CN113095446A - Abnormal behavior sample generation method and system - Google Patents
Abnormal behavior sample generation method and system Download PDFInfo
- Publication number
- CN113095446A CN113095446A CN202110640175.9A CN202110640175A CN113095446A CN 113095446 A CN113095446 A CN 113095446A CN 202110640175 A CN202110640175 A CN 202110640175A CN 113095446 A CN113095446 A CN 113095446A
- Authority
- CN
- China
- Prior art keywords
- sample
- class
- center
- samples
- distribution
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/2433—Single-class perspective, e.g. one-against-all classification; Novelty detection; Outlier detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
- Complex Calculations (AREA)
Abstract
The invention discloses a method and a system for generating an abnormal behavior sample, wherein a sample with the minimum sum of distances with the characteristics of other samples, namely the mass center of the base class is searched from the base class with sufficient samples, and the mass center is used as the center of the base class; calculating Euclidean distances between the current new class sample and centers of all base classes, and selecting k base classes closest to the new class sample as nearest neighbor base classes; calculating the average value of the centers of the k base classes to serve as an approximate class center, and calculating the midpoint between the approximate class center and the new class sample to serve as a final new class distribution center; and constructing a sample generator based on Gaussian distribution by using the final new class distribution so as to randomly generate virtual samples. The method not only estimates the distribution center of the new class more accurately, but also effectively solves the problem of poor classification effect on similar classes in the learning of few samples.
Description
Technical Field
The invention relates to the field of deep learning, in particular to a method and a system for generating an abnormal behavior sample.
Background
Campus security problems are more and more emphasized in all social circles at present, and behaviors damaging campus security such as trampling, overlooking, fighting and the like frequently occur. At present, campus safety events in China are in a multiple situation, the difficulty in prevention and control is increased, and campus safety faces unprecedented severe challenges. However, due to the concealment, the burstiness and the frequent nature of the campus security events, the campus security mechanism with high efficiency of campus security guarantee and early prevention is difficult to implement. In recent years, the development of computer vision technology provides possibility for campus abnormal behavior safety monitoring and early warning. The abnormal behavior recognition set is efficient and accurate, potential safety hazards of school parks are restrained in a sprouting state, and early warning and intervention are actively carried out in advance. Therefore, abnormal behavior identification and intelligent early warning based on the computer vision technology have become important research problems in the fields of campus security monitoring, anti-terrorism maintenance, group event early warning and the like. The past abnormal behavior sample generation method is mainly based on deep learning, and the support of a large-scale data set is required.
The great success of deep learning in the field of image processing depends to a large extent on the appearance of large-scale labeled datasets, but deep learning models are prone to overfitting when the number of samples is limited. Therefore, few sample learning is a very promising and challenging computer vision direction. The method simulates the thinking mode of human cognition on new things, namely, the method can accurately identify an unseen object through a few examples, which seems to be an effective method for teaching a machine how to realize the new things like human beings, and further draws the distance between artificial intelligence and human intelligence[1]。
However, it is challenging and practical to learn some knowledge from a limited variety and number of samples and to deduce it into new categories. Most studies on this problem are still on the classification task[2]The existing methods are still unsatisfactory and far from industrial applicationAnd (4) leveling. On one hand, the shortage of the sample inevitably brings about the problem of poor generalization capability of the model; on the other hand, it is extremely difficult to accurately estimate the class distribution using a small number of samples.
The purpose of the few-sample learning is to improve the utilization rate of samples, and how to explore how to utilize a small number of marked samples enables the model to achieve the performance which is comparable to or even better than that of the traditional deep learning model. For the problem of rare labeled samples, there are two ideas, one is to use the labeled part of the sample to label the unlabeled sample with false label when there is a large amount of unlabeled sample; another approach is to generate a large number of virtual samples using labeled samples when the samples are finite. The former problem is not the data size, but the sample labeling problem, if combined with the relevant experts or algorithms, can be effectively alleviated. The latter is a typical problem of small sample size and is relatively tricky. However, if we assume that the class distribution obeys a Gaussian or Gaussian-like distribution[3]Then we can estimate the distribution of the whole class by knowing only the center and the variation range of the class distribution, however, it is extremely difficult to estimate the class distribution from a limited sample, and the key is whether the class center can be accurately estimated.
Performing distribution correction on new class by means of statistical data of base class[4]Is a novel and effective few-sample classification method. However, in the process of performing distribution correction on similar classes, the two corrected distribution centers may be too close to each other, which may cause a phenomenon of poor classification effect on similar classes, and further affect the final classification performance of few samples.
Taking a one-shot as an example, for any shot in each small sample task, we refer to it as a support picture (support image), and assume that its feature vector iss={s 1,s 2,…s mM represents a characteristic dimension, the category of the characteristic dimension is c, and the existing distribution correction method firstly searches and corrects in a plurality of basessNearest k base class centersx i ={x i1 ,x i2 ,…x im }, (i=1, 2, …k) Then, they are compared with the feature vector of the support picturesCarrying out averaging operation, and finally taking the average value as a pairsEstimation of the distribution center of the new class to which it belongs.
The method has remarkable effect, but in some special cases, such as supporting samplessWhen the Deviation from the true distribution is large, the corrected distribution Center may also have a Deviation of different degrees, and we call this phenomenon as Center correction Deviation (Center correction Deviation). Virtual samples generated from deviating distribution centers may lack reliability and if two rectified distribution centers are too close, there may be an overlap phenomenon based on the samples they generate, which we call Sampling Confusion (Sampling fusion). The main reason for the two phenomena is that the estimated deviation of the distribution center of the new class is large.
(1) In summary, the conventional method needs a large amount of labeled data, which not only needs to consume a large amount of manpower and material resources, but also is likely to cause a long tail distribution problem in a data set due to the scarcity of some abnormal behavior data.
(2) The existing method has difficulty in accurately estimating the distribution center of the new class from a limited sample.
The existing method is likely to have the phenomenon of cross or overlap in the distribution estimation of similar abnormal behavior classes, and the similar classes cannot be effectively distinguished.
Disclosure of Invention
The invention aims to solve the technical problem that aiming at the defects of the prior art, the invention provides the abnormal behavior sample generation method and the system thereof, and the accuracy of the new class distribution estimation is improved.
In order to solve the technical problems, the technical scheme adopted by the invention is as follows: an abnormal behavior sample generation method comprises the following steps:
s1, searching a sample with the minimum sum of distances between the sample and the characteristics of other samples from the base class with sufficient samples, namely the centroid of the base class, and taking the centroid as the center of the base class;
s2, calculating Euclidean distances between the current new class sample and centers of all base classes, and selecting k base classes closest to the new class sample as nearest neighbor base classes;
s3, calculating the average value of the centers of the k base classes as an approximate class center, and calculating the midpoint between the approximate class center and the new class sample as a final new class distribution center;
and S4, constructing a sample generator based on Gaussian distribution by using the final new class distribution so as to randomly generate virtual samples.
The great success of deep learning in the field of image processing benefits from the appearance of large-scale data sets, however, when the sample size is limited, the model can be easily over-fitted, which seriously affects the performance of the abnormal behavior classification model. Due to the particularity and scarcity of abnormal behaviors, a large number of samples are difficult to capture, and a method for generating abnormal behavior samples is provided for pain spots. All operations on the samples are on the level of extracted features, and the reliability of generating the samples is improved because most interference information is filtered by the extracted features. Then, for each sample from the new class, the k base class centers closest to it are searched and the average of these base class centers is taken as the approximate class center, since the approximate class is similar to this new class both in features and semantically, and their feature variations and distributions may be very similar. Then, the midpoint between the approximate class center and the new class sample is calculated as an estimate of the new class center, which is to calibrate the distribution of the new class so that the estimated new class center is closer to the real position, thereby constructing a sample generator based on the gaussian distribution to generate a large number of abnormal behavior virtual samples. The abnormal behavior sample generation method is different from the existing data generation method, and the abnormal behavior sample generation method is closer to the real data distribution state theoretically based on data distribution estimation, so that the generated sample has higher reliability. In addition, the method can be established on any common feature extractor and sample generator, does not need any additional parameter, only operates the features of the samples, and is simple in algorithm and easy to implement.
The invention utilizes ResNet50 to extract the characteristics of the sample, different from the common network, ResNet50 introduces jump connection, which improves the information circulation and avoids the gradient disappearance problem caused by the network being too deep.
And inputting the generated virtual sample into a logistic regression classifier for training to obtain a final abnormal behavior classifier.
The logistic regression classifier is selected, and the logistic regression classifier is simple in structure, easy to implement and good in effect, and is more favorable for verifying the reliability of the samples generated by the logistic regression classifier.
In step S1, the base class center calculation process includes: calculating the gradient of the current iteration by using the following formula(ii) a The anchor value for the current iteration is updated using:until finding out the anchor point with the minimum sum of the distances to other sample points, namely obtaining a base center; wherein the content of the first and second substances,x 0is the anchor point after the previous round of iteration,in order to be the anchor point after the update,x ifor the other samples of the same type,nis the number of samples, and α is the learning rate. The process iteratively searches for a sample point with the minimum sum of distances to other sample points, and the whole class distribution is regarded as a homogeneous geometric body, so that the center of the base class found has certain mathematical theoretical support.
In step S4, the final new class distribution center is used as an input of a gaussian distribution function, so as to construct a sample generator based on gaussian distribution, and obtain a randomly generated virtual sample.
The method takes the estimated new class center as the mean value of the Gaussian distribution function and takes the random value between 0.2 and 0.5 as the variance of the function, and a large number of experimental results show that the effect is best when the variance is the random number between 0.2 and 0.5.
After step S4, the method further includes: and training a logistic regression classifier by using the randomly generated virtual samples to obtain an abnormal behavior classifier.
The invention also provides an abnormal behavior sample generation system, which comprises computer equipment; the computer device is configured or programmed for performing the steps of the above-described method.
Compared with the prior art, the invention has the beneficial effects that: the method not only estimates the distribution center of the new class more accurately, but also effectively solves the problem of poor classification effect on similar classes in the learning of few samples. In addition, the class centroid estimation algorithm provided by the invention can more accurately estimate the distribution of the base class by searching the centroid of the base class. Compared with the existing distribution correction method, the reasonability and effectiveness of the method can be proved by an error theory, and the relatively accurate base class distribution estimation is favorable for further improving the reliability of subsequent new class distribution estimation.
Drawings
FIG. 1 is a network framework for identifying abnormal behavior based on a small sample according to an embodiment of the present invention.
Fig. 2 is a schematic visualization diagram of a two-order center estimation algorithm according to an embodiment of the present invention.
Fig. 3(a) and 3(b) are t-SNE visualizations of the true new class distribution and the estimated new class distribution of the present invention, respectively, according to an embodiment of the present invention.
Detailed Description
Embodiments of the invention include two-order center estimation (TCE) and centroid-like estimation (CCE). Firstly, extracting a feature vector for a picture in a data set by taking ResNet50 as a feature extractor; then, inputting the feature vector into a CCE module to estimate the center of the base class, and estimating the center of the base class through enough samples so as to provide support for the estimation of a new class (namely, a class with scarce samples); the output of the CCE block is then used as input to a TCE block to estimate the center of the new class, which is used to solve the problem of calibration bias so as to bring the estimated class center closer to the actual location. Based on the estimated class distribution, a sample generator based on Gaussian distribution is constructed to generate enough virtual samples for training a logistic regression classifier, and finally an abnormal behavior classifier is obtained. The principle of the above method is shown in fig. 1.
Performing distribution correction on new class by means of statistical data of base class[4]Is a novel and effective few-sample classification method. However, in the process of performing distribution correction on similar classes, the two corrected distribution centers may be too close to each other, which may cause a phenomenon of poor classification effect on similar classes, and further affect the final classification performance of few samples.
Taking a one-shot as an example, for any shot in each task with few samples, we refer to it as a support picture (supported image), and assume that its feature vector iss={s 1,s 2,…s mM represents a characteristic dimension, the category of the characteristic dimension is c, and the existing distribution correction method firstly searches k base class centers which are most adjacent to s in a plurality of base classesx i ={x i1 ,x i2 ,…x im }, (i=1, 2, …k) Then, they are compared with the feature vector of the support picturesCarrying out averaging operation, and finally taking the average value as a pairsEstimation of the distribution center of the new class to which it belongs.
The method has remarkable effect, but in some special cases, such as supporting samplessWhen the Deviation from the true distribution is large, the corrected distribution Center may also have a Deviation of different degrees, and we call this phenomenon as Center correction Deviation (Center correction Deviation). Virtual samples generated from deviating distribution centers may lack reliability and if two rectified distribution centers are too close, there may be an overlap phenomenon based on the samples they generate, which we call Sampling Confusion (Sampling fusion). The main reason for the two phenomena is that the estimated deviation of the distribution center of the new class is large. To address this problem, the present invention proposes a two-step centering method to more accurately estimate the distribution center of the new class. Specifically, in the first stage, samples are estimated and supportedsThe center of k nearest neighbor base class distribution centers; in the second stage, the estimation of the first stageCenter and support samplessThe center between, i.e. the center of the center, the principle visualization of this method is shown in fig. 2.
To accurately correct the distribution of a new class with only a few samples, it is necessary to ensure that the distribution estimate for the base class is reliable enough, since estimating the distribution center of the new class requires the use of distribution data of nearest neighbor base classes. However, the number and distribution of base class samples are irregular, which presents a great challenge to estimating their distribution centers. In order to accurately estimate the distribution center of the base class, we propose a class centroid estimation algorithm that can iteratively search out the sample point, i.e., the class centroid, in which the sum of the distances from one class to other sample points is the smallest.
Wherein the content of the first and second substances,x 0for the anchor point (anchor point) after the previous iteration,is a new anchor point,x ifor the other samples of the same type,nfor the number of samples, α is the learning rate, where α = 0.03.
Notably, anchor points (anchor points) for each iteration roundx 0All from the result of the gradient descent of the previous round. In each iteration process, the gradient is calculated according to the formula (1), then the value of the anchor point (anchor point) is updated according to the formula (2), and finally the anchor point with the minimum sum of the distances to other sample points is found out and used as the estimation of the class centroid.
Fig. 3(a) illustrates the true new class distribution, and fig. 3(b) illustrates the new class distribution estimated by the present invention. Obviously, the true new class distribution has the phenomenon of crossing or overlapping the distribution of some classes without any intervention, and the new class distribution estimated by the method of the invention has larger inter-class spacing and intra-class compactness and is easier to classify.
The method comprises the steps of firstly estimating the distribution of a base class by using a class centroid estimation algorithm, then estimating the distribution of a new class by using the distribution data of the base class, and finally generating a large number of virtual samples based on the estimation result to train an abnormal behavior classifier. The specific process is as follows:
the first step is as follows: randomly sampling 5 types of picture samples from a data set to form a less-sample task, and performing feature extraction operation on pictures in the data set by using ResNet50 as a feature extractor to respectively obtain feature vectors of the corresponding pictures; searching a sample with the minimum sum of distances with other sample characteristics from samples with sufficient base classes by using a class center-of-mass estimation algorithm (CCE), namely the center of mass of the base class, and taking the sample as the center estimation of the base class;
the second step is that: calculating the Euclidean distance between each new class image sample in the less-sample task and the center of all base classes obtained in the last step to serve as the distance measurement between the new class image sample and each base class, and selecting k base classes closest to the new class image sample to serve as nearest neighbor base classes;
the third step: with the distribution centers of k base classes in nearest neighbors, we estimate the distribution center of a new class in two stages using a two-stage center estimation algorithm (TCE): in the first stage, the average value of all nearest neighbor base class centers is calculated as the approximate class center of the new class sample. The second stage, calculating the midpoint between the approximate class center and the new class sample as the final new class distribution center estimation;
the fourth step: and constructing a sample generator based on the Gaussian distribution function by using the estimated new class distribution center, randomly generating a large number of virtual samples, and then taking the generated virtual samples as the input of a logistic regression classifier to train an abnormal behavior classifier.
Reference to the literature
[1]Wang Y, Yao Q, Kwok J T, et al. Generalizing from a few examples: A survey on few-shot learning[J]. ACM Computing Surveys (CSUR), 2020, 53(3): 1-34.
[2]Chen W Y, Liu Y C, Kira Z, et al. A Closer Look at Few-shot Classification[C]//International Conference on Learning Representations. 2019.
[3]Hu Y, Gripon V, Pateux S. Leveraging the feature distribution in transfer-based few-shot learning[J]. arXiv preprint arXiv:2006.03806, 2020.
[4]Yang S, Liu L, Xu M. Free Lunch for Few-shot Learning: Distribution Calibration[J]. arXiv preprint arXiv:2101.06395, 2021.
Claims (8)
1. An abnormal behavior sample generation method is characterized by comprising the following steps:
s1, searching a sample with the minimum sum of distances between the sample and the characteristics of other samples from the base class with sufficient samples, namely the centroid of the base class, and taking the centroid as the center of the base class;
s2, calculating Euclidean distances between the current new class sample and centers of all base classes, and selecting k base classes closest to the new class sample as nearest neighbor base classes;
s3, calculating the average value of the centers of the k base classes as an approximate class center, and calculating the midpoint between the approximate class center and the new class sample as a final new class distribution center;
and S4, constructing a sample generator based on Gaussian distribution by using the final new class distribution so as to randomly generate virtual samples.
2. The abnormal behavior pattern generation method according to claim 1, wherein in step S1, the characteristics of the pattern are extracted using ResNet50 as a characteristic extractor.
3. The abnormal behavior sample generation method according to claim 1, further comprising:
and S5, taking the virtual sample as the input of a logistic regression classifier, and training the logistic regression classifier to obtain a final abnormal behavior classifier.
4. The abnormal behavior sample generation method according to claim 1, wherein in step S1, the base class center calculation process includes: calculating the gradient of the current iteration by using the following formula(ii) a The anchor value for the current iteration is updated using:until finding out the anchor point with the minimum sum of the distances to other sample points, namely obtaining a base center; wherein the content of the first and second substances,x 0is the anchor point after the previous round of iteration,in order to be the anchor point after the update,x i for the other samples of the same type,nis the number of samples, and α is the learning rate.
5. The abnormal behavior sample generation method according to claim 1, wherein in step S4, a sample generator based on gaussian distribution is constructed by taking the final new class distribution center as a mean value of the gaussian distribution function and taking a random value as a variance of the gaussian distribution function, so as to obtain the randomly generated virtual sample.
6. The abnormal behavior sample generation method according to claim 5, wherein the random value ranges from 0.2 to 0.5.
7. The abnormal behavior sample generation method according to any one of claims 1 to 6, further comprising, after step S4: and training a logistic regression classifier by using the randomly generated virtual samples to obtain an abnormal behavior classifier.
8. An abnormal behavior sample generation system, comprising a computer device; the computer device is configured or programmed for carrying out the steps of the method according to one of claims 1 to 7.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110640175.9A CN113095446B (en) | 2021-06-09 | 2021-06-09 | Abnormal behavior sample generation method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110640175.9A CN113095446B (en) | 2021-06-09 | 2021-06-09 | Abnormal behavior sample generation method and system |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113095446A true CN113095446A (en) | 2021-07-09 |
CN113095446B CN113095446B (en) | 2021-09-03 |
Family
ID=76664534
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110640175.9A Active CN113095446B (en) | 2021-06-09 | 2021-06-09 | Abnormal behavior sample generation method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113095446B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2023053569A1 (en) * | 2021-09-28 | 2023-04-06 | 株式会社Jvcケンウッド | Machine learning device, machine learning method, and machine learning program |
CN116385807A (en) * | 2023-05-30 | 2023-07-04 | 南京信息工程大学 | Abnormal image sample generation method and device |
Citations (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7372471B1 (en) * | 2004-11-02 | 2008-05-13 | Nvidia Corporation | System and method for single-sample virtual coverage anti-aliasing |
CN101432691A (en) * | 2006-05-02 | 2009-05-13 | 爱特梅尔公司 | Method and system for controlling timing in a processor |
CN102254033A (en) * | 2011-08-05 | 2011-11-23 | 西安电子科技大学 | Entropy weight-based global K-means clustering method |
CN102676657A (en) * | 2012-04-18 | 2012-09-19 | 盛司潼 | Sequencing image recognition system and sequencing image recognition method |
CN103020654A (en) * | 2012-12-12 | 2013-04-03 | 北京航空航天大学 | Synthetic aperture radar (SAR) image bionic recognition method based on sample generation and nuclear local feature fusion |
CN104598880A (en) * | 2015-03-06 | 2015-05-06 | 中山大学 | Behavior identification method based on fuzzy support vector machine |
US20160025615A1 (en) * | 2014-07-22 | 2016-01-28 | SynTouch, LLC | Method and applications for measurement of object tactile properties based on how they likely feel to humans |
CN109993090A (en) * | 2019-03-25 | 2019-07-09 | 浙江大学 | The iris center positioning method of forest and image grayscale feature is returned based on cascade |
US20190272333A1 (en) * | 2018-03-01 | 2019-09-05 | King Fahd University Of Petroleum And Minerals | Heuristic for the data clustering problem |
CN110298450A (en) * | 2019-05-21 | 2019-10-01 | 北京大学第三医院(北京大学第三临床医学院) | A kind of virtual sample generation method based on production confrontation network |
CN112135248A (en) * | 2020-10-09 | 2020-12-25 | 西安建筑科技大学 | WIFI fingerprint positioning method based on K-means optimal estimation |
CN112184734A (en) * | 2020-09-30 | 2021-01-05 | 南京景瑞康分子医药科技有限公司 | Long-time animal posture recognition system based on infrared images and wearable optical fibers |
CN112462940A (en) * | 2020-11-25 | 2021-03-09 | 苏州科技大学 | Intelligent home multi-mode man-machine natural interaction system and method thereof |
-
2021
- 2021-06-09 CN CN202110640175.9A patent/CN113095446B/en active Active
Patent Citations (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7372471B1 (en) * | 2004-11-02 | 2008-05-13 | Nvidia Corporation | System and method for single-sample virtual coverage anti-aliasing |
CN101432691A (en) * | 2006-05-02 | 2009-05-13 | 爱特梅尔公司 | Method and system for controlling timing in a processor |
CN102254033A (en) * | 2011-08-05 | 2011-11-23 | 西安电子科技大学 | Entropy weight-based global K-means clustering method |
CN102676657A (en) * | 2012-04-18 | 2012-09-19 | 盛司潼 | Sequencing image recognition system and sequencing image recognition method |
CN103020654A (en) * | 2012-12-12 | 2013-04-03 | 北京航空航天大学 | Synthetic aperture radar (SAR) image bionic recognition method based on sample generation and nuclear local feature fusion |
US20160025615A1 (en) * | 2014-07-22 | 2016-01-28 | SynTouch, LLC | Method and applications for measurement of object tactile properties based on how they likely feel to humans |
CN104598880A (en) * | 2015-03-06 | 2015-05-06 | 中山大学 | Behavior identification method based on fuzzy support vector machine |
US20190272333A1 (en) * | 2018-03-01 | 2019-09-05 | King Fahd University Of Petroleum And Minerals | Heuristic for the data clustering problem |
CN109993090A (en) * | 2019-03-25 | 2019-07-09 | 浙江大学 | The iris center positioning method of forest and image grayscale feature is returned based on cascade |
CN110298450A (en) * | 2019-05-21 | 2019-10-01 | 北京大学第三医院(北京大学第三临床医学院) | A kind of virtual sample generation method based on production confrontation network |
CN112184734A (en) * | 2020-09-30 | 2021-01-05 | 南京景瑞康分子医药科技有限公司 | Long-time animal posture recognition system based on infrared images and wearable optical fibers |
CN112135248A (en) * | 2020-10-09 | 2020-12-25 | 西安建筑科技大学 | WIFI fingerprint positioning method based on K-means optimal estimation |
CN112462940A (en) * | 2020-11-25 | 2021-03-09 | 苏州科技大学 | Intelligent home multi-mode man-machine natural interaction system and method thereof |
Non-Patent Citations (2)
Title |
---|
DUEN-REN LIU等: "knowledge maps for composite e-services:a mining-based system platform coupling with recommendations", 《SCIENCEDIRECT》 * |
李文璟等: "基于对称KL距离的用户行为时序聚类方法", 《电子与信息学报》 * |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2023053569A1 (en) * | 2021-09-28 | 2023-04-06 | 株式会社Jvcケンウッド | Machine learning device, machine learning method, and machine learning program |
CN116385807A (en) * | 2023-05-30 | 2023-07-04 | 南京信息工程大学 | Abnormal image sample generation method and device |
CN116385807B (en) * | 2023-05-30 | 2023-09-12 | 南京信息工程大学 | Abnormal image sample generation method and device |
Also Published As
Publication number | Publication date |
---|---|
CN113095446B (en) | 2021-09-03 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Zhang et al. | Efficient temporal sequence comparison and classification using gram matrix embeddings on a riemannian manifold | |
Yuan et al. | Robust visual tracking with correlation filters and metric learning | |
Chen et al. | Veram: View-enhanced recurrent attention model for 3d shape classification | |
CN113095446B (en) | Abnormal behavior sample generation method and system | |
CN106355604B (en) | Tracking image target method and system | |
CN109190446A (en) | Pedestrian's recognition methods again based on triple focused lost function | |
Chitta et al. | Training data subset search with ensemble active learning | |
CN111325115A (en) | Countermeasures cross-modal pedestrian re-identification method and system with triple constraint loss | |
CN104992452B (en) | Airbound target automatic tracking method based on thermal imaging video | |
CN110008828B (en) | Pairwise constraint component analysis measurement optimization method based on difference regularization | |
Wang et al. | Exploiting global camera network constraints for unsupervised video person re-identification | |
CN112434599B (en) | Pedestrian re-identification method based on random occlusion recovery of noise channel | |
CN111126820A (en) | Electricity stealing prevention method and system | |
CN110852152A (en) | Deep hash pedestrian re-identification method based on data enhancement | |
CN116361734A (en) | Data-driven blockchain anomaly detection method and terminal | |
CN111860097B (en) | Abnormal behavior detection method based on fuzzy theory | |
Zhao et al. | G3DOA: Generalizable 3D descriptor with overlap attention for point cloud registration | |
Jiang et al. | Robust action recognition based on a hierarchical model | |
Li et al. | Few-shot meta-learning on point cloud for semantic segmentation | |
Nayak et al. | Exploiting spatio-temporal scene structure for wide-area activity analysis in unconstrained environments | |
Xudong et al. | Pedestrian detection and tracking with deep mutual learning | |
CN111353538B (en) | Similar image matching method based on deep learning | |
Li | An improved face detection method based on face recognition application | |
CN111832475A (en) | Face false detection screening method based on semantic features | |
Yang et al. | SESR: Self-Ensembling Sim-to-Real Instance Segmentation for Auto-Store Bin Picking |
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 |