CN104933438A - Image clustering method based on self-coding neural network - Google Patents
Image clustering method based on self-coding neural network Download PDFInfo
- Publication number
- CN104933438A CN104933438A CN201510293670.1A CN201510293670A CN104933438A CN 104933438 A CN104933438 A CN 104933438A CN 201510293670 A CN201510293670 A CN 201510293670A CN 104933438 A CN104933438 A CN 104933438A
- Authority
- CN
- China
- Prior art keywords
- network
- clustering
- class
- cluster centre
- utilizes
- 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 OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses an image clustering method based on a self-coding neural network, and the method is mainly used for the field of unsupervised learning image clustering. The method comprises the steps: initializing a network and updating two parts of the clustering network, and the method mainly relates to a process of image clustering. The method comprises the steps: firstly employing a self-coding network to form an eight-layer neural network, obtaining an initial weight through random initialization, carrying out the random grouping of data, and obtaining an initial image clustering center; secondly adding in-class and inter-class clustering constraint in a self-coding network model, employing a target function with the clustering constraint to achieve the training of the model so as to update a network structure; and finally employing the updated network to obtain the characteristic of the corresponding image, carrying out clustering at a characteristic layer, and updating clustering grouping. The step, proposed by the method, of adding the in-class clustering constraint in the self-coding network model can enable samples of the same type to be distributed more densely in a characteristic space, and the added inter-class clustering constraint can enable samples of different types to be discriminated in the characteristic space through height. With the help of the highly-nonlinear mapping and unsupervised learning capability of a self-coding network, the method is very suitable for image clustering. The method can achieve the performance better than a conventional image clustering algorithm.
Description
[technical field]
The present invention relates to image procossing, machine learning, particularly based on the image clustering method of own coding neural network.
[background technology]
Along with the growth of information handling needs, image clustering needs the algorithm of efficiently and accurately badly.All there is the restriction that accuracy rate is low and computation complexity is high in tradition class algorithm such as K mean algorithm etc.Along with the rise of unsupervised learning and own coding neural network, the image clustering based on own coding neural network obtains the attention of researcher.
Different clustering constrains directly affects Clustering Effect, we propose a kind of newly consider in class and the clustering algorithm retrained between class based on own coding neural network and simultaneously.Image can be mapped to feature space from original data space by autoencoder network by the method, and adds in class and between class at feature space and retrain, and upgrades autoencoder network to obtain best Clustering Effect by iteration.Wherein, constraint can ensure the distribution of compacting of similar sample in class, retrain between class can reach inhomogeneity sample mutually away from.This method had both achieved and image had been mapped to feature space from luv space, also achieved and retrained in the class of feature space and between class, well solved the image clustering problem under large data background.
[summary of the invention]
In order to solve prior art Problems existing, the object of this invention is to provide a kind of image clustering method based on own coding neural network, as shown in Figure 1, comprising the following steps:
Step S1, utilizes eight layers of autoencoder network to set up clustering network structure, and utilizes the weight of this network as initial weight.
Step S2, to add in class and constraint function between class at the coding layer of autoencoder network, makes similar sample near its cluster centre, different cluster centre mutually away from.
Step S3, by all sample random packet, and is mapped to feature space respectively by autoencoder network, then calculates the mean value of the feature representation of all groups, as the cluster centre of this group.
Step S4, utilizes and adds constrained learning autoencoder network between the interior constraint of class and class, upgrade network weight, carry out image clustering.
Step S5, the network after the renewal utilizing S4 to obtain, calculates the feature representation of all samples, compares with cluster centre before, and sample is dispensed to nearest cluster centre.
Step S6, utilizes in S5 and obtains feature representation and calculate the average of the feature representation often organized by new grouping as new cluster centre.
Step S7, utilizes the cluster centre upgraded in S6 to replace the cluster centre of clustering constrain function in S2.
Step S8, forwards S4 to and circulates, until reach frequency of training or clustering network convergence.
According to method of the present invention, sample can be projected to feature space from original image space by own coding neural network, and add in class and between class at feature space and retrain, sample is compacted more at the distribution within class of feature space, different cluster centre mutually away from, achieve good image clustering result.
[accompanying drawing explanation]
Fig. 1 is the process flow diagram of the image clustering method based on own coding neural network.
[embodiment]
Each detailed problem involved in technical solution of the present invention is described in detail below in conjunction with accompanying drawing.It should be pointed out that described embodiment is only intended to be convenient to understand, any restriction effect is not play to the present invention.Fig. 1 is process flow diagram of the present invention, as shown in Figure 1, said method comprising the steps of:
Step S1, sets up the own coding neural network of eight layers, and the transport function of network is s i gmoi d function.Wherein first 4 layers is coding network, and latter 4 layers is decoding network, and initial network weight is given at random.
Step S2, the coding layer being added in autoencoder network to add in class and constraint function between class, and the objective function of whole like this network comprises three parts: retrain between constraint in own coding constraint, class, class.Objective function can specifically be expressed as:
Wherein
corresponding n-th sample x
nat the cluster centre of feature space, c
irepresent i-th cluster centre, f (x
n) represent sample x
nthrough the expression of coding network at feature space,
represent the feature representation f (x of sample
n) through the output of decoding network.
Step S3, is divided into k group at random by all samples, obtains the initial clustering label L=[l of each sample
1, l
2..., l
n], wherein N represents N number of sample, and l represents the label of sample, the classification namely.Then, all samples are obtained characteristic of correspondence layer by initial network and expresses, calculating often organizes the average of expression as initial cluster centre C.
Step S4, utilizes all sample repetitive exercise own coding neural networks, the objective function in Optimization Steps S2.
Step S5, utilizes the network weight after upgrading, recalculates the feature representation of all samples, and compare with cluster centre before, again divide into groups according to distance.
Step S6, utilizes the sample after upgrading in grouping to calculate every class mean, as the cluster centre after renewal.
Step S7, utilizes the cluster centre after upgrading to upgrade the constraint of objective function.
Step S8, forwards step S4 to and loop iteration, until autoencoder network convergence, or frequency of training reaches the upper limit.
[embodiment]
In order to describe the specific embodiment of the present invention in detail, illustrate for certain large-scale handwritten numeral image data set.This data set comprises 5000 images, comprises the 0-9 digital picture that totally 10 classes are different respectively, and often opening image size is 30*30.Utilize the clustering method proposed to carry out image clustering to this database, sample is divided into 10 classes when non-supervisory.Concrete steps are as follows:
Step S1, utilize eight layers of autoencoder network to set up clustering network structure, network structure is (900-500-200-100-30-100-200-500-900), and the transport function of network is sigmoid function.Wherein first 4 layers is coding network, and latter 4 layers is decoding network, and initial network weight is given at random.
Step S2, to add in class and constraint function between class at the coding layer of autoencoder network, makes similar sample near its cluster centre, different cluster centre mutually away from.The objective function of whole like this network comprises three parts: retrain between constraint in own coding constraint, class, class.Objective function is specifically expressed as:
Wherein
corresponding n-th sample x
nat the cluster centre of feature space, c
irepresent i-th cluster centre, f (x
n) represent sample sample x
nthrough the expression of coding network at feature space,
represent the feature representation f (x of sample
n) through the output of decoding network.
All 5000 samples are divided into 10 groups by step S3 at random, obtain the initial clustering label L=[l of each sample
1, l
2..., l
n], wherein N represents N number of sample, and l represents the label of sample, the classification namely.Then, all samples are obtained characteristic of correspondence layer by initial network and expresses, calculating often organizes the average of expression as initial 10 cluster centre C.
Step S4, utilizes all sample repetitive exercise own coding neural networks, the objective function in Optimization Steps S2.
Step S5, utilizes the network weight after upgrading, recalculates the feature representation of all samples, and compare with the cluster centre of 10 before, be again divided into 10 groups according to distance.
Step S6, the sample after utilization upgrades in grouping calculates the averages of 10 groupings, as 10 cluster centres after renewal.
Step S7, utilizes the constraint of objective function in the cluster centre step of updating S2 after upgrading.
Step S8, forwards step S4 to and loop iteration, until autoencoder network convergence, or frequency of training reaches 50 times.
The above, be only one of the specific embodiment of the present invention, and protection scope of the present invention is not limited thereto.Any people being familiar with this technology, in the technical scope disclosed by the present invention, can understand the conversion or replacement expected, should contain within the scope of the present invention.Therefore, protection scope of the present invention should be as the criterion with the protection domain of claims.
Claims (3)
1., based on an image clustering method for own coding neural network, mainly comprise step:
Step S1, utilizes eight layers of autoencoder network to set up clustering network structure, and utilizes the weight of this network as initial weight.
Step S2, to add in class and constraint function between class at the coding layer of autoencoder network, makes similar sample near its cluster centre, different cluster centre mutually away from.
Step S3, by all sample random packet, and is mapped to feature space respectively by autoencoder network, then calculates the mean value of the feature representation of all groups, as the cluster centre of this group.
Step S4, utilizes and adds constrained learning autoencoder network between the interior constraint of class and class, upgrade network weight, carry out image clustering.
Step S5, the network after the renewal utilizing S4 to obtain, calculates the feature representation of all samples, compares with cluster centre before, and sample is dispensed to nearest cluster centre.
Step S6, utilizes in S5 and obtains feature representation and calculate the average of the feature representation often organized by new grouping as new cluster centre.
Step S7, utilizes the cluster centre upgraded in S6 to replace the cluster centre of clustering constrain function in S2.
Step S8, forwards S4 to and circulates, until reach frequency of training or clustering network convergence.
2. method according to claim 1, utilizes own coding neural network to carry out image clustering.
3. method according to claim 1, utilizes in class simultaneously and between class, constraint adds the objective function of autoencoder network as clustering constrain.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510293670.1A CN104933438A (en) | 2015-06-01 | 2015-06-01 | Image clustering method based on self-coding neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510293670.1A CN104933438A (en) | 2015-06-01 | 2015-06-01 | Image clustering method based on self-coding neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN104933438A true CN104933438A (en) | 2015-09-23 |
Family
ID=54120598
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510293670.1A Pending CN104933438A (en) | 2015-06-01 | 2015-06-01 | Image clustering method based on self-coding neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104933438A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109086805A (en) * | 2018-07-12 | 2018-12-25 | 华南理工大学 | A kind of clustering method constrained based on deep neural network and in pairs |
| CN109983480A (en) * | 2016-11-15 | 2019-07-05 | 谷歌有限责任公司 | Train Neural Networks with Clustering Loss |
| CN110309853A (en) * | 2019-05-20 | 2019-10-08 | 湖南大学 | Medical Image Clustering Method Based on Variational Autoencoder |
| CN110858812A (en) * | 2018-08-24 | 2020-03-03 | 中国移动通信集团浙江有限公司 | Network element cutover and watching method and device |
| WO2021197032A1 (en) * | 2020-04-01 | 2021-10-07 | 支付宝(杭州)信息技术有限公司 | Clustering system and method |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20050216207A1 (en) * | 2004-03-24 | 2005-09-29 | Illumina, Inc. | Artificial intelligence and global normalization methods for genotyping |
| CN103530689A (en) * | 2013-10-31 | 2014-01-22 | 中国科学院自动化研究所 | Deep learning-based clustering method |
| CN104618852A (en) * | 2015-01-28 | 2015-05-13 | 广东电网有限责任公司电力调度控制中心 | Hierarchical clustering-based data aggregating method and system |
-
2015
- 2015-06-01 CN CN201510293670.1A patent/CN104933438A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20050216207A1 (en) * | 2004-03-24 | 2005-09-29 | Illumina, Inc. | Artificial intelligence and global normalization methods for genotyping |
| CN103530689A (en) * | 2013-10-31 | 2014-01-22 | 中国科学院自动化研究所 | Deep learning-based clustering method |
| CN104618852A (en) * | 2015-01-28 | 2015-05-13 | 广东电网有限责任公司电力调度控制中心 | Hierarchical clustering-based data aggregating method and system |
Non-Patent Citations (2)
| Title |
|---|
| CHUNFENG SONG 等: "Auto-encoder Based Data Clustering", 《LECTURE NOTES IN COMPUTER SCIENCE（LNCS）》 * |
| 孙艳丰 等: "基于改进Fisher准则的深度卷积神经网络识别算法", 《北京工业大学学报》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109983480A (en) * | 2016-11-15 | 2019-07-05 | 谷歌有限责任公司 | Train Neural Networks with Clustering Loss |
| CN109983480B (en) * | 2016-11-15 | 2023-05-26 | 谷歌有限责任公司 | Train Neural Networks Using Clustering Loss |
| CN109086805A (en) * | 2018-07-12 | 2018-12-25 | 华南理工大学 | A kind of clustering method constrained based on deep neural network and in pairs |
| CN109086805B (en) * | 2018-07-12 | 2020-07-28 | 华南理工大学 | Clustering method based on deep neural network and pairwise constraints |
| CN110858812A (en) * | 2018-08-24 | 2020-03-03 | 中国移动通信集团浙江有限公司 | Network element cutover and watching method and device |
| CN110309853A (en) * | 2019-05-20 | 2019-10-08 | 湖南大学 | Medical Image Clustering Method Based on Variational Autoencoder |
| WO2021197032A1 (en) * | 2020-04-01 | 2021-10-07 | 支付宝(杭州)信息技术有限公司 | Clustering system and method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110059878B (en) | Photovoltaic power generation power prediction model based on CNN LSTM and construction method thereof | |
| CN103530689B (en) | A kind of clustering method based on degree of depth study | |
| CN113313164B (en) | Digital pathological image classification method and system based on super-pixel segmentation and graph convolution | |
| CN104933438A (en) | Image clustering method based on self-coding neural network | |
| CN110543563B (en) | Hierarchical text classification method and system | |
| CN110728295B (en) | Semi-supervised landform classification model training and landform graph construction method | |
| CN112101430A (en) | Anchor frame generation method for image target detection processing and lightweight target detection method | |
| Yao et al. | Clustering of typical wind power scenarios based on K-means clustering algorithm and improved artificial bee colony algorithm | |
| CN113128671B (en) | A method and system for dynamic prediction of service demand based on multimodal machine learning | |
| CN104850864A (en) | Unsupervised image recognition method based on convolutional neural network | |
| CN116226467B (en) | Community discovery method of graph convolution neural network based on node structural features | |
| CN110084136A (en) | Context based on super-pixel CRF model optimizes indoor scene semanteme marking method | |
| CN115131557A (en) | Lightweight segmentation model construction method and system based on activated sludge image | |
| CN114912489A (en) | Signal modulation identification method | |
| Yang et al. | A modification of adaptive moment estimation (adam) for machine learning | |
| CN114707655A (en) | Quantum line conversion method, quantum line conversion system, storage medium and electronic equipment | |
| CN119166861A (en) | Network prediction method and device based on graph neural network, electronic device and medium | |
| CN108985378A (en) | A kind of domain adaptive approach based on mixing interleaving depth network | |
| CN116758349A (en) | Hyperspectral image classification method based on multi-scale superpixel node aggregation graph convolutional residual network | |
| CN105976018A (en) | Discrete Pigeon Swarm Algorithm for Optimal Layout of Sensors for Structural Health Monitoring | |
| Zhang et al. | Improvement of K-means algorithm based on density | |
| CN113378731B (en) | Vector extraction method of green space water system based on convolutional neural network and energy constraint | |
| Zhang et al. | Center-guided Classifier for Semantic Segmentation of Remote Sensing Images | |
| CN112949590B (en) | Cross-domain pedestrian re-identification model construction method and system | |
| CN115273645B (en) | Map making method for automatically clustering indoor surface elements |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20150923 |