CN109299707A - A kind of unsupervised pedestrian recognition methods again based on fuzzy depth cluster - Google Patents
A kind of unsupervised pedestrian recognition methods again based on fuzzy depth cluster Download PDFInfo
- Publication number
- CN109299707A CN109299707A CN201811279325.2A CN201811279325A CN109299707A CN 109299707 A CN109299707 A CN 109299707A CN 201811279325 A CN201811279325 A CN 201811279325A CN 109299707 A CN109299707 A CN 109299707A
- Authority
- CN
- China
- Prior art keywords
- fuzzy
- pedestrian image
- label
- pedestrian
- feature
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- 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
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Biophysics (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Probability & Statistics with Applications (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The embodiment of the invention discloses a kind of unsupervised pedestrian recognition methods again based on fuzzy depth cluster, this method comprises: extracting pedestrian image feature using pedestrian image feature extraction network model；It constructs fuzzy depth clustering network and initializes；Learn new feature space and cluster centre using fuzzy depth clustering network, distributes fuzzy label for no label pedestrian image；Pedestrian image feature extraction network model is trained using reliability sample；Alternately training is until reliability sample reaches saturation；Test pedestrian image feature is extracted using the pedestrian image feature extraction network model that training obtains, obtains unsupervised pedestrian recognition result again by calculating characteristic distance.The present invention learns new feature space using fuzzy depth clustering network, be conducive to the cluster of complicated pedestrian image and the distribution of fuzzy label, utilize the reliability sample training feature extraction network with fuzzy label, reduce the risk of overfitting, so that improve unsupervised pedestrian identifies matched accuracy again.
Description
Technical field
The invention belongs to computer vision, pattern-recognition, field of artificial intelligence, and in particular to one kind is based on fuzzy deep
Spend the unsupervised pedestrian recognition methods again of cluster.
Background technique
Pedestrian identifies that the target of (Re-identification) is to judge whether the pedestrian occurred under different cameras belongs to again
In same a group traveling together, the subproblem of image retrieval can be regarded as.With the continuous development of society, more and more cameras are pacified
It is attached to public place, such as: market, cell, campus, airport.By the monitoring to public place pedestrian, collection can be convenient
The crime process of offender solves a case for the police and provides more clues.Video monitoring has carried out proper restraint to pedestrian simultaneously,
The social public security of people is ensured.Since the shooting angle of pedestrian image is different, illumination condition is poor, resolution ratio is low, pedestrian
Posture such as constantly changes at the reasons, so that pedestrian identifies again not only has researching value but also pole as current computer vision field one
Has the heat subject of challenge.
Currently, convolutional neural networks have been widely used for pedestrian identifies field again, and pedestrian can be effectively improved
The performance identified again, but the pedestrian image that the calculating of convolutional neural networks needs largely to have label, and the cost of labeling process is non-
Chang Gao.In order to solve this problem, unsupervised pedestrian identifies again comes into being.Unsupervised pedestrian identifies again does not need pedestrian image
Label and closer to practical application.Based on convolutional neural networks, unsupervised pedestrian identifies again is broadly divided into two major classes: the
One kind learns depth characteristic using unsupervised domain adaptive technique from source domain (having label) and aiming field (no label).Wang et al.
Propose semantic feature and identification feature that a transportable model comes learning objective domain；Zhong et al. considers phase simultaneously
The invariance and domain connectivity of machine, with depth characteristic more broadly in learning objective domain.Second class is by prediction without label pedestrian
The pseudo label of image realizes that unsupervised pedestrian identifies again.Wu et al. proposes a kind of dynamic sampling strategy, gradually estimates pedestrian
The label of image, while optimizing depth network pedestrian is learnt with this and indicate feature；Fan et al. is poly- by alternately updating k-means
Class and CNN model obtain pedestrian image label and learn the feature for having identification.For complicated pedestrian image, original feature
Space has the non-linear and inseparability of height, in addition, it is single to cluster pedestrian image of the imparting without label using k-means
Label be easy to cause the over-fitting of CNN.
Summary of the invention
The invention aims to solve the nonlinearity of original feature space and inseparability and utilize k-means
Cluster obtain single label to unsupervised pedestrian recognition result is affected again the problem of, be based on for this purpose, the present invention provides one kind
The unsupervised pedestrian recognition methods again of fuzzy depth cluster.
In order to realize the purpose, the present invention proposes a kind of unsupervised pedestrian side of identification again based on fuzzy depth cluster
Method the described method comprises the following steps:
Step S1 determines pedestrian image feature extraction network model, and utilizes the pedestrian image feature extraction network mould
Type extracts pedestrian image feature for pedestrian image, wherein the pedestrian image is without label image；
Step S2 constructs fuzzy depth clustering network and initializes to it；
Step S3 learns new feature space and cluster centre using the fuzzy depth clustering network after initialization, and is
The pedestrian image of no label distributes fuzzy label；
Step S4 determines reliability sample from the pedestrian image with fuzzy label, and utilizes the reliability sample
The pedestrian image feature extraction network model is trained；
Step S5, alternately the training fuzzy depth clustering network and pedestrian image feature extraction network model are until selection
Reliability sample reach saturation, obtain training completion pedestrian image feature extraction network model；
Step S6 obtains test pedestrian image, extracts institute using the pedestrian image feature extraction network model that training obtains
The feature for stating test pedestrian image, by calculating in the feature for testing pedestrian image and picture library between the feature of pedestrian image
Distance unsupervised pedestrian recognition result again can be obtained.
Optionally, the step S1 the following steps are included:
Step S11 determines pedestrian image feature extraction network model；
Step S12 obtains incoherent tape label pedestrian image, and utilizes the incoherent tape label pedestrian image pair
It is initialized in the pedestrian image feature extraction network model；
Step S13 is mentioned using the pedestrian image that the pedestrian image feature extraction network model after initialization is not tape label
Feature is taken, fisrt feature is obtainedWherein, ziIndicate the feature of the i-th width not pedestrian image of tape label, N indicate without
The quantity of the pedestrian image of label.
Optionally, the step S2 the following steps are included:
Step S21 constructs fuzzy depth clustering network, wherein the fuzzy depth clustering network can be configured such that comprising one
A depth network and a cluster loss；
Step S22 initializes the fuzzy depth clustering network.
Optionally, the step S22 the following steps are included:
Step S221 initializes the depth network parameter in the fuzzy depth clustering network；
Step S222 initializes the cluster centre of the fuzzy depth clustering network.
Optionally, the step S3 the following steps are included:
Step S31, using the depth network in the fuzzy depth clustering network after initialization by the fisrt featureIt is mapped to a new feature space, study obtains new pedestrian image feature second feature
Step S32 clusters the second feature using Fuzzy c-means Clustering to obtain cluster centre and multiple collection
Group, the objective function by minimizing Fuzzy c-means Clustering is updated the cluster centre, and every width row is calculated
Degree of membership of people's image to each cluster；
Cluster is calculated using the second feature, the cluster centre of update and the degree of membership that is calculated in step S33
Loss；
Step S34 updates the parameter of depth network in the fuzzy depth clustering network using the cluster loss iteration
And cluster centre, obtain the Optimal cluster centers of depth network in the fuzzy depth clustering network；
Step S35 obtains the fuzzy label of no label pedestrian image using the second feature and Optimal cluster centers.
Optionally, the objective function of the Fuzzy c-means Clustering indicates are as follows:
Wherein, N indicates the quantity of no label pedestrian image, and K indicates the number of cluster centre, and m > 0 indicates fuzzy coefficient, cj
Indicate the feature vector of j-th of cluster centre, aijIndicate i-th of pedestrian image to j-th of cluster degree of membership,For bound term, indicate every width pedestrian image to the sum of the degree of membership of all clusters be 1.
Optionally, the step S33 the following steps are included:
Step S331 calculates fuzzy point based on the second feature, the cluster centre of update and the degree of membership that is calculated
With fij；
Step S332, according to the fuzzy allocation fijTarget Assignment is calculated；
Cluster loss is calculated using the KL divergence of the fuzzy allocation and Target Assignment in step S333.
Optionally, in the step S35, the mould that will be calculated according to the second feature and the Optimal cluster centers
Paste distribution is defined as the fuzzy label of the no label pedestrian image.
Optionally, in the step S4, confidence level is selected to be higher than the pedestrian image with fuzzy label of a preset threshold
As the reliability sample.
Optionally, the step S4 the following steps are included:
Step S41 selects confidence level to be higher than the pedestrian image with fuzzy label of a preset threshold as reliability sample
This；
Step S42 utilizes pedestrian image feature extraction network model described in the reliability sample training with fuzzy label.
The invention has the benefit that the present invention learns new feature space, new spy using fuzzy depth clustering network
Sign space is conducive to the cluster of complicated pedestrian image and the distribution of fuzzy label, and is instructed using the reliability sample with fuzzy label
Practice ResNet-50 network and reduce the risk of overfitting so that ResNet-50 network training process specification is smooth, to improve
Unsupervised pedestrian identifies matched accuracy again.
It should be noted that the present invention obtained project of national nature science fund project No.61501327,
No.61711530240, No.61501328, Tianjin Natural Science Fund In The Light key project No.17JCZDJC30600, Tianjin teacher
Model university " young scientific research top-notch personnel incubation program " No.135202RC1703, the open class of pattern-recognition National Key Laboratory
Inscribe fund No.201700001, No.201800002, China national fund for studying abroad No.201708120040,
The subsidy of No.201708120039.
Detailed description of the invention
Fig. 1 is a kind of unsupervised pedestrian recognition methods again based on fuzzy depth cluster according to an embodiment of the invention
Flow chart；
Fig. 2 be according to the present invention an embodiment alternately the fuzzy depth clustering network of training and ResNet-50 network implementations without
The structure flow chart that supervision pedestrian identifies again；
Fig. 3 is the structural block diagram of depth network in fuzzy depth clustering network according to an embodiment of the invention.
Specific embodiment
In order to make the objectives, technical solutions and advantages of the present invention clearer, With reference to embodiment and join
According to attached drawing, the present invention is described in more detail.It should be understood that these descriptions are merely illustrative, and it is not intended to limit this hair
Bright range.In addition, in the following description, descriptions of well-known structures and technologies are omitted, to avoid this is unnecessarily obscured
The concept of invention.
Fig. 1 is a kind of unsupervised pedestrian recognition methods again based on fuzzy depth cluster according to an embodiment of the invention
Flow chart illustrates that some specific implementation flows of the invention are as described in Figure 1 by taking Fig. 1 as an example below, described based on fuzzy depth
Cluster unsupervised pedestrian again recognition methods the following steps are included:
Step S1 determines pedestrian image feature extraction network model, and utilizes the pedestrian image feature extraction network mould
Type extracts pedestrian image feature for pedestrian image, wherein the pedestrian image is without label image, as shown in Figure 2
1. and 2. step；
Further, the step S1 the following steps are included:
Step S11 determines pedestrian image feature extraction network model；
In an embodiment of the present invention, the pedestrian image feature extraction network model is chosen as ResNet-50 network.
Step S12 obtains incoherent tape label pedestrian image, and utilizes the incoherent tape label pedestrian image pair
It is initialized in the pedestrian image feature extraction network model；
Step S13 is mentioned using the pedestrian image that the pedestrian image feature extraction network model after initialization is not tape label
Feature is taken, fisrt feature is obtainedWherein, ziIndicate the feature of the i-th width not pedestrian image of tape label, N indicate without
The quantity of the pedestrian image of label.
In an embodiment of the present invention, using the final full articulamentum of ResNet-50 network as pedestrian image feature zi, dimension
Number is 2048.
Step S2 constructs fuzzy depth clustering network and initializes to it；
Further, the step S2 the following steps are included:
Step S21 constructs fuzzy depth clustering network；
In an embodiment of the present invention, the fuzzy depth clustering network can be configured such that comprising a depth network and one
Cluster loss.Wherein, the depth network can be made of five full articulamentums, and the number of neuron may respectively be 2048,512,
512,1024 and 256, Fig. 3 show the structural frames of depth network in fuzzy depth clustering network according to an embodiment of the invention
Figure, as shown in figure 3, the input of the depth network is fisrt featureOutput is be mapped to new feature space new
Pedestrian's feature, second featureWherein, xiIndicate the new feature of the i-th width not pedestrian image of tape label, dimension is
256。
Step S22 initializes the fuzzy depth clustering network.
Wherein, the step S22 the following steps are included:
Step S221 initializes the depth network parameter in the fuzzy depth clustering network；
In an embodiment of the present invention, using stack autocoder for the depth in the fuzzy depth clustering network
Degree network parameter is initialized.
Step S222 initializes the cluster centre of the fuzzy depth clustering network.
In an embodiment of the present invention, net is clustered for the fuzzy depth using clustering methods such as Fuzzy c-means Clusterings
The cluster centre of network is initialized.
Step S3 learns new feature space and cluster centre using the fuzzy depth clustering network after initialization, and is
The pedestrian image of no label distributes fuzzy label；
Further, the step S3 the following steps are included:
Step S31, using the depth network in the fuzzy depth clustering network after initialization by the fisrt featureIt is mapped to a new feature space, study obtains new pedestrian image feature second featureWherein, xiIt can
To indicate are as follows:
xi=φ (zi)
Wherein, φ indicates the mapping of depth network.
Step S32 clusters the second feature using Fuzzy c-means Clustering to obtain cluster centre and multiple collection
Group, the objective function by minimizing Fuzzy c-means Clustering is updated the cluster centre, and every width row is calculated
Degree of membership of people's image to each cluster, wherein the objective function of the Fuzzy c-means Clustering are as follows:
Wherein, N indicates the quantity of no label pedestrian image, and K indicates the number (quantity of cluster) of cluster centre, the table of m > 0
Show fuzzy coefficient, cjIndicate the feature vector of j-th of cluster centre, aijIndicate i-th of pedestrian image being subordinate to j-th cluster
Degree,For bound term, indicate every width pedestrian image to the sum of the degree of membership of all clusters be 1, aijAnd cjIt can be with
It indicates are as follows:
Cluster is calculated using the second feature, the cluster centre of update and the degree of membership that is calculated in step S33
Loss；
Wherein, the step S33 the following steps are included:
Step S331 utilizes following formula meter based on the second feature, the cluster centre of update and the degree of membership that is calculated
Calculate fuzzy allocation fij:
Wherein, β indicates freedom degree.
In an embodiment of the present invention, β is set as 1.
Step S332, according to the fuzzy allocation fijTarget Assignment is calculated, wherein the Target Assignment tijIt can be with
It indicates are as follows:
Wherein, uj=ΣifijIndicate the collection group frequency of j-th of cluster.
Cluster loss is calculated using the KL divergence of the fuzzy allocation and Target Assignment, wherein described in step S333
Cluster loss indicates are as follows:
Step S34 updates depth network in the fuzzy depth clustering network using the continuous iteration of the cluster loss
Parameter and cluster centre, finally obtain the Optimal cluster centers of depth network in the fuzzy depth clustering network；
In an embodiment of the present invention, obscuring the maximum the number of iterations of depth clustering network may be configured as 30.
Step S35 obtains the fuzzy label of no label pedestrian image using the second feature and Optimal cluster centers.
In an embodiment of the present invention, it will be calculated according to the second feature and the Optimal cluster centers fuzzy
Distribution is defined as the fuzzy label of the no label pedestrian image.
Step S4 determines reliability sample from the pedestrian image with fuzzy label, and utilizes the reliability sample
The pedestrian image feature extraction network model is trained；
Wherein, when determining reliability sample, confidence level is selected to be higher than the pedestrian with fuzzy label of a preset threshold
Image is as the reliability sample.
Further, the step S4 the following steps are included:
Step S41 selects confidence level to be higher than the pedestrian image with fuzzy label of a preset threshold as reliability sample
This；
Wherein, the index of the i-th width pedestrian image may be expressed as:
Selection vector r=[r is introduced herein1,r2,...,rN], if the i-th width pedestrian image is selected as reliability sample,
Then ri=1, otherwise, ri=0.So, the selection criterion of the reliability sample can indicate are as follows:
Wherein, ξ indicates the threshold value of reliability samples selection, is a positive number.
As can be seen from the above formula that when the distance between pedestrian's feature to corresponding cluster centre vector is less than ξ, the pedestrian
Image is reliability sample.In addition, bound term therein can guarantee at least one reliability sample of each cluster.
Step S42 utilizes pedestrian image feature extraction network model described in the reliability sample training with fuzzy label.
Wherein, the loss function of the pedestrian image feature extraction network model can indicate are as follows:
Wherein, qij∈ [0,1] indicates that the i-th width pedestrian image belongs to the prediction probability of j-th of cluster.
Step S5, alternately the training fuzzy depth clustering network and pedestrian image feature extraction network model are until selection
Reliability sample reach saturation, obtain training completion pedestrian image feature extraction network model, as the step in Fig. 2 3. and
Shown in 4.；
Step S6 obtains test pedestrian image, extracts institute using the pedestrian image feature extraction network model that training obtains
The feature for stating test pedestrian image, by calculating in the feature for testing pedestrian image and picture library between the feature of pedestrian image
Distance unsupervised pedestrian recognition result again can be obtained.
In an embodiment of the present invention, the distance is chosen as Euclidean distance, in this embodiment, the test pedestrian figure
Euclidean distance between picture and the feature of pedestrian image in picture library is smaller, just illustrates to test pedestrian image in pedestrian image and picture library
Between similarity it is higher, and then can be obtained by unsupervised pedestrian recognition result again.
Using online disclosed pedestrian, identification database, such as will be in DukeMTMC-reID database as test object again
Training set as incoherent tape label pedestrian image for initializing ResNet-50, by Market-1501 database
Training set as not tape label pedestrian image alternately the fuzzy depth clustering network of training and ResNet-50 network and
It is tested on Market-1501 test set, works as K=750, when ξ=0.85, unsupervised pedestrian identifies that matched accuracy is again
Rank-1=63.4%, mean accuracy mAP=32.4%.It can be seen that the validity of the method for the present invention.
It should be understood that above-mentioned specific embodiment of the invention is used only for exemplary illustration or explains of the invention
Principle, but not to limit the present invention.Therefore, that is done without departing from the spirit and scope of the present invention is any
Modification, equivalent replacement, improvement etc., should all be included in the protection scope of the present invention.In addition, appended claims purport of the present invention
Covering the whole variations fallen into attached claim scope and boundary or this range and the equivalent form on boundary and is repairing
Change example.
Claims (10)
1. a kind of unsupervised pedestrian recognition methods again based on fuzzy depth cluster, which is characterized in that this method includes following step
It is rapid:
Step S1 determines pedestrian image feature extraction network model, and utilizes the pedestrian image feature extraction network model pair
Pedestrian image feature is extracted in pedestrian image, wherein the pedestrian image is without label image；
Step S2 constructs fuzzy depth clustering network and initializes to it；
Step S3 learns new feature space and cluster centre using the fuzzy depth clustering network after initialization, and is no mark
The pedestrian image of label distributes fuzzy label；
Step S4, from fuzzy label pedestrian image in determine reliability sample, and using the reliability sample for
The pedestrian image feature extraction network model is trained；
Step S5, alternately the training fuzzy depth clustering network and pedestrian image feature extraction network model are until what is selected can
Reach saturation by property sample, obtains the pedestrian image feature extraction network model of training completion；
Step S6 obtains test pedestrian image, extracts the survey using the pedestrian image feature extraction network model that training obtains
The feature for trying pedestrian image, by calculate between the feature of the test pedestrian image and the feature of pedestrian image in picture library away from
From unsupervised pedestrian recognition result again can be obtained.
2. the method according to claim 1, wherein the step S1 the following steps are included:
Step S11 determines pedestrian image feature extraction network model；
Step S12 obtains incoherent tape label pedestrian image, and using the incoherent tape label pedestrian image for institute
Pedestrian image feature extraction network model is stated to be initialized；
Step S13 is extracted special using the pedestrian image that the pedestrian image feature extraction network model after initialization is not tape label
Sign, obtains fisrt featureWherein, ziIndicate the feature of the i-th width not pedestrian image of tape label, N indicates not tape label
Pedestrian image quantity.
3. the method according to claim 1, wherein the step S2 the following steps are included:
Step S21 constructs fuzzy depth clustering network, wherein the fuzzy depth clustering network can be configured such that comprising a depth
Spend network and a cluster loss；
Step S22 initializes the fuzzy depth clustering network.
4. according to the method described in claim 3, it is characterized in that, the step S22 the following steps are included:
Step S221 initializes the depth network parameter in the fuzzy depth clustering network；
Step S222 initializes the cluster centre of the fuzzy depth clustering network.
5. the method according to claim 1, wherein the step S3 the following steps are included:
Step S31, using the depth network in the fuzzy depth clustering network after initialization by the fisrt featureIt reflects
It is mapped to a new feature space, study obtains new pedestrian image feature second feature
Step S32 clusters the second feature using Fuzzy c-means Clustering to obtain cluster centre and multiple clusters,
Objective function by minimizing Fuzzy c-means Clustering is updated the cluster centre, and every width pedestrian is calculated
Degree of membership of the image to each cluster；
Cluster damage is calculated using the second feature, the cluster centre of update and the degree of membership that is calculated in step S33
It loses；
Step S34 updates the parameter of depth network in the fuzzy depth clustering network using the cluster loss iteration and gathers
Class center obtains the Optimal cluster centers of depth network in the fuzzy depth clustering network；
Step S35 obtains the fuzzy label of no label pedestrian image using the second feature and Optimal cluster centers.
6. according to the method described in claim 5, it is characterized in that, the objective function of the Fuzzy c-means Clustering indicates are as follows:
Wherein, N indicates the quantity of no label pedestrian image, and K indicates the number of cluster centre, and m > 0 indicates fuzzy coefficient, cjIt indicates
The feature vector of j-th of cluster centre, aijIndicate i-th of pedestrian image to j-th of cluster degree of membership,
For bound term, indicate every width pedestrian image to the sum of the degree of membership of all clusters be 1.
7. according to the method described in claim 5, it is characterized in that, the step S33 the following steps are included:
Step S331 calculates fuzzy allocation based on the second feature, the cluster centre of update and the degree of membership that is calculated
fij；
Step S332, according to the fuzzy allocation fijTarget Assignment is calculated；
Cluster loss is calculated using the KL divergence of the fuzzy allocation and Target Assignment in step S333.
8., will be according to the second feature and institute according to the method described in claim 5, it is characterized in that, in the step S35
State the fuzzy label that the fuzzy allocation that Optimal cluster centers are calculated is defined as the no label pedestrian image.
9. the method according to claim 1, wherein selecting confidence level to be higher than a default threshold in the step S4
The pedestrian image with fuzzy label of value is as the reliability sample.
10. the method according to claim 1, wherein the step S4 the following steps are included:
Step S41 selects confidence level to be higher than the pedestrian image with fuzzy label of a preset threshold as reliability sample；
Step S42 utilizes pedestrian image feature extraction network model described in the reliability sample training with fuzzy label.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811279325.2A CN109299707A (en) | 2018-10-30 | 2018-10-30 | A kind of unsupervised pedestrian recognition methods again based on fuzzy depth cluster |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811279325.2A CN109299707A (en) | 2018-10-30 | 2018-10-30 | A kind of unsupervised pedestrian recognition methods again based on fuzzy depth cluster |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109299707A true CN109299707A (en) | 2019-02-01 |
Family
ID=65145172
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811279325.2A Pending CN109299707A (en) | 2018-10-30 | 2018-10-30 | A kind of unsupervised pedestrian recognition methods again based on fuzzy depth cluster |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109299707A (en) |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109902662A (en) * | 2019-03-20 | 2019-06-18 | 中山大学 | A kind of pedestrian recognition methods, system, device and storage medium again |
| CN110009015A (en) * | 2019-03-25 | 2019-07-12 | 西北工业大学 | EO-1 hyperion small sample classification method based on lightweight network and semi-supervised clustering |
| CN110070538A (en) * | 2019-04-28 | 2019-07-30 | 华北电力大学（保定） | Bolt two-dimensional visual documents structured Cluster method based on form optimization depth characteristic |
| CN110309887A (en) * | 2019-07-09 | 2019-10-08 | 哈尔滨理工大学 | Based on the Fuzzy C-Means Clustering method for detecting abnormality for improving flower pollination |
| CN111401281A (en) * | 2020-03-23 | 2020-07-10 | 山东师范大学 | Unsupervised pedestrian re-identification method and system based on deep clustering and sample learning |
| CN111476168A (en) * | 2020-04-08 | 2020-07-31 | 山东师范大学 | Cross-domain pedestrian re-identification method and system based on three stages |
| CN111507232A (en) * | 2020-04-10 | 2020-08-07 | 三一重工股份有限公司 | Multi-mode multi-strategy fused stranger identification method and system |
| CN111597887A (en) * | 2020-04-08 | 2020-08-28 | 北京大学 | Pedestrian re-identification method and system |
| CN111598004A (en) * | 2020-05-18 | 2020-08-28 | 北京星闪世图科技有限公司 | Progressive-enhancement self-learning unsupervised cross-domain pedestrian re-identification method |
| CN111639561A (en) * | 2020-05-17 | 2020-09-08 | 西北工业大学 | Unsupervised pedestrian re-identification method based on category self-adaptive clustering |
| WO2020186914A1 (en) * | 2019-03-20 | 2020-09-24 | 北京沃东天骏信息技术有限公司 | Person re-identification method and apparatus, and storage medium |
| CN112215275A (en) * | 2020-09-30 | 2021-01-12 | 佛山科学技术学院 | Image processing system and method suitable for K-means algorithm, and recording medium |
| CN113255573A (en) * | 2021-06-17 | 2021-08-13 | 成都东方天呈智能科技有限公司 | Pedestrian re-identification method based on mixed cluster center label learning and storage medium |
| CN118470373A (en) * | 2024-04-02 | 2024-08-09 | 南通大学 | Generic sharing semi-supervised feature selection method for depression brain image |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107506703A (en) * | 2017-08-09 | 2017-12-22 | 中国科学院大学 | A kind of pedestrian's recognition methods again for learning and reordering based on unsupervised Local Metric |
| CN107832672A (en) * | 2017-10-12 | 2018-03-23 | 北京航空航天大学 | A kind of pedestrian's recognition methods again that more loss functions are designed using attitude information |
| CN108288051A (en) * | 2018-02-14 | 2018-07-17 | 北京市商汤科技开发有限公司 | Pedestrian identification model training method and device, electronic equipment and storage medium again |
-
2018
- 2018-10-30 CN CN201811279325.2A patent/CN109299707A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107506703A (en) * | 2017-08-09 | 2017-12-22 | 中国科学院大学 | A kind of pedestrian's recognition methods again for learning and reordering based on unsupervised Local Metric |
| CN107832672A (en) * | 2017-10-12 | 2018-03-23 | 北京航空航天大学 | A kind of pedestrian's recognition methods again that more loss functions are designed using attitude information |
| CN108288051A (en) * | 2018-02-14 | 2018-07-17 | 北京市商汤科技开发有限公司 | Pedestrian identification model training method and device, electronic equipment and storage medium again |
Non-Patent Citations (3)
| Title |
|---|
| HEHE FAN ET AL.: "Unsupervised Person Re-identification: Clustering and Fine-tuning", 《ACM TRANSACTIONS ON MULTIMEDIA COMPUTING, COMMUNICATIONS, AND APPLICATIONS》 * |
| JUNYUAN XIE ET AL.: "Unsupervised Deep Embedding for Clustering Analysis", 《ARXIV》 * |
| VANDANA BHATIA ET AL.: "DFuzzy: a deep learning-based fuzzy clustering model for large graphs", 《KNOWLEDGE AND INFORMATION SYSTEMS》 * |
Cited By (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020186914A1 (en) * | 2019-03-20 | 2020-09-24 | 北京沃东天骏信息技术有限公司 | Person re-identification method and apparatus, and storage medium |
| CN109902662A (en) * | 2019-03-20 | 2019-06-18 | 中山大学 | A kind of pedestrian recognition methods, system, device and storage medium again |
| CN109902662B (en) * | 2019-03-20 | 2023-10-24 | 中山大学 | Pedestrian re-identification method, system, device and storage medium |
| CN110009015A (en) * | 2019-03-25 | 2019-07-12 | 西北工业大学 | EO-1 hyperion small sample classification method based on lightweight network and semi-supervised clustering |
| CN110070538A (en) * | 2019-04-28 | 2019-07-30 | 华北电力大学（保定） | Bolt two-dimensional visual documents structured Cluster method based on form optimization depth characteristic |
| CN110070538B (en) * | 2019-04-28 | 2022-04-15 | 华北电力大学（保定） | Bolt two-dimensional visual structure clustering method based on morphological optimization depth characteristics |
| CN110309887A (en) * | 2019-07-09 | 2019-10-08 | 哈尔滨理工大学 | Based on the Fuzzy C-Means Clustering method for detecting abnormality for improving flower pollination |
| CN111401281A (en) * | 2020-03-23 | 2020-07-10 | 山东师范大学 | Unsupervised pedestrian re-identification method and system based on deep clustering and sample learning |
| CN111401281B (en) * | 2020-03-23 | 2022-06-21 | 山东师范大学 | Unsupervised pedestrian re-identification method and system based on deep clustering and sample learning |
| CN111476168A (en) * | 2020-04-08 | 2020-07-31 | 山东师范大学 | Cross-domain pedestrian re-identification method and system based on three stages |
| CN111476168B (en) * | 2020-04-08 | 2022-06-21 | 山东师范大学 | Cross-domain pedestrian re-identification method and system based on three stages |
| CN111597887A (en) * | 2020-04-08 | 2020-08-28 | 北京大学 | Pedestrian re-identification method and system |
| CN111597887B (en) * | 2020-04-08 | 2023-02-03 | 北京大学 | Pedestrian re-identification method and system |
| CN111507232B (en) * | 2020-04-10 | 2023-07-21 | 盛景智能科技(嘉兴)有限公司 | Stranger identification method and system based on multi-mode multi-strategy fusion |
| CN111507232A (en) * | 2020-04-10 | 2020-08-07 | 三一重工股份有限公司 | Multi-mode multi-strategy fused stranger identification method and system |
| CN111639561A (en) * | 2020-05-17 | 2020-09-08 | 西北工业大学 | Unsupervised pedestrian re-identification method based on category self-adaptive clustering |
| CN111598004A (en) * | 2020-05-18 | 2020-08-28 | 北京星闪世图科技有限公司 | Progressive-enhancement self-learning unsupervised cross-domain pedestrian re-identification method |
| CN111598004B (en) * | 2020-05-18 | 2023-12-08 | 江苏星闪世图科技(集团)有限公司 | Progressive reinforcement self-learning unsupervised cross-domain pedestrian re-identification method |
| CN112215275A (en) * | 2020-09-30 | 2021-01-12 | 佛山科学技术学院 | Image processing system and method suitable for K-means algorithm, and recording medium |
| CN113255573A (en) * | 2021-06-17 | 2021-08-13 | 成都东方天呈智能科技有限公司 | Pedestrian re-identification method based on mixed cluster center label learning and storage medium |
| CN113255573B (en) * | 2021-06-17 | 2021-10-01 | 成都东方天呈智能科技有限公司 | Pedestrian re-identification method based on mixed cluster center label learning and storage medium |
| CN118470373A (en) * | 2024-04-02 | 2024-08-09 | 南通大学 | Generic sharing semi-supervised feature selection method for depression brain image |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109299707A (en) | A kind of unsupervised pedestrian recognition methods again based on fuzzy depth cluster | |
| CN111126360B (en) | Cross-domain pedestrian re-identification method based on unsupervised combined multi-loss model | |
| WO2023087558A1 (en) | Small sample remote sensing image scene classification method based on embedding smoothing graph neural network | |
| CN110414368B (en) | Unsupervised pedestrian re-identification method based on knowledge distillation | |
| CN109034044B (en) | Pedestrian re-identification method based on fusion convolutional neural network | |
| CN111259786B (en) | Pedestrian re-identification method based on synchronous enhancement of appearance and motion information of video | |
| CN110263697A (en) | Pedestrian based on unsupervised learning recognition methods, device and medium again | |
| CN110516095B (en) | Semantic migration-based weak supervision deep hash social image retrieval method and system | |
| WO2022062419A1 (en) | Target re-identification method and system based on non-supervised pyramid similarity learning | |
| CN113705597B (en) | Image processing method, device, computer equipment and readable storage medium | |
| CN112507901B (en) | Unsupervised pedestrian re-identification method based on pseudo tag self-correction | |
| CN110598543B (en) | Model training method based on attribute mining and reasoning and pedestrian re-identification method | |
| CN108764308A (en) | Pedestrian re-identification method based on convolution cycle network | |
| CN108537264B (en) | Heterogeneous image matching method based on deep learning | |
| CN111950372B (en) | Unsupervised pedestrian re-identification method based on graph convolution network | |
| CN104008395B (en) | A kind of bad video intelligent detection method based on face retrieval | |
| CN111027421A (en) | Graph-based direct-push type semi-supervised pedestrian re-identification method | |
| CN107506703A (en) | A kind of pedestrian's recognition methods again for learning and reordering based on unsupervised Local Metric | |
| CN109086672A (en) | A kind of recognition methods again of the pedestrian based on reinforcement learning adaptive piecemeal | |
| CN104298992B (en) | A kind of adaptive scale pedestrian recognition methods again based on data-driven | |
| CN111967325A (en) | Unsupervised cross-domain pedestrian re-identification method based on incremental optimization | |
| CN113569657B (en) | Pedestrian re-identification method, device, equipment and storage medium | |
| CN109214430A (en) | A kind of recognition methods again of the pedestrian based on feature space topology distribution | |
| CN111783526B (en) | Cross-domain pedestrian re-identification method using posture invariance and graph structure alignment | |
| CN109670423A (en) | A kind of image identification system based on deep learning, method and medium |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190201 |
|
| RJ01 | Rejection of invention patent application after publication |