an optimization strategy, an optimal Bezier curve is heuristically searched, and the attack on the classification model is realized only by changing the pixel value of one curve in the image. Meanwhile, the curve does not influence the recognition of the image by human beings at all.
Finally, the differential evolution algorithm does not need gradient information of the model, so that the method can be applied to a wider optimization problem and has the characteristics of fast convergence, strong robustness and the like.
Drawings
Fig. 1 is a specific flowchart of the image classification neural network attack method based on bezier curves according to the present invention.
Fig. 2 is an image presentation of 9 attack success samples, with the original image label and the label of the post-attack model prediction noted below.
Fig. 3 is an iteration loss graph when the differential evolution algorithm optimizes the bezier curve.
FIG. 4 is a display diagram showing the number of hits for the predicted tag numbers when the ResNet50 is attacked by the present invention.
FIG. 5 is a display diagram showing the number of hits of predicted tag numbers when the VGG16 is attacked by the present invention.
Fig. 6 is a presentation diagram of the number of hits of the predicted tag numbers when using the present invention to attack DenseNet 201.
Detailed Description
To further illustrate the summary of the invention, the computing process, and the effectiveness of the method, reference is made to the following detailed description taken in conjunction with the accompanying drawings and the following examples:
the implementation flow chart is shown in figure 1.
And (3) building three neural network models according to the requirements of the step 1: ResNet50, VGG16 and DenseNet201, corresponding pre-training parameters are downloaded and loaded into the model, and the parameters in the program and the correctness of the model are ensured by testing the pre-training parameters once after the loading is finished. Meanwhile, in order to facilitate the calling of the subsequent steps, the input and output interfaces of the model are set.
And (3) selecting one thousand image samples meeting the same distribution from the ImageNet data set according to the requirements of the step (2), and then respectively carrying out parameter fine adjustment on the three models built in the step (1) by using the image samples, wherein the training times of the whole data set are required to be not more than 10, so that the over-fitting phenomenon is prevented, and the judgment accuracy of the image samples is improved to 100%.
Randomly generating Bezier curves on the basis of the original image according to the requirements of step 3, i.e. by randomly generating P0，P1，P2Three points are used as key points of the bezier curve by using the formula (3) b (t) ═ 1-t)2P0+2t(1-t)P1+t2P2And calculating the specific coordinate value of the Bezier curve, and randomly generating the RGB value of the Bezier curve. Here, it should be noted that: to ensure that the coordinate point position of the bezier curve does not exceed the maximum range of the image, the weight of the bezier curve needs to be set to 2, and the curve width needs to be a random value.
And (4) inputting a large number of Bezier curves generated in the step (3) into a differential algorithm according to the requirements of the step (4) to carry out iterative optimization. The specific settings of the differential evolution algorithm are as follows: population number 200, iteration number 10. The differential evolution algorithm obtains a new population at each iteration, and the new population has smaller confidence of the real label on the three image classification models than the old population, that is, the iterated images are more easily classified into other types by mistake during classification. And finally obtaining the optimal Bezier curve which makes the model classification wrong through iterative computation of a plurality of differential evolution algorithms.
And (5) adding the Bezier curve found in the step (4) to the image sample according to the requirement of the step (5), so that the three image classification neural networks can be wrongly classified.
And (5) repeating the steps 3-5 to add Bezier curves to all the images. After the Bezier curve of one thousand images is added, the results of the images on three image classification models are recorded as follows:
the error classification rate of the VGG16 model is 97.7%, the average iteration number of the differential evolution algorithm is 1.516, and the loss value is 0.0582.
The error classification rate of the ResNet50 model is 78%, the average iteration number of the differential evolution algorithm is 3.997, and the loss value is 0.2612.
The error classification rate of the DenseNet201 model is 59.3%, the average iteration number of the differential evolution algorithm is 5.692, and the loss value is 0.2859.
The method has the advantages that the attack effect on the classical image classification model is good, and the robustness is good because model gradient information is not needed. As can be seen from the iterative loss diagram of fig. 3, the differential evolution algorithm has a faster convergence rate. Fig. 4-6 are diagrams showing the number of hits of the predicted tag numbers when the present invention attacks ResNet50, and it can be seen that a large number of class tags with prediction errors are concentrated on a few numbers. This illustrates some of the shortcomings of the image classification neural network model, which is more prone to classification errors on these few digital labels.
Claims (6)
1. An image classification neural network attack method based on Bezier curves is characterized by comprising the following steps:
step 1: building a plurality of image classification neural networks and loading pre-training parameters;
step 2: selecting one thousand image samples meeting the same distribution from the ImageNet data set, and carrying out parameter fine adjustment on the constructed neural network by using the image samples;
and step 3: randomly generating a Bezier curve on the basis of an original image;
and 4, step 4: searching an optimal curve in a large number of randomly generated Bezier curves by using a differential evolution algorithm;
and 5: adding the found optimal Bezier curve to the image sample, and misleading the image classification neural network to carry out misclassification on the image sample;
step 6: and (5) repeating the steps 3 to 5, and adding the optimal Bezier curve to all the images.
2. The method for image classification neural network attack based on Bezier curve according to claim 1, characterized in that in step 1, 3 classical image classification neural networks are respectively built: VGG16, ResNet50, densnet 201, and load the respective pre-training parameters.
3. The Bezier curve-based image classification neural network attack method according to claim 1, wherein in the step 2, one thousand images satisfying the same distribution are selected from ImageNet data set so as to better verify the effectiveness of the proposed method; and the samples are utilized to carry out parameter fine adjustment on the neural network, so that the accurate identification rate of each neural network to the samples is ensured to be 100%.
4. The method for image classification neural network attack based on Bezier curve as claimed in claim 1, wherein in step 3, the randomly generated Bezier curve has the same randomized initial value of pixel value and the random generation of position and direction; the theoretical basis of the bezier curve formula used is the bernstein polynomial:
wherein, betaiIs the Bernstein coefficient, bi，n(t) is:
wherein, i is 0, 1, …, n,
when the Bernstein coefficient is a fixed sequence of points lying on a two-dimensional plane (e.g., P)0，P1，P2) The bernstein polynomial becomes the bezier curve formula; by limiting n to 2, can obtainTo the quadratic bezier curve equation:
B(t)＝(1-t)2P0+2t(1-t)P1+t2P2，t∈[0，1] (3)
the Bezier curve can generate a complex smooth curve according to few points; by means of a quadratic Bezier curve formula, a smooth Bezier curve can be drawn by giving three randomly generated coordinate points.
5. The method for image classification neural network attack based on Bezier curves as claimed in claim 1, wherein in the step 4, global optimization is performed on a large number of Bezier curves generated in the step 3 by using a differential evolution algorithm to find the optimal Bezier curve.
6. The method according to claim 1, wherein in step 5, the found optimal bezier curve is added to the image sample, and the neural network obtained in step 2 and subjected to parameter fine tuning is used to classify the image sample, and the classification result is recorded.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011050109.8A CN112215272A (en) | 2020-09-29 | 2020-09-29 | Bezier curve-based image classification neural network attack method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011050109.8A CN112215272A (en) | 2020-09-29 | 2020-09-29 | Bezier curve-based image classification neural network attack method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112215272A true CN112215272A (en) | 2021-01-12 |
Family
ID=74052381
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011050109.8A Pending CN112215272A (en) | 2020-09-29 | 2020-09-29 | Bezier curve-based image classification neural network attack method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112215272A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113011386A (en) * | 2021-04-13 | 2021-06-22 | 重庆大学 | Expression recognition method and system based on equally divided characteristic graphs |
Citations (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2005310070A (en) * | 2004-04-26 | 2005-11-04 | Canon Inc | Device and method of information processing |
| CN101859444A (en) * | 2009-04-08 | 2010-10-13 | 佳能株式会社 | Image processing apparatus and image processing method |
| CN103049907A (en) * | 2012-12-11 | 2013-04-17 | 深圳市旭东数字医学影像技术有限公司 | Interactive image segmentation method |
| CN103942552A (en) * | 2014-02-11 | 2014-07-23 | 北京大学 | Character image vectorization method and system based on framework instruction |
| CN104714743A (en) * | 2013-12-12 | 2015-06-17 | 鸿合科技有限公司 | Soft pen realization method and device based on touch screen |
| CN105893984A (en) * | 2016-04-29 | 2016-08-24 | 北京工业大学 | Face projection method for facial makeup based on face features |
| CN106506899A (en) * | 2015-09-08 | 2017-03-15 | 东莞市智觉电子科技有限公司 | A kind of image acquisition based on machine vision and method for reconstructing and device |
| CN108038491A (en) * | 2017-11-16 | 2018-05-15 | 深圳市华尊科技股份有限公司 | A kind of image classification method and device |
| CN108090914A (en) * | 2017-12-18 | 2018-05-29 | 辽宁师范大学 | Color image segmentation method based on statistical modeling and pixel classifications |
| US20180211102A1 (en) * | 2017-01-25 | 2018-07-26 | Imam Abdulrahman Bin Faisal University | Facial expression recognition |
| CN108665420A (en) * | 2017-03-31 | 2018-10-16 | 西安电子科技大学 | Infrared DIM-small Target Image background suppression method based on variation Bayesian model |
| CN108805957A (en) * | 2018-06-07 | 2018-11-13 | 青岛九维华盾科技研究院有限公司 | A kind of vector drawing generating method and system based on bitmap images adaptivenon-uniform sampling |
| CA3065891A1 (en) * | 2017-07-19 | 2019-01-24 | Charite Universitatsmedizin Berlin | Method for estimating shape parameters of the fovea by optical coherence tomography |
| CN110008653A (en) * | 2019-02-26 | 2019-07-12 | 西北工业大学 | A kind of aviation centrifugal pump blade profile optimum design method |
| CN110362089A (en) * | 2019-08-02 | 2019-10-22 | 大连海事大学 | A method of the unmanned boat independent navigation based on deeply study and genetic algorithm |
| AU2019257544A1 (en) * | 2015-05-05 | 2019-11-28 | Kyndi, Inc. | Quanton representation for emulating quantum-like computation on classical processors |
| CA3112697A1 (en) * | 2018-09-14 | 2020-03-19 | Squarehead Technology As | Microphone arrays |
-
2020
- 2020-09-29 CN CN202011050109.8A patent/CN112215272A/en active Pending
Patent Citations (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2005310070A (en) * | 2004-04-26 | 2005-11-04 | Canon Inc | Device and method of information processing |
| CN101859444A (en) * | 2009-04-08 | 2010-10-13 | 佳能株式会社 | Image processing apparatus and image processing method |
| CN103049907A (en) * | 2012-12-11 | 2013-04-17 | 深圳市旭东数字医学影像技术有限公司 | Interactive image segmentation method |
| CN104714743A (en) * | 2013-12-12 | 2015-06-17 | 鸿合科技有限公司 | Soft pen realization method and device based on touch screen |
| CN103942552A (en) * | 2014-02-11 | 2014-07-23 | 北京大学 | Character image vectorization method and system based on framework instruction |
| AU2019257544A1 (en) * | 2015-05-05 | 2019-11-28 | Kyndi, Inc. | Quanton representation for emulating quantum-like computation on classical processors |
| CN106506899A (en) * | 2015-09-08 | 2017-03-15 | 东莞市智觉电子科技有限公司 | A kind of image acquisition based on machine vision and method for reconstructing and device |
| CN105893984A (en) * | 2016-04-29 | 2016-08-24 | 北京工业大学 | Face projection method for facial makeup based on face features |
| US20180211102A1 (en) * | 2017-01-25 | 2018-07-26 | Imam Abdulrahman Bin Faisal University | Facial expression recognition |
| CN108665420A (en) * | 2017-03-31 | 2018-10-16 | 西安电子科技大学 | Infrared DIM-small Target Image background suppression method based on variation Bayesian model |
| CA3065891A1 (en) * | 2017-07-19 | 2019-01-24 | Charite Universitatsmedizin Berlin | Method for estimating shape parameters of the fovea by optical coherence tomography |
| CN108038491A (en) * | 2017-11-16 | 2018-05-15 | 深圳市华尊科技股份有限公司 | A kind of image classification method and device |
| CN108090914A (en) * | 2017-12-18 | 2018-05-29 | 辽宁师范大学 | Color image segmentation method based on statistical modeling and pixel classifications |
| CN108805957A (en) * | 2018-06-07 | 2018-11-13 | 青岛九维华盾科技研究院有限公司 | A kind of vector drawing generating method and system based on bitmap images adaptivenon-uniform sampling |
| CA3112697A1 (en) * | 2018-09-14 | 2020-03-19 | Squarehead Technology As | Microphone arrays |
| CN110008653A (en) * | 2019-02-26 | 2019-07-12 | 西北工业大学 | A kind of aviation centrifugal pump blade profile optimum design method |
| CN110362089A (en) * | 2019-08-02 | 2019-10-22 | 大连海事大学 | A method of the unmanned boat independent navigation based on deeply study and genetic algorithm |
Non-Patent Citations (2)
| Title |
|---|
| "增量极坐标编码的贝赛尔曲线智能优化算法", 《智能系统学报》, vol. 12, no. 6, pages 841 - 847 * |
| ALAA THARWAT: "Intelligent Bézier curve-based path planning model using Chaotic Particle Swarm Optimization algorithm", 《CLUSTER COMPUTING》, vol. 22, pages 4745, XP036918957, DOI: 10.1007/s10586-018-2360-3 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113011386A (en) * | 2021-04-13 | 2021-06-22 | 重庆大学 | Expression recognition method and system based on equally divided characteristic graphs |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wu et al. | Object detection based on RGC mask R‐CNN | |
| CN113378632B (en) | Pseudo-label optimization-based unsupervised domain adaptive pedestrian re-identification method | |
| CN109977918B (en) | Target detection positioning optimization method based on unsupervised domain adaptation | |
| CN108564129B (en) | Trajectory data classification method based on generation countermeasure network | |
| CN109063719B (en) | Image classification method combining structure similarity and class information | |
| CN110619059B (en) | Building marking method based on transfer learning | |
| WO2022105108A1 (en) | Network data classification method, apparatus, and device, and readable storage medium | |
| CN111368935B (en) | SAR time-sensitive target sample amplification method based on generation countermeasure network | |
| CN111931801B (en) | Dynamic route network learning method based on path diversity and consistency | |
| CN115908908B (en) | Remote sensing image aggregation type target recognition method and device based on graph attention network | |
| CN110263855B (en) | Method for classifying images by utilizing common-basis capsule projection | |
| CN114283320B (en) | Branch-free structure target detection method based on full convolution | |
| CN114708479B (en) | Self-adaptive defense method based on graph structure and characteristics | |
| CN115358305A (en) | Incremental learning robustness improving method based on iterative generation of boundary samples | |
| CN115019102B (en) | Construction method and application of countermeasure sample generation model | |
| CN113705724B (en) | Batch learning method of deep neural network based on self-adaptive L-BFGS algorithm | |
| CN107274425A (en) | A kind of color image segmentation method and device based on Pulse Coupled Neural Network | |
| CN112215272A (en) | Bezier curve-based image classification neural network attack method | |
| CN111783688B (en) | Remote sensing image scene classification method based on convolutional neural network | |
| CN117372724A (en) | Progressive image recognition method and system based on sub-mode optimization constraint | |
| Khoo et al. | Evaluation of genetic operators and solution representations for shape recognition by genetic algorithms | |
| Liu et al. | Swarm intelligence for classification of remote sensing data | |
| CN111652246B (en) | Image self-adaptive sparsization representation method and device based on deep learning | |
| CN111797732B (en) | Video motion identification anti-attack method insensitive to sampling | |
| CN108805221A (en) | Handwritten Digit Recognition 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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20210112 |